NV40 Coming @ Comdex Fall / nVIDIA Software Strategies

To be honest I wasn't even thinking about units like you are, but more from a CPU viewpoint. So lets say that we have memory load/store units, FPU add/sub units and FPU multiply/divide units and maybe some very exotic ones. So, a pixel shader for one pixel might use from every available resource, just like a CPU schedules instructions.

 

Shouldn't this schedule set the alarm bells ringing? Deja vu all over again...

 

On the contrary, NV40 has some very consevative specs right now, which makes me believe, that it'll be here for christmas...

Do not overhype it by yourselfs There will be some new features (shaders 3.0 is the main part of NV40), but it won't be "a revolution". Though, there'll be a programmable displacement mapping...

 

Seems to me that both Loci and NV40 should be "launched" this Comdex. Both with availability Q1 2004. Just a hunch though.

VS/PS 3.0 and "programmable displacement mapping" isn't going to matter much at all for gamers. May be an important marketing check-box though. And of course more features is always "fun" for developers and demowriters to play with. PS/VS 3.0 v.s 2.0 is going to be similar to PS 1.4 vs. PS 1.1-1.3.

What will make or break these cards is the same thing that makes and breaks every other card: benchmarks. And by the time these cards come out "Doom3" will of course be the benchmark to win, for better or worse. The real key to NV40 vs. Loci is going to be about performance. And right now, the rumors are similar performance increases for both.

NV40 could have an advantage in that, if both parts perform similarly, and NV40 also has "more features", that becomes a logical winner. On the other hand, "brand new architectures" (as NV40 is being touted) often come with driver baggage and issues, not to mention higher risks of not getting the part out on time, so Loci stands to have an easier road to production and release.

 

I wouldn't necessarily be 100% sold on the idea that ATI's .130nm high end part (whatever the damned thing is called) will map 100% to R300/R350.

As for the idea of 'Unified shaders' there are degree's to which these things can be 'unified'. I'd hazzard a guess that VS and PS will still be relatively separate but they may start sharing some resources, which will ultimately lead down the path of a fully unified shader system. Given that NV30 is still very similar to NV25 in much of its configuration and I'd be srprised if they made a jump like fully unifying everything in one go - most unlike NVIDIA's development path.

 

Could you be more cryptic?

I'm not expecting "just" a die shrunk R350...but I'm also not expecting full PS/VS 3.0 functionality either. I'm more expecting upgrades (possibly very significant) to memory bandwidth technology moreso than I am in PS/VS functionality. (I also hope that they continue to increase AA and aniso quality levles). In all honesty, assuming that R500 is scheduled for Fall '04, and is at least PS/VS 3.0 or even DX10.....ATI should be able to get away with "faster" R300 cores until then.

On a non-related note: F@H:....I see you're not going to be caught easily...don't make me whip out another 600 Mhz of power!

 

Yes. My cryptivity knows no bounds!

IMO, and there's not much firm basis for this at the moment, memory probably won't be radically different from what we see at the moment, AA I doubt they'll be much change there - I expect good things from NV40's AA though. ATI reps will state they have PS/VS3.0 "when its needed".

Ohhhh, I'm scared by an extra 600MHz!!!

Joe, I'm cantering at the moment. My main rig isn't 24/7 yet and I've got a springdale review to do meaning that the main rig gets the upgrade to 3.06GHz (presently its 2.5) and then I'll have a 3GHz 800MHz FSB CPU/Mobo in the test rig to tick over with when its not testing!

 

Maybe the Nv40 will use SOI/LOW-K or?
Could someone clear upp this mess with the ATI codenames!
R360, R390, Loci and R400/420. What are they??

 

I hope so...that is perhaps the biggest reason why I won't even consider the NV3x core products vs. the R300 core ones. NV3x AA "sucks."

Hmmmm...interesting. That's quite different from saying they'll have it "when Microsoft Supports it". Sounds more like 3dfx saying they'll support 32 bit color "when its needed".

PhE4r Me!

 

I think the NV40 will use SOI, but not Low K. Could be wrong though.

DegustatoR: I agree PS3.0./VS3.0. isn't particularly ambitious. The PPP is quite nice, as you say, but yes, it's nothing "revolutionary". But do you have facts proving there's no dynamic allocation?


Uttar

 

... well ... err... 8)

 

Sorry Kristof, didn't think about that, hehe

I should have made my sentence clearer:
PS3.0./VS3.0. isn't particularly ambitious CONSIDERING nVidia has already developped CineFX, which is very near to it in many ways, and that they have had a year between NV30 and NV40.

A TBDR with PS3.0./VS3.0., on the other hand... I'd like a review sample of that, you know


Uttar

 

Well, I know of at least one person in the queue beforehand

 

Don't worry, but maybe we can share it? j/k
You get the left tiles, and you get the right one, maybe? I don't know how that's technically possible, but eh...

BTW, you bring an interesting point by saying the NV40 could be *partly* shared.
Actually, it would make a lot of sense. Having dedicate texturing functionality in the VS is insane, and having unused branching in the PS would also be strange. So you could simply share that, and not the rest. No idea how you'd do that technically, though, and how much cheaper / easier to implement it would be compared to a fully shared part.


Uttar

 

Kristof, besides the and why would it be ambitious over the CineFX? nVidia seem fairly close to the VS 3.0 spec (mainly except for the cool texture lookup/address). Will the static and dynamic flow control in the PS 3.0 be very complicated and silicon expensive to implement?

Throw us laymen an easy bone...

 

Which nv35 cards are you referring to? I said in my original post (which apparently you didn't read very carefully) that the two lower-priced nv35 reference designs would probably ship in late June to early July (even though I still think it very odd that even now there are no reviews of these products--that I've seen, anyway.) I think it's accurate to say that the 5900U will be in volume production by nVidia's OEMs in August, but you can of course think what you like.

Heh-Heh...not to worry. In only a few weeks there'll be no guessing about it...

Is that what you guys over at nVnews think?.... (You pretty much deserved that.)

Again, I think a volume ship date of August for the 5900U in volume is likely accurate. I expect the other, cheaper, slower variants of the nv35 reference design will ship in either late June or early July.

I've seen nothing about "R360" from ATi. Could you provide a link on "R360" or "speed bumped nv35's" that contains credible official information that would lead you to draw such conclusions? If not, then how could you make them? Uh, oh--don't tell me--if "nvNews forum posts declare it--it has to be true", right? *chuckle*

 

You mean me, right!? I'd build a new system around such a card right now! I'd beta test a bunch of games, and give lots of feedback, and I'd be SO interested in testing it, and...

Hmm? What? You meant who?

Oh.






Wavey is a bad human being, I tell you!

 

AFAIK R360 is something R3x0-based chip on 150 nano - around July, IMHO.

IIRC R390/Loci/R400/420: they are the same solution, the new high-end chip, probably on 130 nano. But! It isn't the same as the original (year old), brand-new hyppasuppa' architecture-based R400: that became R500... so, current R390/Loci/R400/420/whateveryouwant/Stanley-cup will be kinda 'intermediate' high-end chip 'til the R500 hits the market...

PS: Everything based on rumours (MuFu, hehe), of course.

EDIT: typos and text

 

Dave,

Do you know something about NV40's AA possibilities? IE have you spoken to somebody from nVidia, and they said something like...

"Yeah, yeah...We're well aware of the fact that most people, in the know, regard our AA as sub-standard...But I can promise you that it will all change with our next gen. part."

Something like that?