speculated Nvidia Roadmap - NV50 (G80) and beyond - just speculation

[megadrive88]

from reading some of the threads on Beyond3D, and listening to Jen Hsun in the recent Morgan Stanley Semiconductor Conference,
I've concluded (perhaps incorrectly) the following:

G80 ~ NV50 (or a revised NV5X) will probably be on 65nm, is due late this year, has more decoupling of the pipeline going on, but is not a fully unifed shader architecture.

'G81' - just a spring 2007 65nm die shrink of G80~NV50, that is, if G80~NV50 starts out on 90nm

G90 ~ 'NV55' (or a revised NV5X refresh) still not a full unified shader architecture, probably due in H2 2007.

then, NV60 (call it G100 if you will) will probably be full unified shader architecture, and is probably due in late 2008.

comments? additions, subtractions, etc ? :)

[killahsin]

my personal speculation is that g80 will be 80nm and the refresh will be 65.

[MUYA]

80nm makes sense as that is the next node that TSMC has right? Maybe the rumoured use of AMD fabs will facilitate the move to 65nm if AMD has that process working or maybe they will just use TSMC...but USA has been poo-pooed only for this generation. It might still make it for g80?

[megadrive88]

In the Morgan Stanley webcast, I thought Jen Hsun said their upcoming next-gen GPU would be on 65nm.

[Mr_LoL]

May I please kindly ask what full USA is?

Never mind I know now. Where are my tablets.

[agentkay]

May I please kindly ask what full USA is?


Unified Shader Architecture. The idea behind it is quite good, but unless you have an extremly fast and efficient scheduler, I think full USA can become a big pain in the ass as well.

[retsam]

i think g80 will have some sort of integrated physics engine.

[Redeemed]

Hah! I saw "USA" and thought you guys were talking about the United States of America for some reason! Haha! I was about ready to go all rambo and defend my country. ;-)

Ya'll scared me there for a second. :D :D ;)

[tieros]

I like the idea of the unified API, but there's no reason to unify the shaders in hardware unless the finished product performs better. They would have to give up any shader-specific hardware optimizations and incur the overhead of a non-trivial scheduler.

If all it does is boost the synthetic benchmark scores, don't bother. If the minimum FPS gets sporatic due to shader starvation, don't bother. If you end up with an excessively clocked, bazillion transistor GPU that needs a 1500W PSU and a tri-slot cooling solution, ... Well, you get the picture :D

[agentkay]

I like the idea of the unified API, but there's no reason to unify the shaders in hardware unless the finished product performs better. They would have to give up any shader-specific hardware optimizations and incur the overhead of a non-trivial scheduler.

If all it does is boost the synthetic benchmark scores, don't bother. If the minimum FPS gets sporatic due to shader starvation, don't bother. If you end up with an excessively clocked, bazillion transistor GPU that needs a 1500W PSU and a tri-slot cooling solution, ... Well, you get the picture :D

Totally agree with that! :applaude:

[megadrive88]

I watched gamespot's feature on Vista and DirectX10 (or Direct3D10) and they said the vertex shaders and pixel shaders are now just shaders. does that mean Direct3D10 *is* unified and requires it in hardware? previously I thought DirectX10 / Direct3D10 did not require unified shaders.

[tieros]

The DX10 shaders all support the same language features, so in that way they are unified.

But the implementation details are all abstracted by the drivers. The shader may execute on specific-purpose hardware, or be sent to a generic shader, or it may even be run in software.

For that matter, the entire "card" can be done in software. It's just not likely to win any speed records :D

[JoKeRr]

I'm curious to know how far is OpenGL going now days, since DirectX see to be quite popular.

[Redeemed]

You know, I've wondered the same thing my self. After all, looking at QuakeIV and DoomIII OpenGL is a very capable API. I just wonder why it isn't used more often than it is.

[tieros]

I think that one of the main reasons that DirectX took the market share away from OpenGL is that DirectX is controlled by a single entity. OpenGL suffers from the same issues that other "design by committee" projects do.

DirectX also handles audio, keyboard, etc. which makes it attractive as a game development platform.

OpenGL will work as well, if not better, in Vista as it does in XP. It's just not going to be used by MS for coding the GUI, which makes perfect sense.

[Red_Shift]

i think g80 will have some sort of integrated physics engine.
For what? I'm for PhysX card all the way. I don't like idea of using transistors for physics' shaders when those could be used for graphics shaders. The shaders maybe be capable of doing both physics and graphics but still it would sacrifice performance.
As far as I know, for now nvidia strategy is to use the havok fx engine, and their ideal situation is to have a SLI system where one card does physics and the other does graphics, but if you're on high-end buying 2 graphic cards is more expensive than buying one graphic card and one physics card. On mainstream (7600GT price zone) it might get a bit cheaper but physX engine is much better than havok fx, and for physics processing nothing is more powerful than Ageia's PPU. With so strong retail partners (Asus and BFG, Asus is spread worldwide) I think Ageia will be very sucessfull.

[Redeemed]

I agree about the PPU entirely. The last thing I want is for my video card(s) to be doing both Phsyics and Graphics- even in SLi. I guarantee that If you went out and purchased to of the best 8900GTX Uber Extreme Editions for SLi, you wouldn't be gaming at a resolution higher than 1280x1024 with everything maxed as only one card would be actually working on the graphics. The other would be too busy with physics and such. Now, I don't know about you guys, but if I purchase an SLi setup it is so I can have the best graphical experience possible. And 1280x1024 is not that. If I'm gonna' be dumping around $1200 or more for a pair of graphics cards, I had better be able to play my games with the highest possible settings for my graphics. And with a video card having to be dedicated to doing physics, that just wont happen.

Sorry nVidia, I am not supporting you on this at all. Any money I invest for physics, already has AGEIA written on it in big, bold print.

[borntosoul]

i dont know, if the card is doing physics then its shader/graphic performance is going to be compromised somewhere - id like to see the cards really maxed out doing what they meant to be doing. im trying to think outside the square but i cant see much benifit for doing physics on the GPU right now. anyway what do we have dual core CPUs for? plus sometime late 2007 or 2008 we might see quad core CPUs.