The Inquirer Trying to Scoop B3D!

[demalion]

How the heck is mentioning Hellbinder's "4x4" comment anything remotely resembling "anti-nvidia bias"? It seems likely now that Hellbinder's description may have been based on the description Randell was replying to...was he asking anything more than that?

It's the mentality.

That one line question is an exhibit of a "mentality"? What does your response exhibit? Should we stop here and ponder the "mentalities" exhibited by the question and your response to it? I did ask you if there was other text you had in mind, so I presume this means "yes"?

Err..what? Where did that come from? The observed information has the GF FX exhibiting performance equivalent to a 4 pipe architecture in many cases. Why is he a "moron" for asking the question you quoted? Or, is there some other piece of text you forgot to quote that is the reason for your description?

Are you missing this whole ideal? Instead of a 4 pipe architecture, the architecture is instead composed of an array of processing elements, ALUs, whatever. From there, these can be formed into virtual pipelines - based upon whats being rendered.

Hmm...I think you are missing something that has been said a few times now: if it can't output the pixels to the screen, what is the point of all those "elements"? It is not like the GF FX is significantly outperforming the R300 in shader computations (this is WITH a significant clock advantage).

Thus, the whole paradigm of a 4 or 8 pipe architecture allways exhibiting the same, set-piece preformance is overwith. As Kirk stated in the quote that you all glazed over in your continued nVidia bashing.

Eh!? That reads like you started with David Kirk's comments, and are defining the world by it, with any and all observed performance to date not mattering at all. What exactly is "set-piece" performance, perhaps the meaning of this term is the cause of my confusion.

What we've actually observed is the nv30 fails to calculate pixel output significantly more quickly in comparison to the competition's design, is limited in its ability to output what it has calculated to the screen compared to the competition's design, has more design complexity than the competition, and this comparison with the competition is with the advantage of significantly higher clock speed.
Looking at your statements so far, it seems to me that you are the one who "glazed over" important issues.

Is this statement supposed to make sense? "if you're still dumb enough to think,"? Eh? The simple observation is the R300 exhibits behavior consistent with being able to output 8 pixels per clock, and the nv30 does not (most of the time)...the "why" is under investigation, and it doesn't take "anti-nvidia" bias to wonder about it in the meantime.

It takes anti-nVidia bias to dismiss the answer:

Pipes don't mean as much as they used to. In the [dual-pipeline] TNT2 days you used to be able to do two pixels in one clock if they were single textured, or one dual-textured pixel per pipe in every two clocks, it could operate in either of those two modes. We've now taken that to an extreme. Some things happen at sixteen pixels per clock. Some things happen at eight. Some things happen at four, and a lot of things happen in a bunch of clock cycles four pixels at a time. For instance, if you're doing sixteen textures, it's four pixels per clock, but it takes more than one clock. There are really 32 functional units that can do things in various multiples. We don't have the ability in NV30 to actually draw more than eight pixels per cycle. It's going to be a less meaningful question as we move forward...[GeForceFX] isn't really a texture lookup and blending pipeline with stages and maybe loop back anymore. It's a processor, and texture lookups are decoupled from this hard-wired pipe.

What does it take to dismiss the observation that many elements of this description could apply to the R300 as well, and that it seems to be more effective at getting the tasks done (with less transistors, and lower clockspeed... )?

And instead talk about nVidia's lying and failure, ohh, and throw in some 3dfx bashing too.

So...your problem is that people are criticizing nvidia for calling this design "8x1" when it, so far, can't seem to actually output 8 pixels a clock in any useful circumstance? Why shouldn't they? Nvidia could have picked some other description that isn't misleading compared to its abilities, couldn't they?

Did you ever pause to contemplate that perhaps the reason why the nV30's preformance isn't fixed like the R300's, is because it is using the approach that Kirk talked about?

I have little doubt it is using the approach Kirk is talking about, to some unspecified degree. I am also observing that so far it gets equivalent work done at 500 MHz as the R300 at 325 MHz, and using more transistors. Is it really so hard to understand why that might be criticized?
Did you ever pause to contemplate that the reason why the nv30 is being criticized is because compared to the R300, and the statements nvidia made concerning its design, it deserves it?

If you have an array of processing elements, then you can only use their flexibility to do so much as you only have so many elements to share.

Well, that seems to be a flaw in the execution of the concept as done in the nv30, doesn't it? The advantage we've perhaps observed so far, to my knowledge, is enhanced simple vertex processing power. Do you want to search for my comments on the Quadro FX to see if I acknowledge that...would that satisfy your idea of how the nv30 design should be discussed?

Thus, it won't have the consistentcy of a fixed pipeline, but a virtual pipeline has more plasticity and - as stated - can achieve 16-odd ops in some tasks.

You keep echoing the words in David Kirk's comments (in a way that looks mangled to me) without explaining how discussing the way the design behaves like a 4x? architecture (not just in synthetic tests, either) is displaying "anti-nvidia bias".

You mean in 1997, when performance mattered...as opposed to now, when...?
Are you saying the nv30's performance isn't relevant now, or do you think there is some other reason this 4x? discussion is being brought up?

What? Preformance matters, but so does architectural elegence and effeciency. I think you have absolutly no idea what I'm talking about.

I'm all for architectural elegance and efficiency, believe me, but could you relate that description to the actual nv30 for me please? You do realize that it isn't David Kirk's description of an architecture we are "attacking" when we comment on 4x? for the nv30, but the specific realization (to whatever extent) of that architecture that the is the nv30?

I do fear, mind you, that this topic is sticking and growing in presence because of two factors (1) Ignorance to the Nv30's true design

What does "true design" mean? Are there "true pixels" being rendered by the nv30 that the benchmarks and games aren't accounting for? You seem to be using catch phrases to counter "criticism" based on observed performance. I don't understand the validity. Why don't you stop calling people ignorant for basing their comments on observations you decide to ignore?

(b) the rampent anti-nVidia rhetoric thats rallying around this ideology that the Nv30 has only "4pipes" and is thus inferior in the nomenclature race with the R300s "8pipes"

What are you talking about, "nomenclature race"?! Are you reading anything anyone has said about this? This 4x? speculation (seemingly confirmed now) is based on the actual performance of the nv30, not some nomenclature contest!

nvidia said nv30 is 8x1.
nv30 underperforms expectations, and the performance is evaluated.
nv30 is theorized to actually be more accurately described as 4xSomething.
You come in and criticize people for describing it as 4xSomething.

Ack!

 

[KimB]

Hmm...I think you are missing something that has been said a few times now: if it can't output the pixels to the screen, what is the point of all those "elements"? It is not like the GF FX is significantly outperforming the R300 in shader computations (this is WITH a significant clock advantage).

It matters for two things:
1. Odd numbers of textures. If only single-texturing suffers, then it's not much of a problem (since it's pretty much bandwidth-limited here anyway).
2. Fragment programs. The fragment processing power is independent of the number of pixels that can be output per clock. I would really like to see some benchmarks with relatively long programs that look at 8-bit int, 16-bit fp, and 32-bit fp processing power.

 

[Dave H]

As a side note, has anybody checked texturing speed with an odd number of textures? It seems possible that the NV30 would be capable of, say, 3-texture performance at the speed of an 8x1 pipeline (depending on how flexible the pipelines are).

If this is true, then the NV30 working as a 4-pipeline architecture when single-texturing is used will be of little consequence, as it will usually be bandwidth-limited in that situation anyway.

This is indeed the critical question, something I failed to appreciate before in two ways. (Thanks, Chalnoth. )

1) If NV30 turns out to handle e.g. 3-texture situations like a normal 4x2 pipeline, than this would indeed hinder performance in certain real-world situations, compared to a hypothetical 8x1 NV30.

2) If, on the other hand, NV30 manages to achieve the same throughput as an 8x1 would, then the pipeline design does reasonably deserve to be called especially flexible, for it would essentially be a true 8x1 with a decoupled "write to framebuffer" unit limited to a throughput of 4 per clock.

As I think about it, I'm beginning to think NV30 can pass this test, because this is simply the most sensible design for it to have. OTOH, one doesn't often go wrong by underestimating the NV30...

Very interesting! Villagers demand a benchmark!!

 

[Crusher]

Hmm...I think you are missing something that has been said a few times now: if it can't output the pixels to the screen, what is the point of all those "elements"? It is not like the GF FX is significantly outperforming the R300 in shader computations (this is WITH a significant clock advantage).

It can output pixels to the screen. I don't recall the screenshots being completely black.

So...your problem is that people are criticizing nvidia for calling this design "8x1" when it, so far, can't seem to actually output 8 pixels a clock in any useful circumstance? Why shouldn't they? Nvidia could have picked some other description that isn't misleading compared to its abilities, couldn't they?

Since when is something not useful simply because it's not the final value in the frame buffer?

I have little doubt it is using the approach Kirk is talking about, to some unspecified degree. I am also observing that so far it gets equivalent work done at 500 MHz as the R300 at 325 MHz, and using more transistors. Is it really so hard to understand why that might be criticized?

Criticism is fine, as long as you're criticizing things that are actually wrong. For instance, complaining that it's a year late, the drivers aren't as efficient in shading ops as they should be, and it's loud, are all valid criticisms. Saying it's crap because it's barely better than the 9700 Pro isn't what I would consider a valid criticism. If its operation does have more in common with a 4x2 style architecture than an 8x1 style architecture, then there are probably going to be several instances when the FX should be producing less pixels per clock than the 9700 Pro. In fact, one of these instances would be single texturing. And I don't think I have to point you to the article on this website that describes how commonplace single texturing is in games Indeed, the fact is that the FX outperforms the 9700 Pro in most of those games, wherein there are several areas that it should be doing worse (i.e. it's doing 4 pixels/clock at 500 MHz, and the 9700 Pro is doing 8 pixels/clock at 325 Hz, equivalent to 4 pixels/clock at 650 Hz). In that respect, looking at actual performance in situations where it should be severly handicapped, it seems to be performing quite well.

nv30 underperforms expectations, and the performance is evaluated.

The only two reasons it underperforms expectations, is that people had the erroneous idea that being delayed would make it significantly faster, and that it was supposed to stomp all over the R300 no matter what. I find it humorous that this will probably be the best performing "flop" the world has ever seen. Not to detract from the accomplishment of the R300, which certainly is a fine product, but a noticable percentage of people around here would fly off their rockers at the smallest bit of criticism anyone gave it, and the same crowd seems to be unwilling to give the NV30 any credit for going through delays and problems and still end up both on the market and in reasonably decent shape compared to the competition. Instead, people have been doing nothing but criticizing every aspect of it, even making unfounded claims like the one you made eariler about pixel output being meaningless unless it's the final pixel.

I know people think I'm an nvidiot, and dismiss most of my ramblings as someone who's blind, but the only reason it seems that way is that most of my posts are countering what I consider to be unfair, one sided criticism. Of course, the gate swings in both directions, but the net flow seems to be coming from the red and white side.

 

[RoOoBo]

So, how do you know the fragment/pixel back-end of the nV30 isn't based around 32 processing elements that can form upto 8 virtual pipelines, with the number actual dependent upon the task/fragment being worked on. IMHO, if the architecture is anything like this, regardless if it's generally utputting in a 4 pixel comparable - it's a hell of alot smarter of a design than anything else this side of the P10.

Complexity. Don't design something too hard to control when a simple design is as efficient or more efficient than the complex solution. Sorry but until someone explains why and how a 'sea of units' is implemented and efficiently used I will think it is just PR.

Complexity always has a cost.

BTW with a 4 pixel pipes it would made sense something like 2 16bit FP shader instruction per clock and pipe and 1 32 bit FP shader instruction per clock per pipe. That could explain what they say about the 8 shader instruction per clock.

 

[Althornin]

The only two reasons it underperforms expectations, is that people had the erroneous idea that being delayed would make it significantly faster, and that it was supposed to stomp all over the R300 no matter what. I find it humorous that this will probably be the best performing "flop" the world has ever seen.

People were GIVEN the idea it would stomp over the R300.
All of the critcism (or the vast majority of it) is just backlash to the overly positive propaganda and marketing. Products "flop" when public perception of them fails. This happens when a device doesnt live up to expectations. Expectations that were created by nVidia.

It wont be the last time a great product was trashed because it failed to live up to the hype, and it certainly isnt the first. I dont get all the "confusion" you exibit over this backlash.

Hype up the product, and you ahve two possibilities:
1) It lives up to the hype, and is likely very successful.
2) It doesnt live up to the hype, and no matter if it is competative, it has a chance of failing because of public opinion and backlash to said hype.

2) is what happened here.

 

[Crusher]

People were GIVEN the idea it would stomp over the R300.
All of the critcism (or the vast majority of it) is just backlash to the overly positive propaganda and marketing. Products "flop" when public perception of them fails. This happens when a device doesnt live up to expectations. Expectations that were created by nVidia.

It wont be the last time a great product was trashed because it failed to live up to the hype, and it certainly isnt the first. I dont get all the "confusion" you exibit over this backlash.

Hype up the product, and you ahve two possibilities:
1) It lives up to the hype, and is likely very successful.
2) It doesnt live up to the hype, and no matter if it is competative, it has a chance of failing because of public opinion and backlash to said hype.

2) is what happened here.

Oh please, as if NVIDIA is the only company with a marketing department. The Parhelia was a *MUCH* bigger disappointment, and it was hyped as being the exact same kind of huge leap in graphics processing ability, etc. etc. The only difference is people didn't give Matrox's marketing as much attention as they do to NVIDIA's. Because of that, when the Parhelia came out, the general tone was like "oh well, good try at least, better luck next time." When NVIDIA does it, and comes out with a much more competitive product in the end, the public response is more like: "I based my life around every single marketing phrase you thew at me and blindly believed it all, and this is the crap you fling in my face?!?!"

All the people who say they never listen to marketing and just look at the product are now lynching NVIDIA over the very things they claim they pay no attention to. Talk about hypocracy.

 

[THe_KELRaTH]

I agree that with Perhelia was a dissapointment and as you say "better luck next time" was the general feeling but in the case of the NV30 it's been close on a year of PR - now thats alot of PR!
(and even back then the NV30 was quoted as having 8 rendering pipelines).

On another note would I be correct in saying that the use of multi textures will be less important as more complex PS shaders are used?

 

[Hellbinder]

The hells wrong with you people. The amount of Anti-nVidia bias in here nowadays is sickening.


Like this comment above, are you a moron? Why do you have such a closed mind about this? Kirk just stated, if you're still dumb enough to think, that the era of fixed function pipelines with TCUs bolted on to a set-piece pipeline is over.

The time when you have a 4*4 or 8*2 architecture is over. Whether this is in the nV30 is irrelevent as you're still thinking like we did way back in 1997. With the rise in transistor budgets has come the move to added programmability while increasing preformance - at a fundimental level, the architecures are the same.

So, how do you know the fragment/pixel back-end of the nV30 isn't based around 32 processing elements that can form upto 8 virtual pipelines, with the number actual dependent upon the task/fragment being worked on. IMHO, if the architecture is anything like this, regardless if it's generally utputting in a 4 pixel comparable - it's a hell of alot smarter of a design than anything else this side of the P10.

The point is, it's too early to state that it's case A or B. For you to do it is not only ignorant, it shows just how rabid many of you are in your hatred of a particular IHV.

Give me a break Vince....

The bottom line is that in several critical ways it only showws half the performance expected from a 8 pipeline card. Nvidia has been leading people to believe that it is because of immature drivers. Which is obviously not the case. I think Dave H above covered pretty well what They could be construing as 16 operations etc.

To take it a step further and start saying that everyone is extremely Anti-Nvidia-biased in this case is just silly. None of this would even be happening if Nvidia had simply addressed what the Nv30 was from the beginging. The problem is that they have been Snow jobing everyone for about a year now, and stacking the deck with mounds of Techno-Jargon.

You simply cant call the Nv30 an *8 pipeline card* when in reality is is a 4x2 for all traditional rendering situations. This is the same treatment that ATi or anyone else would be receiving if a similar revelation occured.

You simply cant deny that in spite of all your or Nvidias posturing on the matter it has exactly 1/2 the DX9 shader power it should be displaying as an 8 pipe card. Its pretty obvious now that the issues J.C. and others were calling *Driver issues*.. never were.

 

[KimB]

]You simply cant deny that in spite of all your or Nvidias posturing on the matter it has exactly 1/2 the DX9 shader power it should be displaying as an 8 pipe card. Its pretty obvious now that the issues J.C. and others were calling *Driver issues*.. never were.

No, it isn't exhibiting that at all. The performance looks to be, under some benchmarks, half that of the R300. The R300 is running at a lower clock speed. I still say it's better to wait a little bit (i.e. until wide availability) until judging the shader performance.

However, a purely-synthetic shader that, say, just does a ton of madd's would probably not be limited by driver problems, and should be able to expose the actual peak shader performance of the NV30. But any complex shader has the potential of poor drivers currently getting in the way of performance.

 

[Crusher]

(and even back then the NV30 was quoted as having 8 rendering pipelines).

And who's to say it doesn't? Only NVIDIA knows what they have built, and all anyone can say so far is that it performs similar to a 4x2 architecture, in some situations. If the Inquirer is to be believed, and the quote they give is accurate, then it is also similar to an 8x1 architecture in other situations. What Vince is trying to say, and it's hard to decipher from the attitude he says it with, is that it's neither of those, but something more flexible that acts like one or the other depending on the situation. There may even be situations where it acts like neither of those, who knows? You can't really test these things accurately, and it could even change between driver revisions. Marketing people always want to quantify their products using familiar terms, so they probably called it 8x1 just because it has the capability to act like that, and that's what they think sounds the best when describing it.

Back in the day, you had the original GeForce that had 4 pipelines, and could do 4 single textured pixels per clock. But when it had to do dual textured pixels, it needed to combine texturing units and was only capable of doing 2 textures per clock. The GF2 added a 2nd texture unit to each pipe, and was then able to do 4 single or dual texture pixels per clock. GeForce 3 came along and it was the same scenario all over again, only this time with dual and quad textured pixels (4 dual/clock, 2 quad/clock). The GF1 was a 4x1 and a 2x2 architecture. The GF3 was a 4x2 and a 2x4. Who's to say the FX can't be a 8x1 and a 4x2 at the same time for different situations? I think the reason it's hard to define has less to do with NVIDIA's marketing, and more to do with the fact that the "pipeline" has swollen and bled into other areas to the point that it's not possible to distinguish such characteristics anymore.

 

[Randell]

The hells wrong with you people. The amount of Anti-nVidia bias in here nowadays is sickening.


Like this comment above, are you a moron? Why do you have such a closed mind about this? Kirk just stated, if you're still dumb enough to think, that the era of fixed function pipelines with TCUs bolted on to a set-piece pipeline is over.

phew a bit touchy arent we? I think enough people told you off for that response already.

Have you read this, I think you should - this isnt the console forum, though I guess it is a bit hotter these days

http://www.beyond3d.com/forum/viewtopic.php?t=4445

BTW dont ever try to work out my 'mentality' from a one line, throw away comment.
oh, and I assume the apology will be via pm as I dont see one here yet.

 

[demalion]

Hmm...I think you are missing something that has been said a few times now: if it can't output the pixels to the screen, what is the point of all those "elements"? It is not like the GF FX is significantly outperforming the R300 in shader computations (this is WITH a significant clock advantage).

It can output pixels to the screen. I don't recall the screenshots being completely black.

"it"="elements", not "nv30". Yes, "it" should have been "they", but, err...I didn't say the R300 was infinitely faster, so maybe you could give me the benefit of the doubt about what I meant?

So...your problem is that people are criticizing nvidia for calling this design "8x1" when it, so far, can't seem to actually output 8 pixels a clock in any useful circumstance? Why shouldn't they? Nvidia could have picked some other description that isn't misleading compared to its abilities, couldn't they?

Since when is something not useful simply because it's not the final value in the frame buffer?

What isn't useful is calling something 8x1 when it can't exhibit the characteristics associated with the name in any useful circumstance. Exactly as I said the first time, and strangely resembling-but-completely-different-than the question you posed.

I have little doubt it is using the approach Kirk is talking about, to some unspecified degree. I am also observing that so far it gets equivalent work done at 500 MHz as the R300 at 325 MHz, and using more transistors. Is it really so hard to understand why that might be criticized?

Criticism is fine, as long as you're criticizing things that are actually wrong. For instance, complaining that it's a year late, the drivers aren't as efficient in shading ops as they should be, and it's loud, are all valid criticisms. Saying it's crap because it's barely better than the 9700 Pro isn't what I would consider a valid criticism.

Well, what I'd say (and I presume you want to know since you are pointing this out in reply to me...) is that it compares poorly to the 9700 Pro by barely outperforming it with a very large clock advantage. Oh, wait a second, I did say that, and not what you stated, so I'm not sure why you included the comment...?

If its operation does have more in common with a 4x2 style architecture than an 8x1 style architecture, then there are probably going to be several instances when the FX should be producing less pixels per clock than the 9700 Pro.

Exactly. But, err...why did they call it 8x1 then?

In fact, one of these instances would be single texturing. And I don't think I have to point you to the article on this website that describes how commonplace single texturing is in games Indeed, the fact is that the FX outperforms the 9700 Pro in most of those games, wherein there are several areas that it should be doing worse (i.e. it's doing 4 pixels/clock at 500 MHz, and the 9700 Pro is doing 8 pixels/clock at 325 Hz, equivalent to 4 pixels/clock at 650 Hz). In that respect, looking at actual performance in situations where it should be severly handicapped, it seems to be performing quite well.

With vertex processing running at 500 MHz as well, and factoring that into your comparison, are you sure the GF FX is performing "quite well" in comparison to the R300? How about performance with texture filtering? I suppose I need some examples for discussion of this, as the GF FX performance leads I remember show the GF FX excelling in multi-texturing without texture-filtering.

nv30 underperforms expectations, and the performance is evaluated.

The only two reasons it underperforms expectations, is that people had the erroneous idea that being delayed would make it significantly faster, and that it was supposed to stomp all over the R300 no matter what.

How about we try a complete quote, Crusher?

nvidia said nv30 is 8x1.
nv30 underperforms expectations, and the performance is evaluated.
nv30 is theorized to actually be more accurately described as 4xSomething.
You come in and criticize people for describing it as 4xSomething.

Gee, I didn't mean that sentence as an absolute statement after all, but as part of a brief description of what Vince did. Imagine that, quoting it in isolation sort of distorts what I was saying, doesn't it?

I find it humorous that this will probably be the best performing "flop" the world has ever seen. Not to detract from the accomplishment of the R300, which certainly is a fine product, but a noticable percentage of people around here would fly off their rockers at the smallest bit of criticism anyone gave it, and the same crowd seems to be unwilling to give the NV30 any credit for going through delays and problems and still end up both on the market and in reasonably decent shape compared to the competition.

Yes, it is a high performing part. No, it is not in "reasonably decent shape" compared to the competition, and yes I'm aware that that evaluation is an opinion...but why is it an invalid one? No, I will not "fly off my rockers" at a criticism of the R300 as long as the criticism is substantiated (sometimes I don't even bother if it isn't, imagine that), but I do wonder why is it ok to do this in response to substantiated criticism of the nv30. No, I am not part of some "crowd" (that's a new one, isn't it? ) that is unwilling to give the nv30 credit, but I do happen to be one person that doesn't see reason to give it the credit some choose to.
I'd suggest your time would be better spent on pointing out why I should give it credit based on analysis (Hmm...Chalnoth seems to be doing this so far...I guess that's why I'm not having this type of discussion with him yet?) instead of an ideological dispute (As Vince has been doing, and you seem to want to continue).
No one is saying the final evaluation of the nv30 is complete, or else we wouldn't be looking forward to Wavey's review, would we?

Instead, people have been doing nothing but criticizing every aspect of it, even making unfounded claims like the one you made eariler about pixel output being meaningless unless it's the final pixel.

Umm...you mean the silly distortion you made of my statement?

I know people think I'm an nvidiot, and dismiss most of my ramblings as someone who's blind, but the only reason it seems that way is that most of my posts are countering what I consider to be unfair, one sided criticism. Of course, the gate swings in both directions, but the net flow seems to be coming from the red and white side.

Ack, that "sides" stuff again. I think that your dividing things in such a way causes you to read my text less than logically. By the way, one-sided criticism need not necessarily be "unfair"...I'm at a loss as to how to get across the simple idea that just because they are "your side" doesn't make criticism of them unfounded. I'm curious, how do you think I reacted to the Quack issue at Rage3D?

Will it help if you go search these forums and see that I've defended "your side" on occasion? Can we get past this "don't hold this opinion" and to a "hey, I think your opinion is wrong for these reasons" (with reasons more than "I want it to be") type of discussion?

 

[THe_KELRaTH]

If it turns out that the NV30 does in fact have a flexible pipeline where the format can change in order to provide the highest level of efficiency why on earth didn't the NV PR ppl supply this info.

 

[Crusher]

It seems you're viewing me as defending Vince as opposed to simply addressing your statements for their own value. I never said Vince's response was called for, and I think the only time I mentioned it was when I was talking about one of the underlying points he made. That point being that you can't blame something for being 4x2 instead of 8x1 if it's really neither of those things, and I extended line of reasoning with my own view, which was to say the reason marketing called it 8x1 was probably because some engineer informed them that it's capable of producing 8 single textured pixels per clock under some circumstances. That doesn't mean it's capable of running a single textured fillrate test at 8 pixels per clock, but that doesn't make it a useless ability either, as you seem to claim it is. Wait, I better make a complete quote of exactly what you said so you can't claim I'm manipulating your meaning again...

What isn't useful is calling something 8x1 when it can't exhibit the characteristics associated with the name in any useful circumstance.

And from the Inquirer article:

"GeForce FX 5800 and 5800 Ultra run at 8 pixels per clock for all of the following:
a) z-rendering
b) stencil operations
c) texture operations
d) shader operations"

Therefore, your statement means that you consider z-rendering, stencil operations, texture operations, and shader operations to all be useless. I don't agree with you on that point, and I don't get the impression many other people would, either.

 

[PiNkY]

Without moving too far off-topic... The "dual texturing won't matter in the future" phrase certainly seems to have some valid points to it, yet IMO the question remains at which timeframe "future" is linked to in that context. I am not too sure that we will cross the point in nv30's/nv35's lifespan where 3-4 texture lookups/pixel (i guess that will be pretty much the average amount for the next two years) are completly masked by fragment processing. Especially complex shaders will propably be confined to very few polygons in a given scene for quite some time, while lots of polygons will not use shaders at all (or very simple ones...). In those cases, multitexturing-capable hardware will still have perfomance advantages. So having Yx2 hardware doesn't seem to be a bad idea to me for the time being (especially since it doesn't seem to be too costly to implement (die-space-wise) and benefits almost any application).

 

[PiNkY]

Edit: Double Post...

 

[Hellbinder]

Therefore, your statement means that you consider z-rendering, stencil operations, texture operations, and shader operations to all be useless. I don't agree with you on that point, and I don't get the impression many other people would, either.

I dont think you are following correctly. In the examples given 8 seperate pixels in a *proper* sense are not being worked on in paralell. which is why there are so many posts and comments on the issue. You need to read between the lines a little more carefully.

Its PS and VS operations are clearly not happening (IMO) at 8 pixels but 4. That is including what Chalnoth was suggesting. Which if he stops to think will see the point im making entirely. If clock speeds were Equal then you would expect to see exactly 1/2 performance from 1/2 the pipelines. But Nv30 also has a 175mhz core clock speed advantage. Which then explains how it can be showing 1/2 the performance and still have a little room for driver enhanement. Because of the 175mhz core advantage.

When you break it down it is painfully clear that the Nv30 in many cases is only operating at 4 pixels per clock.

I think people are forgetting about the Nv30ultras clock speed advantage. Its almost 200mhz. Thats 2/3 the total core speed of the 9700pro.

 

[martrox]

Vince & Crusher,

The main reason for this debate is not the validity of the 4x2 vs. 8x1, but the estimates that nVidia is at best misleading everyone & at worst lying about the configuation of the NV30. How it performs is something else, and all you guys need to do is just go to any nVidia forum and see what your peers (ardent nVidia supporters) are saying. Most are extremely disappointed by what they have been told..... And, as many have said, this is a byproduct of what nVidia PR has led the public to believe. It's a case of comparing PR spin to facts. Granted, not all the facts are in, and most here are not dismissing this. In fact, we wnat more information, which nVidia doesn't seem to want to supply. Just how hard would it be to:

1) admit it's a 4x2, or
2) explain IF it's a 8x1 just want is going on to make it appear as a 4x2, or
3) IF it's some sort of hybred - as Vince alludes - then just explain it in ANY (complex or simple) so that IF it's too much for some of us "morons" to understand, I'm sure there are many here that can put it into layman (morons?) terms......

 

[THe_KELRaTH]

Heh, it maybe a tad difficult to get any info from NVnews forum as any threads that might put Nvidia in a bad light are being closed.