R420 might only support pixel shader 2.0

[rwolf]

Here is the thread where I found it....

http://www.rage3d.com/board/showthread.php?s=&threadid=33743797

Here is the original source...

Google translation of www.3dcenter.org article...

http://translate.google.com/translate?u=http%3A%2F%2Fwww.3dcenter.org%2Fartike l%2F2004%2F02-22_a.php&langpair=de%7Cen&hl=de&ie=UTF-8&oe=UTF-8&prev=%2Flanguage_tools

[Demirug]

R420 supports a little bit more. It will be able to run long 2.0 shader. This will allow to say it is a 2.X chip.

[Hyp-X]

The lack of PS3.0 support is one thing, but the article seem to say there won't be any 3.0 shader support.

That means no VS3.0 which I find much more disappointing. :(

[DemoCoder]

R420 supports a little bit more. It will be able to run long 2.0 shader. This will allow to say it is a 2.X chip.

Makes sense if they use F-Buffer to extend instruction limit, and perhaps add arbitrary swizzle and predictation.

[bloodbob]

R420 supports a little bit more. It will be able to run long 2.0 shader. This will allow to say it is a 2.X chip.

Makes sense if they use F-Buffer to extend instruction limit, and perhaps add arbitrary swizzle and predictation.

Well if they used F-Buffer it would be a very nice thing for the R350 owners since they are still waiting for it (Just like the R300 are waiting for SSAA). One assumes if the instruction length is no longer a problem if you use F-buffer therefore you could emulate arbitrary swizzles on R350 (assuming enough temps are aviable ) without to much of a problem.

If the R420 don't support PS3.0 its going to be a long wait before ATI will support PS3.0 support.

[DemoCoder]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

Perhaps this is the reason behind Dave's "Shader 3.0 inflection or deflection" post. If Dave knows the R420 doesn't support 3.0, but that the NV40 does (as some of Nvidia's TWIMTBP devs seem to imply), then of course, it's the first salvo in an argument to assert that 3.0 won't drive games, won' t be that relevent, and doesn't matter.

Of course, I don't neccessarily disagree. Where the hell is Half-Life2? Where's my killer PS2.0 titles? Where's my games with HDR rendering everywhere? I'm not even sure PS2.0 is an "inflection point" given the paucity of titles at the moment. Vast majority of games I play on my R300 are < DX9 titles. Primarily benefit I have derived from my R300 is IQ.


Should be an interesting period ahead. I would say that a card capable of PS3.0/VS3.0 is interesting from the standpoint of being able to fully handle GLSlang, which is cool.

[jvd]

Well this is nothing like what i've been hearing . But oh well any day now and we will find out .

[Hanners]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

That seems like a pretty weird assumption - I would say the tide here will turn against ATi for not driving their feature set forward (Assuming nVidia's PS 2.0 performance is on or close to a par with ATi's this time around that is).

Anyway, if this rumour is true, it's pretty much suicide on ATi's part, from a marketing position at least.

[kyleb]

ya if it goes down like that i will be happy with nvidia and disspointed with ati, but i think DemoCoder is just projecting how he would react if nvidia was sticking to 2.0 and ati were the ones delivering 3.0. :P

regardless, i don't see any reason at this point to conclude one way or another as to the validity of the claim.

[martrox]

Well, if that's true, - R420 not having 3.0 - and if nVidia has redone their FSAA and corrected the problems with their filtering - AND done away with the driver cheats - I guess I'll be forced to take a walk on the dark side........ :wink:

[DemoCoder]

Well let's wait and see. In the past, it hasn't been true. No one cared when Nvidia introduced new features like 32-bit color, or T&L. The argument then was, even if other cards lacked new features, it wasn't important, only performance on old games, or on the previous year's API spec mattered. I think the only way an IHV could be unassailable is if they delivered 3.0 with hardly any performance problems. Otherwise, people will just argue that dynamic branches, or what have you, are worthless, if they have a serious speed penalty.

Pretty much, framerate and IQ (aliasing) has been the only reliable indicators on this board that most people can agree on. Everyone else seems split on features, depending on which IHVs have it, and which "camp" they are in.

[Guden Oden]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

I believe that post of yours qualifies as a rant, and should be clearly marked as such to properly inform the public! :lol:

Anyway, I disagree with your assertions. Assume R420 does not support PS3.0, but instead has awesomely sped up PS2.0 instead (pretty big assumption I agree), why would anyone be dissatisfied?

Is it BAD that it won't run non-existant software, but instead will be running upcoming software much faster? I can't see ANYTHING wrong with that, apart from they'll lose a few points on the geekoid hardware websites for not providing a checkbox feature.

Does the market really need further fragmentation when even the current shader model barely has any support at all? Seems pretty pointless to criticize ATi for not driving tech forward when there's no software-based reason to invest in current tech.

Well, that's my take on it anyway... Fanbois may choose to disagree, but I care not. :lol:

[Demirug]

I know this is not a proof but IMHO a indication.

The new version of ASHLI (http://www.ati.com/developer/ashli.html) allready supports R420. I am played a little bit around with it. If i use the PS 2.X and R420 Setting I was able to generate long shaders (> 450 instruction) but i was not able to do more than 4 dependent read levels in one pass. If I try this dependet read test with the NV30 profil ASHLI stop working with an internal error. No unlimited dependent read = no PS 3.0 support. Maybe this is only a bug in ASHLI.

[Dave Baumann]

Perhaps this is the reason behind Dave's "Shader 3.0 inflection or deflection" post. If Dave knows the R420 doesn't support 3.0, but that the NV40 does (as some of Nvidia's TWIMTBP devs seem to imply), then of course, it's the first salvo in an argument to assert that 3.0 won't drive games, won' t be that relevent, and doesn't matter.

Salvo? That just stems from some conversations I had with some people before Christmas.

[Edit] : Of course, we'll gloss over the fact the very next news post (http://www.beyond3d.com/#news10000) talks up potential implementations of PS3.0 :roll:

I'm not even sure PS2.0 is an "inflection point" given the paucity of titles at the moment.

With Intel supporting it then it'll be more than that - it will become the baseline target eventually.

[DemoCoder]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

I believe that post of yours qualifies as a rant, and should be clearly marked as such to properly inform the public! :lol:

Anyway, I disagree with your assertions. Assume R420 does not support PS3.0, but instead has awesomely sped up PS2.0 instead (pretty big assumption I agree), why would anyone be dissatisfied?



Hahah, brilliant. By misinterpreting my posts, you proved the exact point I was making. I never claimed anyone would be disappointed by the R420 if it didn't have 3.0 shaders. I asserted the reverse. Try rereading my post. Thanks for quickly providing an example to prove my point.


Does the market really need further fragmentation when even the current shader model barely has any support at all? Seems pretty pointless to criticize ATi for not driving tech forward when there's no software-based reason to invest in current tech.

Is it BAD that it won't run non-existant software, but instead will be running upcoming software much faster? I can't see ANYTHING wrong with that, apart from they'll lose a few points on the geekoid hardware websites for not providing a checkbox feature.


I never said any such thing. I merely predicted what your argument would be. Your post qualifies as a rant, because you didn't even read what I wrote.

[DemoCoder]

With Intel supporting it then it'll be more than that - it will become the baseline target eventually.

I agree. PS2.0 has to filter down still further for the benefits to take hold. It's getting close with cheap DX9 cards from ATI and nVidia, but probably will finally reach that state after the R420/NV40 are released, and R300-class cards become the "low end"

[Arun]

NV4x: The only product family to sport true DX9.1 compliancy! :twisted:


Uttar

[DemoCoder]

Well, aside from there being no DX9.1 yet, isn't IMG working on a 3.0 chip? So atleast it will be IMG and Nvidia, and therefore they would not be "the only ones with 3.0"

However, I can see the marketing on the wall already. NVidia will milk 3.0 shaders for all their worth, and the other side will argue non-stop that 3.0 is a marketing PR feature only.

I just hope Nvidia fixed their AA this time. If they deliver performance and 3.0, but same old OGMS FSAA, it won't matter.

[jvd]

well if ati doesn't support p.s 3.0 and nvidia does I will buy nvidia as long as the image quality is on par with ati . p.s 3.0 is actually usable . p.s 2 is as fast or faster than ati .

[DemoCoder]

I don't think ASHLI can be used to determine R420's support level. Is this the only source of the no 3.0 rumors?

Even Cg didn't fully support the NV30 when it was first released. I wouldn't expect ASHLI to support 100% of the R420's capabilities at this moment.

[Ailuros]

Well, aside from there being no DX9.1 yet, isn't IMG working on a 3.0 chip? So atleast it will be IMG and Nvidia, and therefore they would not be "the only ones with 3.0"

However, I can see the marketing on the wall already. NVidia will milk 3.0 shaders for all their worth, and the other side will argue non-stop that 3.0 is a marketing PR feature only.

I just hope Nvidia fixed their AA this time. If they deliver performance and 3.0, but same old OGMS FSAA, it won't matter.

I doubt they'll ever be a any "dx9.1".

I also doubt that NV has been as foolish as to not consider a sparse sampled algorithm and to that up to 8x sample MSAA; if not then I agree. Other elements of personal interest are the possible advancements in the Anisotropy department.

[incurable]

However, I can see the marketing on the wall already. NVidia will milk 3.0 shaders for all their worth, and the other side will argue non-stop that 3.0 is a marketing PR feature only.
And knowing how similar struggles turned out in the past, we might just guess who'll win this one.

I hope NV4x fixes their AA and performance problems.

cu

incurable

[Joe DeFuria]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

That's odd...why would anyone argue that?

(If we continue with an assumption that ATI doesn't support 3.0...)

The key there, is you said if nVidia fixes 2.0 performance. Hey, if nVidia FIXES 2.0 (as good or better 2.0 perperformance than ATI) while delivering 3.0, that's great.

You might see a "holy war" on this board, if nVidia delivers the same crappy 2.0 performance, and bolts 3.0 on top of it. Then you might see people coming down "hard" on nVidia.

[anaqer]

That's odd...why would anyone argue that?
Well, if you make the silent assumption of people here being fanatic nV-haters, like his prophecy implies... :roll:

[christoph]

i asked over at 3dc about the proof or source of the story...

quote from leonidas from 3dc
http://www.forum-3dcenter.org/vbulletin/showthread.php?s=e6ca0550f3b755c01126e0e262dc4488& threadid=126074&perpage=20&pagenumber=7
Die Quelle ist mehr als direkt, direkter geht es gar nicht mehr.
Nicht nur die Position der Quelle ist direkt, sondern auch die Art,
wie sie selber zu den Informationen gekommen, ist ultimativ direkt.
Eingeweihte ahnen womöglich, was ich meine :-).

[jvd]

i asked over at 3dc about the proof or source of the story...

quote from leonidas from 3dc
Die Quelle ist mehr als direkt, direkter geht es gar nicht mehr.
Nicht nur die Position der Quelle ist direkt, sondern auch die Art,
wie sie selber zu den Informationen gekommen, ist ultimativ direkt.
Eingeweihte ahnen womöglich, was ich meine :-).

um english please haha

[elroy]

Babelfish translation:

The source is more than directly, more directly goes it no longer. Not only the position of the source is direct, but also the kind, like it come to the information, is ultimatively direct. Inaugurated ones possibly suspect, what I mean

[Hanners]

Um, English please... ;)

[Sxotty]

Well, if that's true, - R420 not having 3.0 - and if nVidia has redone their FSAA and corrected the problems with their filtering - AND done away with the driver cheats - I guess I'll be forced to take a walk on the dark side........ :wink:

And this post proves democoder is right :)

[Geo]

Babelfish translation:

The source is more than directly, more directly goes it no longer. Not only the position of the source is direct, but also the kind, like it come to the information, is ultimatively direct. Inaugurated ones possibly suspect, what I mean

Hmm. Maybe it just isn't in the first release drivers yet, but the hardware can support it?

[Geo]

Hmm. Maybe it just isn't in the first release drivers yet, but the hardware can support it?

Altho, to argue against my own theory, Dave's comment awhile back about the "tug of war" over the 9.1 moniker would suddenly make sense if R420 would not qualify for that label.

[Evildeus]

Well, it depends on where the information comes from, drivers or something else.

[digitalwanderer]

Hmm. Maybe it just isn't in the first release drivers yet, but the hardware can support it?

Altho, to argue against my own theory, Dave's comment awhile back about the "tug of war" over the 9.1 moniker would suddenly make sense if R420 would not qualify for that label.
Could you expand on that thought a bit please? It's tickling something in my brain and I wanna see if I can figure out what the tickle is.

[Evildeus]

Perhaps 3DCenter and the Inq 8) have the same source :lol: :lol:
Lastly there is word on the NV40 R420 front. Neither card was shown, but several people who would be familiar with both have said Nvidia is going to come back very strong soon. The word 'own' was bandied about, as in Nvidia will own ATI. Not sure what it means, but CeBit should be a very interesting show.
http://www.theinquirer.net/?article=14290 :roll:

[Dave Baumann]

I suspect I know who that "own" comment has come from and I also suspect that he's only dealing with half the deck.

[Evildeus]

Who? :evil: And he doesn't have Ati specs? :?:

[Sabastian]

That would be similar to the way that nvidia didn't directly support PS1.4 in DX8 would it not? I don't recall everyone coming down hard on nvidia because they didn't support PS1.4... well not everyone there were some though. Personally I hope that all IHV support DX next methodically . Surely it would be a negative against ATi if they choose not to support it. But if sales of the R4xx are anything similar to the nv2x series derivatives then surely it will only be a small but influential group of enthusiasts looking for cutting edge technology that will be upset. Most however will be happy to see a faster DX9.0b. I would expect Hardware sites like AnandTech and THG to rail against ATi where they would not against nvidia. Personally if the technology of dx next is exciting I would be disappointed if ATi did not implement hardware support. It would however indicate a role reversal between ATi and nvidia to a degree. ATi has a better record for supporting API features that nvidia did not support.. I suppose if nvidia supports the next generation of dx9 more thoroughly then it will be a sign in some way that ATi is falling behind then? Heh, it is a never ending cycle. I would be surprised though if ATi did not support dx next comprehensively though particularly in light of the development work they are doing WRT the xbox 2.

[John Reynolds]

I'll say this: if NV40 has comparable AA to R420, similar (say within 5-10% across the board) PS 2.0 performance, and 3.0 support whereas R420 doesn't, ATI has a long, tough road ahead of them this year.

[Quitch]

Well, if that's true, - R420 not having 3.0 - and if nVidia has redone their FSAA and corrected the problems with their filtering - AND done away with the driver cheats - I guess I'll be forced to take a walk on the dark side........ :wink:

And this post proves democoder is right :)

All that proves is that PS 3.0 support is not top of martox's list.

I found DC's post to make some huge assumptions that my year of this board would suggest are baseless. I'd have thought DC would know better.

I would simply say that PS 3.0 support is NOT the most important thing in peoples eyes, and if nVidia work on it to the detriment of other more important things, then people will come down hard, and rightly so.

[Evildeus]

Sabastian,
Dany Lepage thinks that PS 3.0 is a huge step and the baseline for the next games.

IMHO, no PS 3.0 support by R420 is pure speculation till we get the final specs...

[PaulS]

The lack of full PS3.0 support in the R420 gives us this as a likely outcome:

- nVidia "winning" the bullet point feature set wars (the DX9.1 moniker mentioned earlier would be far too tempting, I'd say)
- nVidia being the sole PS3.0 supporting IHV
- ATi continuing to beat nVidia in DX9 benchmarks
- ATi continuing to hold the moral high ground in terms of drivers (i.e. the cheats won't disappear)
- nVidia winning the majority of OGL benchmarks

Fairly obvious stuff. Based on a couple of conversations I've had recently, it seems almost certain that nVidia won't beat ATi in DX9, despite improved performance, and that nVidia is now aware of that - even if the likes of Anand/Inq are playing with the wrong information at the moment.

And the comment of "only dealing with half the deck" reinforces that, in my mind. I doubt the "own" comment came from nVidia, given that they apparently now know what they're talking about re: R420.

EDIT: Changed the "maybe" nature of the introduction ;)

[Sabastian]

Sabastian,
Dany Lepage thinks that PS 3.0 is a huge step and the baseline for the next games.

IMHO, no PS 3.0 support by R420 is pure speculation till we get the final specs...

Like I said I would be surprised if ATi were not to support dx next broadly, particularly if you keep in mind the work they are doing with MS with regards to the xbox2.

[whql]

Dany Lepage thinks that PS 3.0 is a huge step and the baseline for the next games.

Well, if NV are touting it to developers then I wouldn't expect Dany to say much else seeing as he writes titles on the TWIMTBP program and is ex-NV.

[Miksu]

Didn't David Nalasco say in a recent interview that R420's focus is performance, not the features? So this would go hand on hand with the possible lack of PS3.0 support.

[Evildeus]

Dany Lepage thinks that PS 3.0 is a huge step and the baseline for the next games.

Well, if NV are touting it to developers then I wouldn't expect Dany to say much else seeing as he writes titles on the TWIMTBP program and is ex-NV.Well, TWIMTBP is just marketing and has never mean more performance contrary to what Dany is saying about PS 3.0 ;). But i'm not a programmer, so who knows :?:

[THe_KELRaTH]

..........And the latest News from the PR Nvidia camp is that the R420 won't even support DX9 at all but will concentrate on offering a whole 10% speed increase for DX8.1...... :lol:

[digitalwanderer]

Assuming the lack of PS3.0 in R420 turns out to be correct (and it's not a given at this point, i guess), I imagine it'll end up being:

- nVidia "winning" the bullet point feature set wars (the DX9.1 moniker mentioned earlier would be far too tempting, I'd say)
- nVidia being the sole PS3.0 supporting IHV
- ATi continuing to beat nVidia in DX9 benchmarks
- ATi continuing to hold the moral high ground in terms of drivers (i.e. the cheats won't disappear)
- nVidia winning the majority of OGL benchmarks

Fairly obvious stuff. Based on a couple of conversations I've had recently, it seems almost certain that nVidia won't beat ATi in DX9, despite improved performance, and that nVidia is now aware of that - even if the likes of Anand/Inq are playing with the wrong information at the moment.

And the comment of "only dealing with half the deck" reinforces that, in my mind. I doubt the "own" comment came from nVidia, given that they apparently now know what they're talking about re: R420.
Thanks Paul, this is a great little summary post! :)

[Bouncing Zabaglione Bros.]

Didn't David Nalasco say in a recent interview that R420's focus is performance, not the features? So this would go hand on hand with the possible lack of PS3.0 support.

I thought PS3 was currently part of DX9? All the recent discussions on whether there would be a DX9.1 pretty much were finished because MS said there was no need for DX9.1, as PS 3.0 is covered in the current DX9 spec. This means IMO that PS 3.0 could easily be considerd to be within the current DX9 spec.

I wouldn't be surprised if this all this is simply the start of Nvidia's ususal campaign of FUD, disinformation, and spoliers against their competitors that they always indulge in during the runup to any major product launches. We saw the same kind of unattributed rumours spring up before R300, NV30 and NV35, and the fan boys carried that onwards in a cry of how Nvidia would "own" ATI - and we saw how that all turned out.

Personally, I think Nvidia will have to do more than just support PS 3.0. Look at how they "supported" PS 2.0. They plugged it as "extended" and able to to extra long shaders, but not at any useful speed. What if Nvidia crow about their PS 3.0 support, only for us to find it is just another marketing tickbox for an unusable feature?

I think we have to see the hardware before we can start listening to unattributed rumours about ATI products that have the style of coming from pro-Nvidia sources via the usual backdoors.

[Dave Baumann]

I think we have to see the hardware before we can start listening to unattributed rumours about ATI products that have the style of coming from pro-Nvidia sources via the usual backdoors.

Errr, the discussion in this case stems from the support withing ATI's own developer tools.

[Ante P]

Apart from the obvious battle of the power point slides with specifications what real matter will it make?
Ie. will many (10+ during 2004) game titles use PS/VS 3.0 (to a larger extent than just novelty support as we saw in GunMetal)? If so will it make a difference in Performance and/or Quality?

From my quite limited understanding PS/VS 3.0 won't really bring much to the table in terms of improved image quality. (ie no higher precision for example) ATi are going to be damned fast at PS2.0 but won't support 3.0, nVidia onm the other hand isa little bitslower at PS.20 but support 3.0 which makes their boards a bit faster than in 2.0 mode.. thus it might all amount to a big nothing.

But then again, personally I think there's much more interesting things. Things that directly impact all games. FSAA and Aniso algorithms for an example. :)
I'd take 8x "ATi-like" FSAA on NV40 over PS3.0 any day o' the weak. ;)

[Evildeus]

According to Mufu:
The pixel pipelines are based on R300. There are functional additions, but they are insufficient to fully support shader model 3.0.
http://www.rage3d.com/board/showthread.php?s=072921629d274ddc69985aaeadd39020&threadid=33743797

[Tim Murray]

I'd take 8x "ATi-like" FSAA on NV40 over PS3.0 any day o' the weak.
Well then, March 24 is your lucky day, dearie. ;)

[Bjorn]

According to Mufu:
The pixel pipelines are based on R300. There are functional additions, but they are insufficient to fully support shader model 3.0.
http://www.rage3d.com/board/showthread.php?s=072921629d274ddc69985aaeadd39020&threadid=33743797

Assuming that the R420 lacks PS 3.0 support, what'll happen with support for centroid MSAA ?

Cause i seem to remember that ATi couldn't support it although the hardware was there, all because it wasn't added in the PS2.0 spec, only PS3.0. Or has that problem been resolved ?

[PaulS]

According to Mufu:
The pixel pipelines are based on R300. There are functional additions, but they are insufficient to fully support shader model 3.0.
http://www.rage3d.com/board/showthread.php?s=072921629d274ddc69985aaeadd39020&threadid=33743797

Assuming that the R420 lacks PS 3.0 support, what'll happen with support for centroid MSAA ?

Cause i seem to remember that ATi couldn't support it although the hardware was there, all because it wasn't added in the PS2.0 spec, only PS3.0. Or has that problem been resolved ?

Well centroid works right now, so it's not much of an issue. Also, the key word in the quote is fully.

[digitalwanderer]

According to Mufu:
The pixel pipelines are based on R300. There are functional additions, but they are insufficient to fully support shader model 3.0.
http://www.rage3d.com/board/showthread.php?s=072921629d274ddc69985aaeadd39020& threadid=33743797
Hmmmm....you don't suppose ATi might be planning on implementing a software solution somehow to run the PS 3.0 shaders aka the way the 5200 is directX 9 compliant the R420 would be too?

[Dave Baumann]

From my quite limited understanding PS/VS 3.0 won't really bring much to the table in terms of improved image quality. (ie no higher precision for example) ATi are going to be damned fast at PS2.0 but won't support 3.0, nVidia onm the other hand isa little bitslower at PS.20 but support 3.0 which makes their boards a bit faster than in 2.0 mode.. thus it might all amount to a big nothing.

PS3.0 isn't that big a leap from NVIDIA's PS2.x support - one of the main differences being arbitary branching support. As has been mentioned in this thread, VS3.0 is probably the more interesting change for the sahder 3.0 generation due to the inclusion of the texture lookups in the VS. VS3.0 brings potential performance increases for handling multiple characters and easier support for displacement mapping, which can ultimately increase IQ. Curiously, though, one of the IHV's not supporting it would not be its biggest block IMO - that will be Intel.

[Evildeus]

Hmmmm....you don't suppose ATi might be planning on implementing a software solution somehow to run the PS 3.0 shaders aka the way the 5200 is directX 9 compliant the R420 would be too?
Could be, but don't you think that the performance would be awfull? :?

[Bouncing Zabaglione Bros.]

I think we have to see the hardware before we can start listening to unattributed rumours about ATI products that have the style of coming from pro-Nvidia sources via the usual backdoors.

Errr, the discussion in this case stems from the support withing ATI's own developer tools.

Pre-release tools. In the past, have ATI's tools been completely up to date well in advance of shipping hardware?

Or is this a usual DB tease and you know for a fact that R420 won't have PS 3.0? Okay, I know you won't answer that question, but can you tell us if you actually know for sure whether R420 will support PS 3.0, or if you are just inferring like the rest of us?

[demalion]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

I don't see it. The benefits of PS 3.0 and VS 3.0 seem pretty evident, and with certain missing functionality there is going to be trouble with API support under DirectX for ATI. There are some things that they might get away with not offering fully in hardware...what with PCI Express and depending on what key features they do offer, but that will only matter when they've managed to deliver them and accordingly at what performance.

In no way does this make 3.0 wasted if NV40 delivers it successfully. At worst, it might make it under-utilized for a time. I don't see an unholy war sprouting up here asserting that...too many would argue against it.

The only dispute against the NV40's PS 3.0 and VS 3.0 support that would seem to have legs is if NV40 added a whole new set of significant performance problems relating to implementing that featureset. Something like this would be required to actually make a case against the NV40. I don't see this either, since it would probably have to decrease performance significantly below NV35 for it to be truly be an argument against it.

Even if the R420 was tremendously faster, that would be an offset for this presumed functionality advantage, not something that makes it useless.

[Tim Murray]

I doubt that PS3.0 will have a huge impact on games.

Now, a huge impact on workstation apps? There's a very good chance (especially since Chris Donahue hinted (http://www.beyond3d.com/forum/viewtopic.php?p=209857#209857) at some workstation apps using D3D in the not-too-distant future).

Also, is GLSLang closer to ps_2_0 or ps_3_0 (I guess the question is more "does GLSLang support branching")?

[Ante P]

I'd take 8x "ATi-like" FSAA on NV40 over PS3.0 any day o' the weak.
Well then, March 24 is your lucky day, dearie. ;)

What happens on March 24th? :?
NV40 is due March 18th.

[Tim Murray]

I'd take 8x "ATi-like" FSAA on NV40 over PS3.0 any day o' the weak.
Well then, March 24 is your lucky day, dearie. ;)

What happens on March 24th. :?
NV40 is due March 18th.
hm, I know CeBit and GDC both end on March 24, so I thought that was most likely. I wasn't really sure. but I guess that would mean NV40 would be launched at CeBit. wonder when and where R420 launches, then. anyone in the know care to share? ;)

[Ante P]

I'd take 8x "ATi-like" FSAA on NV40 over PS3.0 any day o' the weak.
Well then, March 24 is your lucky day, dearie. ;)

What happens on March 24th. :?
NV40 is due March 18th.
hm, I know CeBit and GDC both end on March 24, so I thought that was most likely. I wasn't really sure. but I guess that would mean NV40 would be launched at CeBit. wonder when and where R420 launches, then. anyone in the know care to share? ;)

"A little bit later"®

[digitalwanderer]

Hmmmm....you don't suppose ATi might be planning on implementing a software solution somehow to run the PS 3.0 shaders aka the way the 5200 is directX 9 compliant the R420 would be too?
Could be, but don't you think that the performance would be awfull? :?
Probably, but if it ain't a utilized feature than ATi will be able to slide by on this generation the same way nVidia pretty much slid thru with the FX's crappy shader performance.

[Bjorn]

Probably, but if it ain't a utilized feature than ATi will be able to slide by on this generation the same way nVidia pretty much slid thru with the FX's crappy shader performance.

I wouldn't exactly call it slid thru.

[Ante P]

Probably, but if it ain't a utilized feature than ATi will be able to slide by on this generation the same way nVidia pretty much slid thru with the FX's crappy shader performance.

I wouldn't exactly call it slid thru.

sales wise it kinda did :)

[THe_KELRaTH]

I doubt that PS3.0 will have a huge impact on games.

Now, a huge impact on workstation apps? There's a very good chance (especially since Chris Donahue hinted (http://www.beyond3d.com/forum/viewtopic.php?p=209857#209857) at some workstation apps using D3D in the not-too-distant future).

Also, is GLSLang closer to ps_2_0 or ps_3_0 (I guess the question is more "does GLSLang support branching")?

You could also say that having PS2.0 18 months ago has had little impact too but it's what brought the R300 to the top of the graphics card table.
We've seen what Nvidias' marketing is like when they're behind so it isn't hard to imagine what will happen if they can offer a product that not only is on par performance wise but also offers a more uptodate feature set. You can be sure that PS2.0 will be made to look ancient via NV demo's and new benchmarks.

Who know's - the DX9 support in UT2004 Retail could just be PS3.0 - "The way it's meant to be played"

[ClyssaN]

hmmm ... i guess the next couple months are going to be very interesting ... i'm imagining so many "deja vus" ... now i just have to pick a good seat :)

[John Reynolds]

Who know's - the DX9 support in UT2004 Retail could just be PS3.0 - "The way it's meant to be played"

And thereby exclude all R3xx and NV3x owners from using it. I don't think Epic is going to spend time coding an additional path or effects that benefit absolutely no one when the game is released here in a few weeks.

[NeoCool]

In reply to all,

Direct-X 9.0 comes with built in 3.0 VS/PS shader support, and the R420 should certainly come with full hardware compliance for 3.0 shaders...4.0 VS/PS shaders should come in 2006 with Direct-X 10.

--NeoCool :)

[digitalwanderer]

Probably, but if it ain't a utilized feature than ATi will be able to slide by on this generation the same way nVidia pretty much slid thru with the FX's crappy shader performance.

I wouldn't exactly call it slid thru.

sales wise it kinda did :)
Yup. Considering the amount of people/places screaming at the top of their lungs about how BS it was, people sure seem to still be believing it anyways. :(

I don't think ANYONE is going to come out with anything that supports PS 3.0 anytime soon, although it wouldn't shock me too much if Epic did release UT2k4 with PS 3.0 support if the R420 does it in software just so nVidia could flaunt the hell out of the benchmark that will only run on their flagship product. (I only say it wouldn't shock me because after the lengths nVidia went over the last year I wouldn't put nothing past them. :roll: )

[THe_KELRaTH]

Who know's - the DX9 support in UT2004 Retail could just be PS3.0 - "The way it's meant to be played"

And thereby exclude all R3xx and NV3x owners from using it. I don't think Epic is going to spend time coding an additional path or effects that benefit absolutely no one when the game is released here in a few weeks.

Who says Epic spent the time writing the extra code (Nvidia themselves?). It's not as though they're removing existing support in favour of DX9 rather just adding it.
Also, maybe it won't be in the initial retail release either, but a patch timed with the NV40 release. It's not as though Epic haven't offered extended support to Nvidia cards before.

[PatrickL]

But if there is no PS 3.0 support it should have an impact on sales.

9700 pro 9800 pro and 9800xt are actually powerfull enough for all games. The lack of new feature support will remove the motivation to spend so much money on a new card (or new rig if you go pci-e)

I was ready to pay a premium price and build a new rig based on pci-e and r420 in april may (for my pleasure). If news are true there is no point to do that and i ll be better waiting next fall.

[Bjorn]

Yup. Considering the amount of people/places screaming at the top of their lungs about how BS it was, people sure seem to still be believing it anyways. :(


AFAIK, Ati has sold a lot more mainstream & highend cards (DX9) then Nvidia. And i would say that the biggest reason for that is it's crappy PS 2.0 performance. Sure, the crappy FSAA iq was probably a big reason also but i doubt that it was as important as the PS 2.0 problems.

[Ante P]

In reply to all,

Direct-X 9.0 comes with built in 3.0 VS/PS shader support, and the R420 should certainly come with full hardware compliance for 3.0 shaders

--NeoCool :)

how do you figure
so PS 3.0 has been in DX9 since the start that doesn't mean NV30 and R300 support it sowhy would it mean R420 must support it?

[Miksu]

"A little bit later"®

I was hoping for a same launch date for both of the products. That would have been something.

So, if NV40 is better than Ati thought it would be, is R420 going to be postponed until they can get the clocks high enough? Or are they just going to wait a week or so after NV40's launch and then release the R420? That would make every reviewer to add "You should wait for Ati's offering" to their NV40-articles. Unless NV40-hype is so good, that everyone starts using the word "Own"..

[digitalwanderer]

Yup. Considering the amount of people/places screaming at the top of their lungs about how BS it was, people sure seem to still be believing it anyways. :(


AFAIK, Ati has sold a lot more mainstream & highend cards (DX9) then Nvidia. And i would say that the biggest reason for that is it's crappy PS 2.0 performance. Sure, the crappy FSAA iq was probably a big reason also but i doubt that it was as important as the PS 2.0 problems.
The fact that anyone bought an FX over an R3xx is enough proof that nVidia was successful. ;)

[Mr.Sparkle]

Perhaps this is the reason behind Dave's "Shader 3.0 inflection or deflection" post. If Dave knows the R420 doesn't support 3.0, but that the NV40 does (as some of Nvidia's TWIMTBP devs seem to imply), then of course, it's the first salvo in an argument to assert that 3.0 won't drive games, won' t be that relevent, and doesn't matter.

So what your really trying to say is you believe Dave is just an extension of ATI's marketing team? Or am I interpreting this part of your post wrong? Thanks.

OT: I wonder when this thread will end, I walk away for 20 mins and theres already another 15 to 20 posts to catch up on.

[Bjorn]

The fact that anyone bought an FX over an R3xx is enough proof that nVidia was successful. ;)

And Ati was successful much in the same way with the R8500 series then i assume ?

[John Reynolds]

But if there is no PS 3.0 support it should have an impact on sales.

9700 pro 9800 pro and 9800xt are actually powerfull enough for all games. The lack of new feature support will remove the motivation to spend so much money on a new card (or new rig if you go pci-e)


Large performance gains alone will sell the new boards. How many games were there using 2.0 shaders when the 9700 Pros originally launched in the fall of '02?

[Magic-Sim]

None, but performance gain were tremendous.

[Florin]

The fact that anyone bought an FX over an R3xx is enough proof that nVidia was successful. ;)

I know you're kidding.. you are kidding, right :)

I mean sure not all of Nvidia's FX customers are dual head gamers or Linux users. But some are..

[Joe DeFuria]

I'll just repeat what I said a month ago, back in the inflection/deflection thread:

My personal thoughts are that from a comsumer perspective, PS 3.0 will in fact be to DirectX 9 what PS 1.4 was to DX 8.

That is, there really won't be any games that make you say "I'm really glad I have a PS 3.0 card, instead of a PS 2.0 card."

This is not to say that PS 3.0 cards won't be "worth it"...just not particularly because of the features...but presumably the performance increase that a new generatoin brings. Oh, and of course, we'll need a PS 3.0 card to run Humus' latest demos, so there's always that.

PS 3.0 will be nice to play with for developers and hobbyists, but the "baseline" for DX9 has been set at PS 2.0. All else being equal, having a 3.0 vs. a 2.0 card is obviously better. But all else won't be equal. (edit: to be clear, I have no idea in exactly what way things won't be equal...they just won't be, as it always is with two different architectures.) As far as NV40/R420 is concerned, it will come down to

1) Compatibility
2) PS 2.0+ performance
3) Image quality. (This encompasses many things, including presence or lack of shader replacements, AA and Aniso filtering, etc.)

[Dave Baumann]

Perhaps this is the reason behind Dave's "Shader 3.0 inflection or deflection" post. If Dave knows the R420 doesn't support 3.0, but that the NV40 does (as some of Nvidia's TWIMTBP devs seem to imply), then of course, it's the first salvo in an argument to assert that 3.0 won't drive games, won' t be that relevent, and doesn't matter.

So what your really trying to say is you believe Dave is just an extension of ATI's marketing team? Or am I interpreting this part of your post wrong? Thanks.

At the moment that’s the only way I conclude that sentiment, and I also find it thoroughly offensive. However, I do wonder if it marks more towards DC's own prejudices.

In making the claim he has completely glossed over the content of the piece, without offering comment on the piece as to whether there the precedents bear any historical accuracy, ignores the fact that the next news post highlights a potential use of Shader 3.0 and makes gross assumptions as to when I may or may not have found out any real details on R420. The fact of the matter is when I wrote that piece my "rumour indicator" was pointing towards the support of 3.0.

The point of the piece was to: a.) raise the awareness of "inflection points" because they had not been discussed in those terms here much previously, b.) raise a discussion point about shader 3.0 and whether their adoption would bear any similarities to PS1.4, and c.) ask if any developers if they had any plans for shader 3.0 - the piece scored on all 3 points (especially the final one to my surprise). If this were and "opening salvo", as DC rather disgustingly puts it, then I wouldn’t have drawn more attention to Dany’s subsequent comments by posting them on the front page (and having them linked to numerous news sites). Personally I’m pretty pissed by DC’s comments here and would like an apology for making such gross assumptions and insinuations.

[Mr.Sparkle]

I also found it a very condesending & insulting post as well and am quiet surprised no one called him on that part of the post before I did. DC seems to be a fairly smart person when it comes to 3D knowledge, but comments like that just make my eyes roll and just want avoid his posts completely.

[Evildeus]

But knowing the specs, would make your wonders more insistent, doesn't it? At least it would for me :).

Anyway, Dave or any developper over here, what is your feeling concerning PS 3.0? Same path as PS 1.4, a more important feature, anything else?

PS: It doesn't matter wether you know or not the R420/Nv40 specs (actually i would prefer that you don't know), i just want your view, ie the view of someone with some knowledge and good connections with developpers :)

[Sxotty]

I really didn't get that at all Dave, but I am sure DC can share his own feelings on what he meant.

[Joe DeFuria]

Anyway, Dave or any developper over here, what is your feeling concerning PS 3.0? Same path as PS 1.4, a more important feature, anything else?

http://www.beyond3d.com/forum/viewtopic.php?t=9854 :wink:

[Dave Baumann]

I really didn't get that at all Dave, but I am sure DC can share his own feelings on what he meant.

Well, I can currently only see that he's either suggesting that or I'm a too much of a dolt not to understand when I'm trying to be used by PR if the sentiment of the post came from ATI. Either way, I’m not too enamoured with the intimation.

[Joe DeFuria]

I really didn't get that at all Dave, but I am sure DC can share his own feelings on what he meant.

FWIW, I saw DC's comment the same way Dave and Sparkle did. Doesn't seem to be much in the way of misinterpretation, IMO. Inentional or not, is something DC would have to expand on.

[PaulS]

I really didn't get that at all Dave, but I am sure DC can share his own feelings on what he meant.

Well, I can currently only see that he's either suggesting that or I'm a too much of a dolt not to understand when I'm trying to be used by PR if the sentiment of the post came from ATI. Either way, I’m not too enamoured with the intimation.

No idea how people are reading it as him being an extension of the ATi Marketting department. The comment was merely saying that - if at the time of writing - Dave knew PS3.0 wasn't supported by R420, then only one of the two major IHVs would support it, raising the question of whether it would be a major inflection point, or just fall by the wayside due to a lack of support from ATi.

[Joe DeFuria]

No idea how people are reading it as him being an extension of the ATi Marketting department. The comment was merely saying that - if at the time of writing - Dave knew PS3.0 wasn't supported by R420, then only one of the two major IHVs would support it, raising the question of whether it would be a major inflection point, or just fall by the wayside due to a lack of support from ATi.

Because DC specifically said this:

...then of course, it's the first salvo in an argument to assert that 3.0 won't drive games, won' t be that relevent, and doesn't matter.

DC didn't state it like it was the first "round" of argumentation for both sides...but that it was the first round that would support ATI's position.

[thatdude90210]

I still remember when right before R300's release, one or two Nvidia supporters here trying to stick the Dx8.5 label on it.

[Joe DeFuria]

On a related note...

If someone wants to devise a "what if" scenario for the next holy war...I PRAY that something like this doesn't happen:

1) FutureMark releases a PS 3.0 update of their 3DMark '03 benchmark.
2) It includes a PS 3.0 test which contributes to the final score.

[Evildeus]

Anyway, Dave or any developper over here, what is your feeling concerning PS 3.0? Same path as PS 1.4, a more important feature, anything else?

http://www.beyond3d.com/forum/viewtopic.php?t=9854 :wink:
Thanks, not much from Dave on PS 3.0 at least, but interesting for the replies!

[Ante P]

I still remember when right before R300's release, one or two Nvidia supporters here trying to stick the Dx8.5 label on it.

heh DX 8.5 is reserved for Intels upcoming Expreme Graphics 3

[Sabastian]

One thing for sure is that if ATi chooses not to support PS3.0 fully then it may very well end up being a deflection point. That might be the case particularly where ATi has begun to overtake nvidia in market share. I do think this next year will amount to an (continuation? ;) ) all out no holds bared street fight... If ATi does not support the extension of PS2 to PS3 it is more then likely the spec will be much the same as PS1.4 was to DX8. If nvidia can spur on allot of software development for PS3 then it might amount to something worth while. But that won't be until after the next variety of DX is on the market. Unless development for PS3.0 is taken on mass by developers it will likely end up much the same way PS1.4 went. That is not a horribly big deal IMO until benchmarks such as 3DMark start accrediting PS3.0 support in some way. Until then it won't be that much of a deal and maybe even after that it won't be that important.... The question for me personally is PS3 going to be interesting in terms of its output? Have to wait and see for that.

off to the store to buy a years supply of popcorn. ;)

[Ante P]

On a related note...

If someone wants to devise a "what if" scenario for the next holy war...I PRAY that something like this doesn't happen:

1) FutureMark releases a PS 3.0 update of their 3DMark '03 benchmark.
2) It includes a PS 3.0 test which contributes to the final score.

that would be sooooo much fun
but I also think that Futuremarks popularity would sink
a lot of reviewersstopped using 3dmark last year and I bet even more would aftersuch a move

then again, if they provide the only PS 3.0 test out there...

Humus would need to save us with a benchmarkable demo :)

[Joe DeFuria]

a lot of reviewersstopped using 3dmark last year and I bet even more would aftersuch a move..

Not true, becuase if that happened, nVidia would be praising 3DMark from the rooftops, at which point the llamas would "suddenly have a change of heart" about the benchmark. ;)

[Neeyik]

then again, if they provide the only PS 3.0 test out there...

Except that there are PS3.0 demos available already:

http://3d.rightmark.org/index.shtml

http://www.pvrdev.com/pub/PC/eg/index.htm

[Magic-Sim]

On a related note...

If someone wants to devise a "what if" scenario for the next holy war...I PRAY that something like this doesn't happen:

1) FutureMark releases a PS 3.0 update of their 3DMark '03 benchmark.
2) It includes a PS 3.0 test which contributes to the final score.

They didn't for PS 1.4... who said biased ?

After all the problems nVIDIA gave them, they would surely be keen to follow nvidia a little....

B3D is in the béta program, isn't it ?

[Dave Baumann]

B3D is in the béta program, isn't it ?

Yes. Although I can't comment on what Futuremark are doing next I would suggest that Joe's scenario is highly unlikely.

[Magic-Sim]

a lot of reviewersstopped using 3dmark last year and I bet even more would aftersuch a move..

Not true, becuase if that happened, nVidia would be praising 3DMark from the rooftops, at which point the llamas would "suddenly have a change of heart" about the benchmark. ;)

Is nVIDIA a kind of a sect ? ;)

[Magic-Sim]

B3D is in the béta program, isn't it ?

Yes. Although I can't comment on what Futuremark are doing next I would suggest that Joe's scenario is highly unlikely.

Well.... I'm a bit lazy minded this evening, do you mean that, there won't be an update to 3DMark other than cheat-disabling ? :?:

[PaulS]

Well.... I'm a bit lazy minded this evening, do you mean that, there won't be an update to 3DMark other than cheat-disabling ? :?:

Not including score-changing tests which one IHV can't run != Nothing more than cheat-disabling.

[DemoCoder]

No need to speculate on the meaning of my post, Dave interpreted it correctly. My theory is, no matter how much one can try to remain objective, one can be persuaded to be nice or favorable (even unconsciously) towards another company if they are nice to you. For example, giving one early pre-release information, "scoops", per se. You see it even in the NYT as reporters treat their sources with kidgloves, lest those sources be cut-off.


That said, I apologize to Dave for questioning his credibility. I remember Rev implying that Dave has secret info ("those in the know") and has been gradually trying to leak info, but without actually saying it. :) Then when the no-3.0 post showed up, I remembered Dave's "3.0 inflection or deflection" article which seem to fit Rev's statement, and everything else fell together, so Dave's article to me looked like an early attempt to talk about the R420, but without talking about it. :)

Apparently, Dave didn't know about the R420's specs when he wrote that article. I hereby retract my theory. Sorry, Dave.

[Skinner]

Why are so many already convinced the R420 would be faster in ps2?

Is it based on inside rumors or the fact ATI uses their transistors only for ps2 performance instead of flexibility (=+ps3)?

I mean, the real truth was always hidden until the reaviews where out.

I mean, what do we really know about the NV40, apart from a very few rumors?

[digitalwanderer]

you are kidding, right :)

Well, at least half-kidding...but there is an awful lot of truth in the statement too. ;)

[Magic-Sim]

Well.... I'm a bit lazy minded this evening, do you mean that, there won't be an update to 3DMark other than cheat-disabling ? :?:

Not including score-changing tests which one IHV can't run != Nothing more than cheat-disabling.

Okay, that's fairly explicit, thanks to you two ;)

[Megadrive1988]

Like I said I would be surprised if ATi were not to support dx next broadly, particularly if you keep in mind the work they are doing with MS with regards to the xbox2

^^ I tend to agree with this ^^

[Mariner]

If, as rumoured, R420 doesn't support PS 3.0 then it better have one heck of a lot more performance than NV40 in PS 2.0 shaders.

I assume that it will support VS 3.0 as we've heard rumours that R420 uses the VS capabilities of the 'original' R400?

Even if R420 has much better image quality (through AA and AF) if PS 2.0 performance is only slightly higher than NV40, ATI will be in trouble for this generation of chips. As can be seen from most reviews on the net, as long as the check boxes for AA and AF are filled, little attention is paid to image quality. We've seen that because NV34 is a DX9 chip it has greatly outsold the competing DX8 Radeon chips despite it's rather poor performance. It just goes to show that extra features, however unusable, have a great effect on how well a product sells.

All IMO, of course.

[Magic-Sim]

In the mean time, if ATI don't play the cheat/play fool on people like some other IHV's, they still stand a chance to limit damage.

[Heathen]

In the mean time, if ATI don't play the cheat/play fool on people like some other IHV's, they still stand a chance to limit damage.

especially if they enhance their already great AA IQ and tweak the AF IQ, it should be a stormer of a card. A lot depends on how good the NV40 is in PS2 & IQ and how the game developers react really.

If Nvidia ca produce a car5d which is an all round better bet than the R420 I'll go with them. If not, ATi's got my money for another round.

[Cyborg]

Even if R420 has much better image quality (through AA and AF) if PS 2.0 performance is only slightly higher than NV40, ATI will be in trouble for this generation of chips

Arent we forgetting a little something called price here? If R420 features faster PS2.0 performances, better image quality and lower price than Nv40, why buy NV40? Im talking about gamers here, all they'll see is performances and price in the end, image quality is not really high on the list of priorities and let alone PS3.0. Its sad but its like that, geforce3 didnt suffer against the 8500 because it didnt support PS1.4

[PaulS]

Arent we forgetting a little something called price here? If R420 features faster PS2.0 performances, better image quality and lower price than Nv40, why buy NV40? Im talking about gamers here, all they'll see is performances and price in the end, image quality is not really high on the list of priorities and let alone PS3.0. Its sad but its like that, geforce3 didnt suffer against the 8500 because it didnt support PS1.4

Prices will likely be roughly identical, and the GF3 succeeded over the 8500 because of the latter's bad driver support (at least initially) - the strong nVidia/GF brand name also helped.

Why are so many already convinced the R420 would be faster in ps2?

Two things really: ATi had a far better base from which to build from, and secondly they concentrated significant resources on PS2.0 performance. Did nVidia do the same (and more, to make up for the initial deficit)? I'd suggest not.

[digitalwanderer]

the GF3 succeeded over the 8500 because of the latter's bad driver support (at least initially) - the strong nVidia/GF brand name also helped.
Also the lack of trilinear filtering with aniso on, that was my personal grudge against the 8500.

[nggalai]

Good evening everybody,

as no-one has provided a proper translation, yet, I'll do a free-form paraphrasing:
i asked over at 3dc about the proof or source of the story...

quote from leonidas from 3dc
Die Quelle ist mehr als direkt, direkter geht es gar nicht mehr.
Nicht nur die Position der Quelle ist direkt, sondern auch die Art,
wie sie selber zu den Informationen gekommen, ist ultimativ direkt.
Eingeweihte ahnen womöglich, was ich meine :-).
"The information came to us directly, it's nigh impossible to get that sort of info in a more direct way. Not only is the person in question a direct contact, but the way he/she got the information is as straightworward and firsthand as it gets. Those in the loop might know what I'm talking about. :-)"

We're also taught that R500 will be PS4.0 capable. Gods know what that means, I wasn't even aware that the PS4.0 specs were done, yet.

93,
-Sascha.rb

[Magic-Sim]

Looks like more and more FUD to me....

[nggalai]

Anyway, Dave or any developper over here, what is your feeling concerning PS 3.0? Same path as PS 1.4, a more important feature, anything else?Devs I talked with about PS3.0 were, on the whole, quite giddy to get their hands on it. I am not a tech guy myself, so I understood only a fraction of it, but the word "texkill" was featured prominently in those discussions. Apparently, PS3.0 is considered more of an elegant way to get rid of unwanted work in existing HLSL shaders than an eye candy update. Demirug did a short worst-case (for PS2.0, best case for PS3.0) example program that should run about 8x more effective on PS3.0 than PS2.0, on the same hardware, simply because it can skip stuff that's not needed for the final result. HLSL shaders apparently don't take more than one more compile run to make them use those PS3.0 extras. Perhaps a dev here can get more into it.

We'll probably see PS3.0 support in the not too distant future, simply for speed / efficiency reasons. As a last restort, there's still NV's Unified Shader Compiler. ;)

93,
-Sascha.rb

[Evildeus]

Thx nggalai 8)

[Dave Baumann]

Devs I talked with about PS3.0 were, on the whole, quite giddy to get their hands on it. I am not a tech guy myself, so I understood only a fraction of it, but the word "texkill" was featured prominently in those discussions.

Are you sure that was the right instruction? It seems to be a part of PS2.0:

http://www.beyond3d.com/articles/nv30r300/index.php?p=8#psinstr

[edit] In fact, according to that, it was only NV30 that didn't / doesn't support TEXKILL specifically. Anyone have any update in that?

[Joe DeFuria]

They didn't for PS 1.4... who said biased ?

That's exactly my point...they DIDN'T do that for PS 1.4. So if they DID do that for PS 3.0, they'd have some explaining to do. ;)

[Dave Baumann]

No need to speculate on the meaning of my post, Dave interpreted it correctly. My theory is, no matter how much one can try to remain objective, one can be persuaded to be nice or favorable (even unconsciously) towards another company if they are nice to you.

We're not here as a PR outlet for anyone. Sometimes the purpose of those editorial posts are just to provoke some toughts and discussion.

I hereby retract my theory. Sorry, Dave.

Accepted. Thanks.

[digitalwanderer]

Sometimes the purpose of those editorial posts are just to provoke some toughts
I've had quite a few toughts whilst here. 8)

[digitalwanderer]

Good evening everybody,

as no-one has provided a proper translation, yet, I'll do a free-form paraphrasing:
i asked over at 3dc about the proof or source of the story...

quote from leonidas from 3dc
Die Quelle ist mehr als direkt, direkter geht es gar nicht mehr.
Nicht nur die Position der Quelle ist direkt, sondern auch die Art,
wie sie selber zu den Informationen gekommen, ist ultimativ direkt.
Eingeweihte ahnen womöglich, was ich meine :-).
"The information came to us directly, it's nigh impossible to get that sort of info in a more direct way. Not only is the person in question a direct contact, but the way he/she got the information is as straightworward and firsthand as it gets. Those in the loop might know what I'm talking about. :-)"

We're also taught that R500 will be PS4.0 capable. Gods know what that means, I wasn't even aware that the PS4.0 specs were done, yet.

93,
-Sascha.rb

Thanks Nggalai, that helps a lot! :D

[Xmas]

Devs I talked with about PS3.0 were, on the whole, quite giddy to get their hands on it. I am not a tech guy myself, so I understood only a fraction of it, but the word "texkill" was featured prominently in those discussions.

Are you sure that was the right instruction? It seems to be a part of PS2.0:

http://www.beyond3d.com/articles/nv30r300/index.php?p=8#psinstr

[edit] In fact, according to that, it was only NV30 that didn't / doesn't support TEXKILL specifically. Anyone have any update in that?
TEXKILL is already a part of PS1.0, though its usage is a bit different in PS1.4 and higher. TEXKILL and KILL are the same, it's TEXKILL in DirectX and KILL in OpenGL, IIRC. This is because DX has hard limits on shader instructions and differentiates between tex and arithmetic instructions. All instructions beginning with TEX are considered texture instructions.

[Dave Baumann]

TEXKILL and KILL are the same, it's TEXKILL in DirectX and KILL in OpenGL, IIRC.

That would make sense - most of the NV30 information came from the prerelease OpenGL documentation. I'll update that.

[digitalwanderer]

Ok, here's something that makes me have to ask a dumb question. I read the following post in this thread here (http://www.beyond3d.com/forum/viewtopic.php?t=10368&start=20):
As for the "simplier shader implementation" :roll: I guess they are trying to say that because ATI only supports the 24 bit standard at good speed and quality, it must be "simpler" than the more complex Nvidia implementation that offers FP16 speed at low quality, or FP32 high quality at low speed.
No, they are not.
That's not a very helpful reply :)

To clarify; R300 supports VS2.0 and PS2.0. nv30 supports VS2.0 Extended and PS2.0 Extended. If you go and look up the specs for those versions, you'll see that PS/VS 2.0 is simpler than 2.0 Extended.

In fact 2.0 Extended is much closer to 3.0, than it is to 2.0.
So assuming the "no PS 3.0 hardware support" theory/rumor/whatever is true, will the R420 support 2.0 extended in hardware?

[DemoCoder]

PS2.x allows dynamic branching (it's not required, it's a cap), gradient, arbitrary swizzle, etc but they are optional. It's not clear to me whether ATI will support these are not. The major requirement for 3.0 is 512 instructions minimum, and the removal of texture read limits, plus increased number of registers.

It seems like according to the spec, it is possible to omit dynamic branching, yet still claim shader model 3.0. (although it was be a hollow claim IMHO)

For V3.0, the major change of course, is vertex textures.

[Magic-Sim]

Well, it seems like it is urgent to wait an official/direct comment...

[digitalwanderer]

Well, it seems like it is urgent to wait an official/direct comment...
Wait? :|

Pfffft! :P

Why wait for something official when we can have sooo much fun speculating? 8)

[Hyp-X]

"The information came to us directly, it's nigh impossible to get that sort of info in a more direct way. Not only is the person in question a direct contact, but the way he/she got the information is as straightworward and firsthand as it gets. Those in the loop might know what I'm talking about. :-)"

Isn't it:
"The information came to us directly from nVidia's PR department..."

[MuFu]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

[Demirug]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

[digitalwanderer]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).
"Compliant" and not "compatible"...right?

[Mr.Sparkle]

Well, it seems like it is urgent to wait an official/direct comment...
Wait? :|

Pfffft! :P

Why wait for something official when we can have sooo much fun speculating? 8)

Exactly! :) The boards are always better during the "Silly Season".

More rumours please! :wink:

[T2k]

Man, if this rumor is true, it's going to be an unholy war on this board. Why? Assume hypothetically Nvidia delivers 3.0, with 2.0 performance "fixed". Everyone is going to come down hard on NVidia and say 3.0 is wasted, no one's gonna use it, too early, etc etc (until R500 arrives of course).

That seems like a pretty weird assumption - I would say the tide here will turn against ATi for not driving their feature set forward (Assuming nVidia's PS 2.0 performance is on or close to a par with ATi's this time around that is).

Anyway, if this rumour is true, it's pretty much suicide on ATi's part, from a marketing position at least.

Agreed. That would be recognized as a sign of return of the lazy, arrogant style we've seen from the old ATI in the 90s...

It's just BS.

[John Reynolds]

Allow me to suggest something. One of the main reasons cited for NV30's lackluster performance was because of NVIDIA's prior focus on the X-box. . .the engineering resources required for that sapped them for their next generation PC part. ATI's movement of R400 from late last year to the X-Box 2/R500 could have been to pre-empt the fact that they are going to have to divert resources to meet the X-Box deadlines, so rather than take the hit later they are doing it now. In other words, release the next part (R420) with a lower specification that's build primarily upon their existing architecture and therefore requires less engineering time and resources to finish, and where the key inflection point isn't going to be architectural abilities (i.e. Shader 3.0 is an interim spec) but a relatively small change to PCI-Express, while ensuring that the technology isn't lacking after the X-Box is released. In short, the hit that NVIDIA took post-X-Box, ATI are trying to minimize by taking it pre-X-box2.

Just a thought. And guessing. . .I do not know official specs for either R420 or NV40.

[Fodder]

We've seen that because NV34 is a DX9 chip it has greatly outsold the competing DX8 Radeon chips despite it's rather poor performance. It just goes to show that extra features, however unusable, have a great effect on how well a product sells.
The FX5200 non-Ultra is solidly better than the R9200 non-Pro in DX7 and DX8.0 games (ie. 99% of titles). I'm not sure about the Pro and Ultra, but down here at the bottom of the world they never made it to stores. And who knows, on a shit slow budget card you might appreciate the cheats. ;)

[Magic-Sim]

Well, it seems like it is urgent to wait an official/direct comment...
Wait? :|

Pfffft! :P

Why wait for something official when we can have sooo much fun speculating? 8)

Well, we could say, wait before canning R420 8)

I'm still confident in ATI. I just hope Marketting and cheatwaring isn't going to inflate another time in the chameleon camp :D

[MuFu]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

Sorry, I should clarify - I meant "Demirug's thread and the ensuing surge of presumptious forum activity". :lol:

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).
"Compliant" and not "compatible"...right?

Yep - compliant, "supports PS3.0"; whatever.

If you remember that original "Loci" (sic) post I made here and on R3D last June, I mentioned it probably wouldn't support PS3.0 as it was based on R3x0. Should have stuck to my guns, maybe; the source of that information certainly couldn't be much better (and that little detail was almost a C&P - not my own opinion). Maybe they do have some sort of software-based compliancy now though. :?: If you remember, the idea of both IHVs concentrating on a very fast PS2.0 and only "token" PS3.0 solution was touched upon in that thread.

MuFu.

[Magic-Sim]

By the way, would PS 3.0 run at acceptable speeds with a 2 times faster part (eg in PS 2.0) than R360 ?

[Ailuros]

Allow me to suggest something. One of the main reasons cited for NV30's lackluster performance was because of NVIDIA's prior focus on the X-box. . .the engineering resources required for that sapped them for their next generation PC part. ATI's movement of R400 from late last year to the X-Box 2/R500 could have been to pre-empt the fact that they are going to have to divert resources to meet the X-Box deadlines, so rather than take the hit later they are doing it now. In other words, release the next part (R420) with a lower specification that's build primarily upon their existing architecture and therefore requires less engineering time and resources to finish, and where the key inflection point isn't going to be architectural abilities (i.e. Shader 3.0 is an interim spec) but a relatively small change to PCI-Express, while ensuring that the technology isn't lacking after the X-Box is released. In short, the hit that NVIDIA took post-X-Box, ATI are trying to minimize by taking it pre-X-box2.

Just a thought. And guessing. . .I do not know official specs for either R420 or NV40.

Even if it's just speculation for the time being and if true it'll take more than a year to get an even half way/indirect verification, it does make more than just sense to me.

I don't have a single doubt that a console contract - even if it's just a license - can seriously distract/disorientate any IHV's development timetables. Here's to say (and the PVR folks will probably hate me for saying it) that I'm actually glad that IMG didn't land a console deal. Those that remember how long it took the Neon250 to get on shelves, will probably know what I'm talking about.

Next best thought/point would be that IMG is small in size and resources compared to large IHVs like ATI/NV, yet NVIDIA knows very well how the XBox deal affected their roadmaps of the past. I have good reason to believe that Ti200/500 were never actually planned and there had been official statements that NV30 was to be their spring 2002 product.

What the future may hold I prefer to see upcoming products to be tested/analyzed to have a better picture, since paperspecs and bold marketing claims that circulate behind the curtains leave me personally fairly in the cold.

One aspect I'm most certainly glad about and it was due ATI's large success in 2002/3, is that exclusive contracts ceased to exist at large vendors. It might seem of small importance to some, but at least IMHO it opened the doors to higher degrees of competition, wherever it may come from.

[T2k]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

Ho-ho-ho... :lol:

Nice story about 3dcenter's credibility... :lol:

[AlphaWolf]

The FX5200 non-Ultra is solidly better than the R9200 non-Pro in DX7 and DX8.0 games (ie. 99% of titles).

Not really true from what I have seen. It wins some and loses some. It wins the checkbox battle but doesn't win the framerate battle.

[Ailuros]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

Ho-ho-ho... :lol:

Nice story about 3dcenter's credibility... :lol:

I think you're jumping to conclusions too fast. 3DCenter's recent article about the R4xx line of products, is according to the author's (and administrator's) claim, not based on the thread in question or Demirug's findings in it.

[T2k]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

Ho-ho-ho... :lol:

Nice story about 3dcenter's credibility... :lol:

I think you're jumping to conclusions too fast. 3DCenter's recent article about the R4xx line of products, is according to the author's (and administrator's) claim, not based on the thread in question or Demirug's findings in it.

Did I say that? I did not.

It's just very interesting... such a coincidence, huh? :roll:

[Fodder]

Not really true from what I have seen. It wins some and loses some. It wins the checkbox battle but doesn't win the framerate battle.
R9000/9200 NP seems to be a bit faster than the regular MX440 (plus DX8 of course), while the reference FX5200 is about on par with the faster again MX440-8x, plus DX8 and checkbox DX9.

It may well depend on benchmarks, but everything I've seen points to the 5200 being the superior card.

http://www.nordichardware.com/reviews/graphiccard/2003/Budget_Roundup/index.php


Rampant blind speculation on R420 is more interesting than the performance of budget cards though. :)

[Geo]

Hmm. Maybe it just isn't in the first release drivers yet, but the hardware can support it?

Altho, to argue against my own theory, Dave's comment awhile back about the "tug of war" over the 9.1 moniker would suddenly make sense if R420 would not qualify for that label.
Could you expand on that thought a bit please? It's tickling something in my brain and I wanna see if I can figure out what the tickle is.

Yowza. This one took off since last I looked.

Well, Digi, what I'm pointing at is this thread here: http://www.beyond3d.com/forum/viewtopic.php?t=9837&postdays=0&postorder=asc&highlight=dx9%201&start=20

And this post by Wavy:

Put it like this. There is some pushing and pulling going on, and no-doubt some serious lobbying to Microsoft. It'll become clearer why in the fullness of time.

If my highly speculative theory above is true, then this would be evidence to support it. The idea being that many of the less informed (Anand comes to mind) seemed to be pushing the idea of ps/vs 3.0 compliance being the fabled DX9.1. From a marketing perspective, if this were true, and ATI could not meet the ps/vs 3.0 specs, then their card would NOT be DX9.1 compliant and it would suddenly become clear why there was "pushing and pulling" on this point (sorry, I originally misremembered as "tug of war") with MS.

And, by the way, I don't necessarily believe my own theory. I just saw two interesting pieces of data that looked like they might be related to each other. One of my other hobbies is researching the Confederate secret service during the Civil War. That mind set and experience sometimes allows me intuitive flashes on partial data (the only kind we get in that other field). . and huge clunkers.

That help?

[AlphaWolf]

Not really true from what I have seen. It wins some and loses some. It wins the checkbox battle but doesn't win the framerate battle.
R9000/9200 NP seems to be a bit faster than the regular MX440 (plus DX8 of course), while the reference FX5200 is about on par with the faster again MX440-8x, plus DX8 and checkbox DX9.

It may well depend on benchmarks, but everything I've seen points to the 5200 being the superior card.


Well in toms vga charts for example, the 9200 wins as many dx8 benches as it loses to the 5200, as I said it wins the checkbox battle but for framerate I'd give it no better than a draw with a 9200.

[nelg]

Allow me to suggest something. One of the main reasons cited for NV30's lackluster performance was because of NVIDIA's prior focus on the X-box. . .the engineering resources required for that sapped them for their next generation PC part. ATI's movement of R400 from late last year to the X-Box 2/R500 could have been to pre-empt the fact that they are going to have to divert resources to meet the X-Box deadlines, so rather than take the hit later they are doing it now. In other words, release the next part (R420) with a lower specification that's build primarily upon their existing architecture and therefore requires less engineering time and resources to finish, and where the key inflection point isn't going to be architectural abilities (i.e. Shader 3.0 is an interim spec) but a relatively small change to PCI-Express, while ensuring that the technology isn't lacking after the X-Box is released. In short, the hit that NVIDIA took post-X-Box, ATI are trying to minimize by taking it pre-X-box2.

Just a thought. And guessing. . .I do not know official specs for either R420 or NV40.
How much work will ATI save by supplying only I.P. instead of silicone?

[Joe DeFuria]

How much work will ATI save by supplying only I.P. instead of silicone?

That depends on target breast size. 8)

[nelg]

:oops: :lol:

[Tim Murray]

John, I don't quite buy your Xbox/Xbox 2 explanation--there were more reasons to the NV30's total failure than the Xbox contract: TSMC screwing up, an ATI that was reenergized by the ArtX acquisition, and hubris hubris hubris ("NOBODY needs a 256-bit memory bus!"). The Xbox probably had some influence on the NV30's failure, but I think ATI might be gambling on the way the industry is moving like NVIDIA did with the NV30. Maybe they're betting that PS3.0 will be the equivalent of PS1.4 (in other words, nobody will really care) and that PS2.0 will dominate until DXNext. It's not stupid, but I would say that it's risky.

[Joe DeFuria]

ATI's movement of R400 from late last year to the X-Box 2/R500 could have been to pre-empt the fact that they are going to have to divert resources to meet the X-Box deadlines, so rather than take the hit later they are doing it now.

Yes, that makes complete sense.

From what I heard though, the original R400 (now R-500) turned out to be too ambitious for production on the target process (0.13u), regardless of X-Box2 or not. In other words, ATI would have needed an "inbetween" part after the R300 core, and before the R400/500 core anyway.

That doesn't change the rest of your post, which I pretty much agree with though. Basically, ATI's only real choice was to bring out a new part heavily based on existing tech, otherwise run a big risk of being late. ATI can "Get away" with this strategy, because the R300 core is extremely solid, and for all intents and purposes the PS 2.0 reference design. (It's the same reason why nVidia could get away with not going to PS 1.4 in the DX8 generation.)

[mczak]

Not really true from what I have seen. It wins some and loses some. It wins the checkbox battle but doesn't win the framerate battle.
R9000/9200 NP seems to be a bit faster than the regular MX440 (plus DX8 of course), while the reference FX5200 is about on par with the faster again MX440-8x, plus DX8 and checkbox DX9.

It may well depend on benchmarks, but everything I've seen points to the 5200 being the superior card.


Well in toms vga charts for example, the 9200 wins as many dx8 benches as it loses to the 5200, as I said it wins the checkbox battle but for framerate I'd give it no better than a draw with a 9200.
You mean this one, http://www6.tomshardware.com/graphic/20031229/ ?
Actually, I'd say the FX5200 loses. The trend seems to be (simplified) that the newer the game, the faster the 9200 gets relative to the fx5200. The 9200 loses badly in QuakeIII (which is probably the oldest used game there, additionally it's also using OpenGL, Nvidia's strong domain), loses slightly in some other games, but sometimes even wins against the fx5600. Unfortunately I've not seen a review with these cards using Max Payne2, but I have a feeling that the 9200 would trounce the fx5200 in that title (since it's using PS 1.4).
Other reviews comparing 9200 and fx5200 can be found here,
http://www.anandtech.com/video/showdoc.html?i=1933&p=2 - unfortunately only with the fx5200ultra, but even against this card the 9200 often looks quite good.
Though I wouldn't say the fx5200 only winning the "checkbox battle", it actually offers better options for AA/AF. AA is faster, and AF higher quality (the hacked trilinear mode seems a good compromise for this market segment, compared to the bilinear AF with the 9200).
So, for the same price, the FX5200 might be (or not, depending on your favourite games...) a better deal than a 9200. I believe though the 9200 is cheaper.

[digitalwanderer]

I think ATI might be gambling on the way the industry is moving like NVIDIA did with the NV30. Maybe they're betting that PS3.0 will be the equivalent of PS1.4 (in other words, nobody will really care) and that PS2.0 will dominate until DXNext. It's not stupid, but I would say that it's risky.
Yup.

Basically, ATI's only real choice was to bring out a new part heavily based on existing tech, otherwise run a big risk of being late. ATI can "Get away" with this strategy, because the R300 core is extremely solid, and for all intents and purposes the PS 2.0 reference design. (It's the same reason why nVidia could get away with not going to PS 1.4 in the DX8 generation.)
Yup.

So when does everyone think we'll find out for sure on this one way or another?

[Tim Murray]

So when does everyone think we'll find out for sure on this one way or another?
We'll know for sure once it launches--that's it.

[Fodder]

So when does everyone think we'll find out for sure on this one way or another?
We'll know for sure once it launches--that's it.
But will we know the reasoning?

[digitalwanderer]

So when does everyone think we'll find out for sure on this one way or another?
We'll know for sure once it launches--that's it.
But will we know the reasoning?
Sure, in the fullness-o-time.

Until then, we speculate. 8)

[T2k]

This is fun, after all, isn't it? ;)

Edit: missing word :oops:

[Megadrive1988]

another thing, I wonder if R420's pixel shaders are beyond the 2.0+ that NV30 has, even if short of 3.0

not that it matters.


PS 2.0 will probably be the baseline for a few years, until DX10 comes out.

[Brent]

This is fun, after all, isn't it? ;)

Edit: missing word :oops:

definitely, i love reading good informed speculation

then i can compare it to the real information and see who was right :D lol

[Tim Murray]

This is fun, after all, isn't it? ;)

Edit: missing word :oops:

definitely, i love reading good informed speculation

then i can compare it to the real information and see who was right :D lol
but where's the fun in actually determining who was right and who was wrong? the whole point of the post-launch information flood is to spin your original position in such a way that it makes you look right after all ;)

(PS--just read your Sapphire 9600XT/9800XT review. nice job. I hate to say it, though, but I'm so ripping off the FRAPS line graph stuff. best thing ever. holy crap, did I just say that an [H] review was good? the times, they are a-changin')

[Brent]

This is fun, after all, isn't it? ;)

Edit: missing word :oops:

definitely, i love reading good informed speculation

then i can compare it to the real information and see who was right :D lol
(PS--just read your Sapphire 9600XT/9800XT review. nice job. I hate to say it, though, but I'm so ripping off the FRAPS line graph stuff. best thing ever. holy crap, did I just say that an [H] review was good? the times, they are a-changin')

oh noes :shock:

[Tim Murray]

oh noes :shock:
what, the stealing of the graph idea (and credit will be given, of course) or the fact that [H] is actually writing some of the best video card reviews out there? jeez, that sounds so weird. I remember a few months back when [H] was doing the brilinear articles and making excuses for NVIDIA and yadda yadda yadda.. strange times.

then again, maybe it's just me having a fever and confusing reality with Dig's tirades.

[Brent]

oh noes :shock:
or the fact that [H] is actually writing some of the best video card reviews out there?

that one

[Althornin]

oh noes :shock:
or the fact that [H] is actually writing some of the best video card reviews out there?

that one
yep, only thing they lack are some nice synthetic benchies.

[Brent]

oh noes :shock:
or the fact that [H] is actually writing some of the best video card reviews out there?

that one
yep, only thing they lack are some nice synthetic benchies.

i just read that as: "yep, only thing they lack are some nice synthetic boobies."

i know, i'm weird :(

[/end threadjack]

[Hellbinder]

PS2.x allows dynamic branching (it's not required, it's a cap), gradient, arbitrary swizzle, etc but they are optional. It's not clear to me whether ATI will support these are not. The major requirement for 3.0 is 512 instructions minimum, and the removal of texture read limits, plus increased number of registers.

It seems like according to the spec, it is possible to omit dynamic branching, yet still claim shader model 3.0. (although it was be a hollow claim IMHO)

For V3.0, the major change of course, is vertex textures.
Bingo!

R420 = PS/VS 3.0

Although I find it funny that you find it "hollow" that Ati may not support an "optional" function. I am wondering why that might be?? The minimum Spec is the Minimum Spec. It was the same rules for PS2.0 and its the Same Rules for PS 3.0

Further what if you could claim nearly infinite Instructions far beyond 512? and actually be able to deliver it in a practical usable way into the thousands of instructions?? Why is it that certain people seem to think that Dynamic Branching is so important yet Other features like HDR, Multiple render targets, Floating point volumetric textures, Floating point Cube maps etc.. seem to be of no consiquence??

Interesting.

[Bjorn]

PS2.x allows dynamic branching (it's not required, it's a cap), gradient, arbitrary swizzle, etc but they are optional. It's not clear to me whether ATI will support these are not. The major requirement for 3.0 is 512 instructions minimum, and the removal of texture read limits, plus increased number of registers.

It seems like according to the spec, it is possible to omit dynamic branching, yet still claim shader model 3.0. (although it was be a hollow claim IMHO)

For V3.0, the major change of course, is vertex textures.
Bingo!

R420 = PS/VS 3.0

Although I find it funny that you find it "hollow" that Ati may not support an "optional" function. I am wondering why that might be?? The minimum Spec is the Minimum Spec. It was the same rules for PS2.0 and its the Same Rules for PS 3.0

Further what if you could claim nearly infinite Instructions far beyond 512? and actually be able to deliver it in a practical usable way into the thousands of instructions?? Why is it that certain people seem to think that Dynamic Branching is so important yet Other features like HDR, Multiple render targets, Floating point volumetric textures, Floating point Cube maps etc.. seem to be of no consiquence??

Interesting.


Honestly, it's not performance I think will be the NV40's problem, it's featureset. The NV30 simply lacks alot of DX9 features (e.g. MRT, lots of FP texture formats, etc), plus crappy AA. The NV40 could whip the R420's ass in shaders, but most of the people on this board wouldn't buy it if it had the same old AA.


http://www.beyond3d.com/forum/viewtopic.php?t=10364&start=0

[Hellbinder]

PS2.x allows dynamic branching (it's not required, it's a cap), gradient, arbitrary swizzle, etc but they are optional. It's not clear to me whether ATI will support these are not. The major requirement for 3.0 is 512 instructions minimum, and the removal of texture read limits, plus increased number of registers.

It seems like according to the spec, it is possible to omit dynamic branching, yet still claim shader model 3.0. (although it was be a hollow claim IMHO)

For V3.0, the major change of course, is vertex textures.
Bingo!

R420 = PS/VS 3.0

Although I find it funny that you find it "hollow" that Ati may not support an "optional" function. I am wondering why that might be?? The minimum Spec is the Minimum Spec. It was the same rules for PS2.0 and its the Same Rules for PS 3.0

Further what if you could claim nearly infinite Instructions far beyond 512? and actually be able to deliver it in a practical usable way into the thousands of instructions?? Why is it that certain people seem to think that Dynamic Branching is so important yet Other features like HDR, Multiple render targets, Floating point volumetric textures, Floating point Cube maps etc.. seem to be of no consiquence??

Interesting.


Honestly, it's not performance I think will be the NV40's problem, it's featureset. The NV30 simply lacks alot of DX9 features (e.g. MRT, lots of FP texture formats, etc), plus crappy AA. The NV40 could whip the R420's ass in shaders, but most of the people on this board wouldn't buy it if it had the same old AA.


http://www.beyond3d.com/forum/viewtopic.php?t=10364&start=0
Interesting but that again goes to point out that the areas that Nvidia are lacking (except aa) are generally swept under the carpet. Dynamic Branching and Swizzle seem to be held up as the Golden egg while the rest are unconsiquential unless really pushed to the point.

Just look above.. it will be "hollow" if ATi does not support Dynamic Branching. When Dynamic Branching is no more Required than Floating point Volumetric textures.

[Evildeus]

It's quite rare to see a not so much enthousiastic post on R420 by Hellbinder :shock:

[kemosabe]

Whatever shader specs R420 supports, Fuad claims that final silicon is set and boards are ready for production. :lol:

http://www.theinquirer.net/?article=14302

[Hellbinder]

Besides.. It does not have *exactly* The same old AA ;)

It does support lots of DX9 features and it will be much faster in FP32.

[Bjorn]

Interesting but that again goes to point out that the areas that Nvidia are lacking (except aa) are generally swept under the carpet. Dynamic Branching and Swizzle seem to be held up as the Golden egg while the rest are unconsiquential unless really pushed to the point.

Just look above.. it will be "hollow" if ATi does not support Dynamic Branching. When Dynamic Branching is no more Required than Floating point Volumetric textures.

I don't know. I think there has been a lot of complaints about missing features of the NV30, especially from developers. And as you can see, also from DemoCoder. I seem to remember that one of the devs that writes on this forum said that the biggest problem for him when it came to the NV30 was that it didn't support MRT. And let's of course not forget Half Life 2 :-)

And i can't remember reading anything about the huge problems with lack of features for the R300 series.

[DemoCoder]

The reason why not having dynamic branches sucks is because developers will have to check this CAPS bit and author *two* 3.0 shaders, one that supports dynamic branches, and one that doesn't. To some extent, this can be alleviated in GLSlang, because the compiler can compile the right version before runtime, but there are alot of algorithms that aren't amenable to inlining into predicates. For Microsoft's HLSL, you'll have to compile a bunch of different versions.

The problem with making gradient, branches, et al, optional in 3.0 is that it creates a plethora of target profiles. There aren't any "optional" assembly instructions in PS2.0 Now we have the situation that 3.0 is nearly identical to 2.0, except for instruction limit and a handful of registers. All other features are optional.

I'd rather that ATI ship a card that is PS2.x, and microsoft *require* all the features that are optional PS instructions in 3.0, than to have a situation where two cards can claim Shader Model 3.0, but have vastly different instruction set support.

It's better to have a consistent standard that requires minimal support than to declare everything optional. I came down hard on Nvidia for not having MRT, for example, because I think MRT should be mandatory in DX9. One of the reasons why I like OpenGL better is all the mandatory requirements, so that developers don't have to the headache of checking bazillons of caps, features, and formats. (yes, there are extensions, but it's no where near as fine-grained as the CAPS system)


Simply put, developers should have the peace of mind to write their shaders, and not have to probe a dozen feature flags and provide combinatorial versions for many scenarios.

The "3.0" moniker gives people the impression that it is a *major revision* over "2.0", but in fact, if you compare it to 2.x, it is a minor revision, because all of the features that were optional in 2.x are still optional in 3.0.

I personally feel that if you say you support 3.0, but offer no new branching, gradient, swizzle, or texture read limit differences, it might be legal, but I think its misleading and confusing. Slapping an F-Buffer on top of 2.0 ALUs to extend instruction slot limits and calling them 3.0 seems to be a hollow victory.


Now, if they support vertex textures, they certainly have the right to use the VS3.0 moniker, because that's the major change to VS3.0.

[pcchen]

The "3.0" moniker gives people the impression that it is a *major revision* over "2.0", but in fact, if you compare it to 2.x, it is a minor revision, because all of the features that were optional in 2.x are still optional in 3.0.

I don't have a D3D9 spec, but the document gives me the impression that all PS 2.X functions are "supported" in PS 3.0, except dynamic branch, which is optional (range from 0 to 24, where 0 means "not supported").

For example, arbitrary swizzle is described as in PS 2.X: "If D3DPS20CAPS_ARBITRARYSWIZZLE flag is set, arbitrary swizzle is supported." but in PS 3.0 it's simply "Arbitrary swizzle is supported." All other functions, including gradient instructions, predication, dependent read limit, and texture instruction limits, are also described this way.

[bloodbob]

Further what if you could claim nearly infinite Instructions far beyond 512? and actually be able to deliver it in a practical usable way into the thousands of instructions?? Why is it that certain people seem to think that Dynamic Branching is so important yet Other features like HDR, Multiple render targets, Floating point volumetric textures, Floating point Cube maps etc.. seem to be of no consiquence??

Interesting.

Correct me if I'm wrong but doesn't dynamic branching save a whole load of instruction when you try to emulate dynmaic branching with static branching. You make it clear how useless it is to support more then 512 instructions. I tell you I'm quite sick of this 56ish ARB FP1.0 instruction limit on my R300.

[Dave Baumann]

Whatever shader specs R420 supports, Fuad claims that final silicon is set and boards are ready for production. :lol:

Well, there was at least one paragraph that Fudo got spot on there.

[tEd]

The "3.0" moniker gives people the impression that it is a *major revision* over "2.0", but in fact, if you compare it to 2.x, it is a minor revision, because all of the features that were optional in 2.x are still optional in 3.0.
.

i really don't think that the features are optional in ps3.0. That wouldn't make any sense. It would destroy all the reasons why the ps3.0 spec. was even created.

[DemoCoder]

I don't have a D3D9 spec, but the document gives me the impression that all PS 2.X functions are "supported" in PS 3.0, except dynamic branch, which is optional (range from 0 to 24, where 0 means "not supported").



Actually, you're right, I was looking at an old version of the documentation before Microsoft released the update. Even dynamic branching seems required now,


http://msdn.microsoft.com/library/default.asp?url=/library/en-us/directx9_c/directx/graphics/reference/assemblylanguageshaders/pixelshaders/ps_3_0.asp
Dynamic Flow Control
The device supports dynamic flow control (if bool, break, and break_comp). The depth of nesting ranges from 0 to 24.


Vs the DXSDK docs I have in an old directory

Dynamic Flow Control
If D3DCAPS9.D3DPSHADERCAPS_2_0.DynamicFlowControlDept h is non-zero, the device supports dynamic flow control (if, break, and breakc). The depth of nesting ranges from 0 to 24.

Number of Temporary Registers
The number of temporary registers supported by the device. The range is from 12 to 32.

Static Flow Control Nesting Depth
D3DCAPS9.D3DPSHADERCAPS_2_0.StaticFlowControlDepth is equal to the depth of loop/call nesting. The range is 0 to 4.

Number of Instruction Slots
The number of instruction slots can range from 96 to 1024. The total number of instructions performed, which is always greater than this value because of looping and function calls, is clamped by D3DRS_MAXPIXELSHADERINST.

Arbitrary Swizzle
If D3DPS20CAPS_ARBITRARYSWIZZLE flag is set, arbitrary swizzle is supported. See Arbitrary_Swizzle.

Gradient Instructions
If D3DPS20CAPS_GRADIENTINSTRUCTIONS is set, gradient instructions are supported. See dsx and dsy.

Predication
If D3DPS20CAPS_PREDICATION flag is set, instruction predication is supported. See Predicate.

Dependent Read Limit
If D3DPS20CAPS_NODEPENDENTREADLIMIT is set, there are no dependent read limits.

Texture Instruction Limit
If D3DPS20CAPS_NOTEXINSTRUCTIONLIMIT flag is set, there are no limit on texture instructions.


Ah well, I'm happy. No optional features for 3.0 pixel shaders. Looks like MS made the right decision between the last preliminary documentation release and the final spec.

[Demirug]

The reason why not having dynamic branches The "3.0" moniker gives people the impression that it is a *major revision* over "2.0", but in fact, if you compare it to 2.x, it is a minor revision, because all of the features that were optional in 2.x are still optional in 3.0.

I personally feel that if you say you support 3.0, but offer no new branching, gradient, swizzle, or texture read limit differences, it might be legal, but I think its misleading and confusing. Slapping an F-Buffer on top of 2.0 ALUs to extend instruction slot limits and calling them 3.0 seems to be a hollow victory.

No, if you want to report PS 3.0 in the caps you have to support all features that were optional in 2.X and some more. There is no other way in the context of the DX shader version system. If the caps say PS 3.0 is supported every possible 2.X Shader have to run on this chips, too.

The SDK have a little documentation bug but in the DDK you can find the right setup.

[madshi]

So I'd guess the R420 will have the following shader specs:

PS2.0: full support
VS2.0: full support

PS3.0: limited support (all features except dynamic branching)
VS3.0: full support

What do you think?

[nggalai]

Ah well, we'll see what happens. I know there are still certain parties that remain utterly insistent R420 is PS3.0-compliant, despite Demirug's thread (The Inq for example).

Sorry but which thread do you mean? I am not the person who say that R420 don't have PS 3.0. The only thing I have say is that R420 supports long shaders (2.X) and i have asked (at 3dcenter) if this will the only update? The next day Leonidas writes that there is no PS 3.0 support.

Ho-ho-ho... :lol:

Nice story about 3dcenter's credibility... :lol:That thread was located in our "speculations" subforum and is not related to Leo's article.

93,
-Sascha.rb

[Demirug]

So I'd guess the R420 will have the following shader specs:

PS2.0: full support
VS2.0: full support

PS3.0: limited support (all features except dynamic branching)
VS3.0: full support

What do you think?

If you do not support all 3.0 Feature you have a 2.X Chip.

[madshi]

If you do not support all 3.0 Feature you have a 2.X Chip.
Hmmm... I think we have to seperate PS3.0 and VS3.0. A chip can support VS3.0 while not (fully) supporting PS3.0. If a game engine makes use of VS3.0 features but not of PS3.0 features, such a chip had no disadvantage for this specific game engine. Of course for developers it would be much nicer and simpler if they could just seperate 2.0 chips from 3.0 chips. But I think it should be no problem for a game engine to handle PS and VS seperately. After all those are in DX9 still seperate shaders.

[Demirug]

If you do not support all 3.0 Feature you have a 2.X Chip.
Hmmm... I think we have to seperate PS3.0 and VS3.0. A chip can support VS3.0 while not (fully) supporting PS3.0. If a game engine makes use of VS3.0 features but not of PS3.0 features, such a chip had no disadvantage for this specific game engine. Of course for developers it would be much nicer and simpler if they could just seperate 2.0 chips from 3.0 chips. But I think it should be no problem for a game engine to handle PS and VS seperately. After all those are in DX9 still seperate shaders.

Missunderstanding. I was only talk about your "PS3.0: limited support (all features except dynamic branching) " but it is same same with VS. nV3X miss VS 3.0 because it supports no Texturereads.

[madshi]

Missunderstanding. I was only talk about your "PS3.0: limited support (all features except dynamic branching) " but it is same same with VS. nV3X miss VS 3.0 because it supports no Texturereads.
Oh, yes, then I agree with you.

However, we don't know yet how fast NV40's dynamic branching will be. There were some hints that NVidia might have found a way to implement it cheaply but slowly. In that case it might happen that developers won't use dynamic branching in DX9 cards at all. And that might result in that R420 will have no disadvantage technically, but only in terms of marketing.

Maybe I'm now sounding as if I would desperately trying to defend the R420, no matter what. But that's not my intention. I'm just thinking about possible future scenarios. (I'm saying that just in case someone wants to stick an ATI fan label on me :) ).

[THe_KELRaTH]

I would prefer:-
A new form of AA that handles Alpha Tests so that no matter what - the whole image is smooth.
Full AF (as in not angle based) which is also extended further
Increase in speed so that having Vsync on doesn't cause stuttering when fps drops below Hz setting thereby offering gameplay as smooth as watching a movie on TV (or like older computers such as Atari 8bits etc).

[christoph]

another quote from leonidas from 3dc regarding the source of their assertion from the same thread i linked above:

Auf keinen Fall ist dies unsere Quelle. Wir haben die Sache uns nicht irgendwie aus ein paar Halb-Infos erdacht,
sondern wir haben eine definitive und glasklare Aussage von jemanden, der es gemäß Amt wissen *muß* - und nebenbei mit
dieser Info von ATi höchstselbst ganz offiziell gefüttert wurde. Alles koscher also.


Aber zur Quelle selber: Sorry, kein Kommentar.

translation by me (english is not my first language):
in no way this is our source (demirurgs speculation thread). we didnt jump to conclusions based on some questionable informations,
but we have a definitive, crystal-clear statement from somebody who *has* to know it because of his position - and btw got
this information official and directly from ati itself.so nothing fishy here.

about the source itself: sorry, no comment

[Ante P]

Whatever shader specs R420 supports, Fuad claims that final silicon is set and boards are ready for production. :lol:

Well, there was at least one paragraph that Fudo got spot on there.

I see two. :)

[PatrickL]

Anyway, with all the ATI guys lurking here, i imagine we should have get a post saying that 3d center article was wrong by now, and without breaking any rule no ?

[Proforma]

I mean why even be DX 9 compliant?

People say that PS3.0 doesn't matter because there is no software out for it for a year, but it does matter because some of the average user is going to be using this card in a few years and the programs that I write won't work on his card and that affects my sales and that also means that you might as well just give up DX and OpenGL and go console only.

If this is true then ATI should not even bother putting a card out.
I own a Radeon 9800 Pro and I bought it because of Nvidia's stupid redneck decisions, but ATI looks like they are going the same stupid way.

What is up with these two companies? Are they out to have a competition to see which one can be more stupid?

No disrespect ment to this forum or its members. I am just angry to being excited about purchasing a next generation product and then having a company see what they can do to do stupid things to screw it up.

I upgraded my Geforce 3 (original) to a Radeon 9800 Pro and I have now been advocates for both companies and I have been fans of both companies, but they keep shooting themselves in the foot.

I was going to buy an R420 and be done with this system and sell it next year and go PCI Express, but now I am not so sure of that. I might actually buy an Nvidia product again if this rumor is true.

Microsoft should really get a logo out and say 100 Percent Fully Direct X 9 compliant and award it to companies that have their act together.

However, I know why they won't because nobody is 100 percent Fully Direct X 9 compliant, so you might as well kiss DX 9 goodbye.

160 million transistors at 13 micron with low k,what are they doing with this many transistors if they don't have 3.0 compliance? Thats what I want to know.

[anaqer]

160 million transistors at 13 micron with low k,what are they doing with this many transistors if they don't have 3.0 compliance? Thats what I want to know.

Well, there are the two extra VS units... I'm also speculating on two TMU-s per pipe, but that's just me.

[vb]

Can't ATI "multipass" branching at driver level?

Sure, it wouldn't be the fastest way, but maybe they can afford to...

[DemoCoder]

Let's say you have a loop in your pixel shader anda conditional IF statements that BREAK out of the loop. How do you propose to multipass this semiefficiently?

[Ante P]

160 million transistors at 13 micron with low k,what are they doing with this many transistors if they don't have 3.0 compliance? Thats what I want to know.

Well, there are the two extra VS units... I'm also speculating on two TMU-s per pipe, but that's just me.

Also keep in mind the comments about more precision ATi (think it was Dave orton) made a while back...

[g__day]

Can someone give me a breakdown of how +50 Million transistors were allocated.

Like 5-6 Million for a new Vertex shader unit * 2, then 9-11 Million for a new pixel shader (fp24) * ??? + Z million for n extra TMUs per 8 pipelines
etc.

How does the math best breakdown guys and gals please?

[Magic-Sim]

And 1 transistor targetted to the rendering of PS 3.0, just to have the bullet point feature on the back of the box, rendering it at an asstonishing speed of 0.00000001fps.

Huge PS 2.0 speed and kind of PS 3.0 support 8)

nVIDIA on the contrary chose 0.1fps on both cases :idea:

[anaqer]

nVIDIA on the contrary chose 0.1fps on both cases :idea:

Isn't this getting a bit old already? :roll:

[Chalnoth]

Let's say you have a loop in your pixel shader anda conditional IF statements that BREAK out of the loop. How do you propose to multipass this semiefficiently?
Statically execute at least two loops at a time (depending on the length of the loop), with a compare that, essentially, chooses between doing another pass and outputting the answer of one of the loops done this pass.

[digitalwanderer]

Besides.. It does not have *exactly* The same old AA ;)

It does support lots of DX9 features and it will be much faster in FP32.
"and it will be much faster in FP32", are we talking the R420 or nV40 here?

Whatever shader specs R420 supports, Fuad claims that final silicon is set and boards are ready for production. :lol:

Well, there was at least one paragraph that Fudo got spot on there.

I see two. :)
Which was the other one?

[DemoCoder]

Let's say you have a loop in your pixel shader anda conditional IF statements that BREAK out of the loop. How do you propose to multipass this semiefficiently?
Statically execute at least two loops at a time (depending on the length of the loop), with a compare that, essentially, chooses between doing another pass and outputting the answer of one of the loops done this pass.

But this will vary per-pixel. One pixel might need 1 pass. Another might need 20. The break instruction will be executed dynamically. That means you have to re-render the whole scene, but mask out pixels which don't need any more iterations. Now consider doing this for nesting levels up to the allowed 4 minimum. And how do you efficiently determine when you need no more loops/passes? Read back the framebuffer to figure out you're done, or just always execute the maximum number of passes given by the constant loop register?

This is not what I call "semiefficient". It'll work, like stanford's RTSL worked, but it'll be non-realtime IMHO.

[DemoCoder]

nVIDIA on the contrary chose 0.1fps on both cases :idea:

Isn't this getting a bit old already? :roll:

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.

[digitalwanderer]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:

[Tim Murray]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:
Something like this? (http://www.penny-arcade.com/images/2002/20021213l.gif)

[Ante P]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:

Why not use the Redeemer?

[cthellis42]

There'll still be ashes left to throw at people.

[Ante P]

Which was the other one?

Depends on which one we are referring to as the first one now doesn't it? ;)

[Chalnoth]

But this will vary per-pixel. One pixel might need 1 pass. Another might need 20. The break instruction will be executed dynamically.
Ah, yeah, didn't think about that. But the obvious solution would be to use an "uber buffer" output buffer, where, say, 8 bits are dedicated to one or more flags, with a flag being whether or not to execute said loop. Oh, and remember that for most branching instructions, there should be whole regions where essentially the same shader is executed, which I imagine should help with efficiency.

Anyway, the situation you outlined, with a conditional-length loop, is quite possibly a worst-case scenario for graphics hardware, so it's never going to be all that efficient.

[jvd]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:

Why not use the Redeemer? we still need the sharks with laser beams on it to blow up the horse

[digitalwanderer]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:

Why not use the Redeemer?
Too detached, you don't get the visceral satisfaction the same way you do when you manually beat it a bit by hand.



Which was the other one?

Depends on which one we are referring to as the first one now doesn't it? ;)
Ok, how about "which is the one I don't know which is right?" then?

[Tim Murray]

Ok, how about "which is the one I don't know which is right?" then?
Well, that depends on which one you do know is right... +2 BASTARD POINTS! ;)

At least it seems like people have stopped being so worked up about this and have resigned to wait and see how much it matters. I don't think that poor PS2.0 support killed NV3x sales, so I don't know how much PS3.0 will matter.

[digitalwanderer]

At least it seems like people have stopped being so worked up about this and have resigned to wait and see how much it matters. I don't think that poor PS2.0 support killed NV3x sales, so I don't know how much PS3.0 will matter.
Well, we really don't know nothing yet officially and this is still all speculation...and a whole lot is going to depend on how ATi bills their part, how quickly anything that demonstrates it's weakness comes out, and it's just too early to get all worked up over what may or may not be.

But it IS fun to speculate, and I'll be damned if I don't actually understand all of this and it's ramifications! :shock:

[martrox]

But it IS fun to speculate, and I'll be damned if I don't actually understand all of this and it's ramifications! :shock:

And, just look at all the returning people we haven't seen much of in about, oh, a year and a half! :wink:
Not to mention all the flagwaving some here are doing at the mere thought that this is true! :wink: :wink:

[digitalwanderer]

Not to mention all the flagwaving some here are doing at the mere thought that this is true! :wink: :wink:
I ain't waving no flags, I'm just checking out the crowd. ;)

Gotta agree though, a great thread all around. :)

[Magic-Sim]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:

Humans love doing that. Especially with arrogants :D One year of nVIDIA is a must read ;)

[PaulS]

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.
You forgot the bit where I jump up and down on the dead horse screaming and waving me arms about and then the bit where I whack it a few times with the shovel again for luck and set it aflame. :roll:
Something like this? (http://www.penny-arcade.com/images/2002/20021213l.gif)

Goddamn it, I was going to post that. One of my favourites.

[Joe DeFuria]

One question about "Vertex Texturing."

Is there any significant detriment if the R420 supports VS 3.0, while at the same time does not support FP32 throughout the entire PS pipeline?

I know the R300 (and thus R420) supports FP32 textures, so physically it wouldn't be an issue to have "Vertex Textures" combined with their current FP24 floating point pixel pipelines.

But is there a need to have a scenario like this:

1) Pixel shader does some operations to create a texture / render target.
2) Vertex shaders operate on that target.

So could any reduced precision of FP24 PS operations have a significant usability impact on using the resulting render targets for vertex shader input?

[anaqer]

Goddamn it, I was going to post that. One of my favourites.

Yup, PA is somewhat like The Simpsons to me - anyone creatively quoting it instantly wins the argument.

[Ante P]

So could any reduced precision of FP24 PS operations have a significant usability impact on using the resulting render targets for vertex shader input?

Since it's a FP32 (in PS) capable part why bother thinking about it? ;)

[Joe DeFuria]

Since it's a FP32 (in PS) capable part why bother thinking about it? ;)

Because in all seriousness, I don't know if R420 will be FP32 or not. There are definitely cost pros for sticking with FP24, and I'd be interested in any "real architectural cons" for sticking with FP24 as well.

[cthellis42]

Considering the expected transistor bumps (and a few other comments), it seems rather like ATi is making the FP32 move with R420. Unless there's a few random, totally unexpected changes in there as well.

[hoom]

Like hardware accelerated subdivision surface primitives? (aka trueform3?)
I definitely recall reading something from an ATI guy about that being needed...

[T2k]

nVIDIA on the contrary chose 0.1fps on both cases :idea:

Isn't this getting a bit old already? :roll:

Beat. Horse. Dead. Dig up horse. Beat again. Stand horse up. Kick and beat down. Again.

They got what they deserve - actually, it is still not enough.

No, definitely not yet.

[Joe DeFuria]

Considering the expected transistor bumps (and a few other comments), it seems rather like ATi is making the FP32 move with R420. Unless there's a few random, totally unexpected changes in there as well.

Well, transistor bumps don't mean much to me...I mean, there's supposed to be a very advanced vertex shader, and the "rumors" point to a doubling a PS "performance", clock for clock. They could account for any transistor increase.

Since FP32 is not needed, for PS 3.0 (let alone 2.0), it almost seems like it would be a waste of silicon for ATI to support it. It would be mostly a marketing check box. If, on the other hand, not having FP32 would really cause issues with VS 3.0, then I could see it. This is why I asked the Vertex shader question. (Is anyone going to offer any insight? ;))

I suppose going from FP24 to FP32 may not require that much increase in transistors....but if that were the case then you would think ATI might have just gone with FP32 in R300.

Back when we first leared that R390/420 was "heavily based on R300", I was assuming that R420 would be PS 2.0 and FP24. Then, we started hearing "definitive rumors" that R420 was PS 3.0...I went along with that until this past week when the current "definitive rumor" is PS 2.x. So, I'm wondering if my original hunch was correct all along on both accounts.

[T2k]

nVIDIA on the contrary chose 0.1fps on both cases :idea:

Isn't this getting a bit old already? :roll:

Still much younger than that CRAP we've been told by NVIDIA for a YEAR, my friend.

[Pete]

I don't see why ATI would bother with FP32 in the pixelshaders if R420's not going to support PS3.0.

This might be another case of ATi ratining transistors wisely, but we'll see. We've all expected ATi to blow nV out of the water in DX9 games, but that hasn't really happened yet because, well, there ain't no DX9 games. :?

[Lecram25]

I don't see why ATI would bother with FP32 in the pixelshaders if R420's not going to support PS3.0.

This might be another case of ATi ratining transistors wisely, but we'll see. We've all expected ATi to blow nV out of the water in DX9 games, but that hasn't really happened yet because, well, there ain't no DX9 games. :?

And Tomb Raider: AoD is what?

[digitalwanderer]

And Tomb Raider: AoD is what?
A half-way decent benchmark before they pulled it with the last patch, but the game itself was so sucky as to be nearly unplayable.

How about we say "there haven't been any good dx9 games yet that really take advantage of a lot of dx9 features", could we live in peace with that one and call it a draw? :|

[Geo]

Even assuming ATI can kick butt in current games, I would be concerned for them a year down the line. Remember how we use to hear at great length how many incompatibilities and "bugs" in ATI drivers really weren't, they were just developers who had developed around NV's bugs and now ATI was paying the price even tho their implentation was correct? Remember how that disappeared with DX9? Well, it will be back a little down the road if they hand the field to NV on ps3.0.

[Brent]

FarCry

[Chalnoth]

Of course, there have been some documented instances of ATI's drivers not conforming to spec, too.

But yes, the "underdog" will always have issues, even if the only problem is that the software is optimized for the competitor's hardware, as was the biggest problem for nVidia, back when they were overcoming 3dfx for mindshare. More recently this has likely also hurt the performance of the NV3x line somewhat (I certainly don't claim it's all of the performance hit, or even most, but witness such things as 3DMark2k3 not benchmarking in any partial precision mode). For ATI this was probably more of an issue with the Radeon 8500.

[DemoCoder]

Let's just say that I've been pleasantly underwhelmed by Far Cry. Nice polybump and shadows, but HL2's smoke and mirrors E3 demos still look better. Hopefully, once Valve finishes writing the real game, it will have the same quality. :)

[Hanners]

Which was the other one?

Depends on which one we are referring to as the first one now doesn't it? ;)

Well, we know Fuad got his name right, so that's one thing... ;)

[Ante P]

So if ATi supports FP32 what will happen to FP24 and does it mean they'll also support FP16?

And what if they have both FP24 and FP32, how would an application choose between the two. (Seeing that both are full precision and no hint is required.)

[Miksu]

Kinda interesting comment from MuFu @ Rage3D:

I don't think NV40 is launching at CeBIT now either.

So (if this is true), if Ati was supposed to wait until NV40 is launched before releasing R420, does this mean that we're gonna wait a little longer for both chips?

[Skinner]

Maybe they wait on each other, so no one is coming out :D

I wish it was already april, I want to buy some new toys.

[anaqer]

nVIDIA on the contrary chose 0.1fps on both cases :idea:

Isn't this getting a bit old already? :roll:

Still much younger than that CRAP we've been told by NVIDIA for a YEAR, my friend.

Well, whatever floats your boat, my friend.

[Hanners]

Maybe they wait on each other, so no one is coming out :D

I wish it was already april, I want to buy some new toys.

My pay rise kicks in at the start of April, it couldn't be timed much better. :D

[incurable]

Well, we know Fuad got his name right, so that's one thing... ;)
:shock: Wait, has it really been verified by independent sources that he exists and does indeed turn his head when being called by his name? :D :wink:

Edit: Oh, and on the timing issue, I don't care about the announcements, availability in early May would be fine with me. :)

[PaulS]

Kinda interesting comment from MuFu @ Rage3D:

I don't think NV40 is launching at CeBIT now either.

So (if this is true), if Ati was supposed to wait until NV40 is launched before releasing R420, does this mean that we're gonna wait a little longer for both chips?

I'd hold off on reading too much into that quote for now, since there's some confusion there (or there was last night, at least).

[Sage]

if the rumors about R420 being available in quantity already are true then I can't imagine ATi wasting the jump they have on nVidia. Last time round they were the only ones with DX9 cards on the market for a loooong time and sold quite a many few because of that sole factor. It looks like they could have a product shiped a few months before nVidia, giving them a very nice window where they, once again, have no competition. Why throw away all of those sales because your competitor is being slow?

[digitalwanderer]

if the rumors about R420 being available in quantity already are true then I can't imagine ATi wasting the jump they have on nVidia. ......Why throw away all of those sales because your competitor is being slow?
Because there are an awful lot of R3xx cards still in the channels to be sold and the card makers would probably appreciate a chance to unload their stock on hand of old cards before the new cards come out and make 'em obsolete. :)

[Sage]

Because there are an awful lot of R3xx cards still in the channels to be sold and the card makers would probably appreciate a chance to unload their stock on hand of old cards before the new cards come out and make 'em obsolete. :)
yeah I know, but I think that ATi is in a possition to strike nVidia another hard blow. nVidia is already in a weak possition and allowing them to recouperate would be a big mistake because I doubt they will let themselves be put in such a possition again.

[CapsLock]

if the rumors about R420 being available in quantity already are true then I can't imagine ATi wasting the jump they have on nVidia. ......Why throw away all of those sales because your competitor is being slow?
Because there are an awful lot of R3xx cards still in the channels to be sold and the card makers would probably appreciate a chance to unload their stock on hand of old cards before the new cards come out and make 'em obsolete. :)

It seems as if the R350 is already being priced at mid-range levels in some places @~200USD, the R360 could be brought down to the 300USD level and they would still be making money on it, leaving plenty of room for a 400-500USD uber card that enthusiasts will be unable to resist. I agree with the get a jump on the competition POV. Many systems and buyers work on the whatever is the mostest factor, if Ati get this market they will have the inertia when NV40 hits.

Not that I know all the facts on this kind of situation, just my 2 cents.

Caps

[Joe DeFuria]

yeah I know, but I think that ATi is in a possition to strike nVidia another hard blow. nVidia is already in a weak possition and allowing them to recouperate would be a big mistake because I doubt they will let themselves be put in such a possition again.

Yes, but you don't necessarily "strike nVidia" from a business perspective just by launching your cards first. You strike nvidia by making your OEM customers as satisfied with a combination of products and service.

The "happier" ATI keeps their oem customers, the more these customers would be willing to either

1) Be an ATI shop only
2) Be willing to pay a premium for ATI's parts.

That being said, there is a certain market appeal to launching first, (it can help your brand, as well as allow OEMs to charge a premium to consumers for these parts before the competition hits the market.) So there are of course some cons to "waiting." Picking the right time to "launch" is as much of an art as it is a science.

[CapsLock]

Checking PriceGrabber just now, it seems most 9800P are still close to 300USD, and XT's are around 450USD. Soooo, maybe my analysis wasn't so goodish. :?

Caps

[digitalwanderer]

Yes, but you don't necessarily "strike nVidia" from a business perspective just by launching your cards first. You strike nvidia by making your OEM customers as satisfied with a combination of products and service.

The "happier" ATI keeps their oem customers, the more these customers would be willing to either

1) Be an ATI shop only
2) Be willing to pay a premium for ATI's parts.

That being said, there is a certain market appeal to launching first, (it can help your brand, as well as allow OEMs to charge a premium to consumers for these parts before the competition hits the market.) So there are of course some cons to "waiting." Picking the right time to "launch" is as much of an art as it is a science.
Bang on!

I don't think ATi's main goal is just to beat nVidia, I think their goal is to win customers. ;)