ATI R600

[Joe DeFuria]

That would be my guess.....though I am just guessing.

 

[PaulS]

I wouldn't be quite so quick to categorise NV50, if I were you.

 

[Evildeus]

I wouldn't be quite so quick to categorise NV50, if I were you.

Which means?

 

[PatrickL]

Edit: Missread something

 

[DW Fan!!!!!]

I thought that comment was refferring to the Xbox 2 GPU not nvidia

 

[ninelven]

What about the R420 refresh?

 

[Dave Baumann]

I say people are being grossly overoptimistoc as to when R500 will be available.

 

[jvd]

I say people are being grossly overoptimistoc as to when R500 will be available.

Hmm i was thinking january , febuary dave . The r600 being based mostly on the xbox 2 gpu would also have a speed up as it should come out the end of 2005 .

Unless the r500s time on the market is the one thats going to get eaten up and not the r420s . Though i would suspect ati to do it the other way around

 

[McDusty]

Can anyone point me in the direction of any articles relating to unified shader architectures? Is the only benefit of combining the pixel and vertex shaders that it becomes easier to program? Or is there more to it?

 

[Joe DeFuria]

Can anyone point me in the direction of any articles relating to unified shader architectures? Is the only benefit of combining the pixel and vertex shaders that it becomes easier to program? Or is there more to it?

AFAIK, it's not just easier to program for, but more "flexible" in terms of performance.

With a unified architecture, you don't have a fixed throughput of "vertex shaders" and "pixel shaders". You have a fixed throughput of "shaders" in general. So, you have more flexibility wrt balancing your engine. You can trade off pixel shading power for vertex shading power when needed, and vice versa.

 

[Megadrive1988]

I don't expect R500 this year at all. I expect a R420 refresh this fall/winter.

then R500 spring/summer/fall 2005

Xbox2 Nov 2005 with custom graphics processor partly based on R600, combined with MS graphics design. might be delayed to summer 2006.

R600 not until 2006

 

[KimB]

AFAIK, it's not just easier to program for, but more "flexible" in terms of performance.

Thinking more traditionally, your vertex shader is not sitting idle waiting for the pixel shader when drawing large polygons, and your pixel shader won't be sitting idle when drawing small polygons.

 

[hstewarth]

I wouldn't be quite so quick to categorise NV50, if I were you.

That is true for NV45, R500, R600 or any thing that is way in the future no matter what the rumors. All that can go is past history of companies releasing there chips..

So anybody heard what is after SM 2.0 / SM 3.0 as far as shaders.

 

[ERP]

Can anyone point me in the direction of any articles relating to unified shader architectures? Is the only benefit of combining the pixel and vertex shaders that it becomes easier to program? Or is there more to it?

The real win with combining them is that you can get better use out of your Vector units.

There is normally a very small cache between the vertex shading side and the pixel shading side, so when your drawing big tri's (or using complex pixel shaders) the cash gets filled and your vertex units sit around doing nothing. When you draw small ones, the cache is empty and the pixel shading units sit around and twiddle there thumbs.

If you can reallocate the ALU's to where they are needed you get better utilisation of your resources.

 

[KimB]

So anybody heard what is after SM 2.0 / SM 3.0 as far as shaders.

Well, we've seen some info on where the next DX version is going in terms of shaders. It's heading for a unified stream architecture. No more pixel/vertex pipelines: just stream processors with special functions for doing, for example, the tasks that go between the vertex shader and pixel shader today (not to mention other tasks that are still effectively fixed-function, such as texture filtering and various framebuffer access techniques).

 

[Entz]

Is the traditional 1 VS/1PS system maintained or can you dynamically customize how many VS/PS are run on a given quad (if it is even quad based).

I.e.
Say you have an effect that requires two passes. Both passes share a common VS but each has its own PS program (What they do is irrelevant for this example but think HL2 style surface effects if you must). Combining the shaders is not possible (flexibility/time/performance etc).

Could you then allocate 2 PS "streams" and 1 VS "stream" for the operation and render the whole thing in one pass? OR is this getting a little too generalized (Need a CMT syle processor) and overcomplicated . Perhaps the ability to schedule X number of PS programs would be enough (would work like multi-texturing does today).

 

[Hellbinder]

There is an R450 that will be the refresh part later this year. Although it may no longer have the *exact* nomenclature "R450". R500 is comming some time after that. R500 is roughly 12+ Months away.

And as far as i understood Xbox2 is based on a variation of R500 not R600. Or more accurately a Hybrid part.

 

[AlphaWolf]

And as far as i understood Xbox2 is based on a variation of R500 not R600. Or more accurately a Hybrid part.

Perhaps you should read this.

http://www.beyond3d.com/forum/viewtopic.php?t=11797

 

[Megadrive1988]

the ideal senario IMHO would be for MS to combine the best of R600+ with the best IP that they have (Cagnet, whatever else) into the best graphics processor they can possibly make for mass production, launching Xbox 2 in summer 2006. it would end up that Xbox 2's graphics processor is never a PC part. even though it shares some technology with R600, it has enough of its own technology (the stuff MS holds) that it never gets released on the PC, in the way that GeForce 4 Ti 4200 was very close to the Xbox GPU.

 

[Hellbinder]

And as far as i understood Xbox2 is based on a variation of R500 not R600. Or more accurately a Hybrid part.

Perhaps you should read this.

http://www.beyond3d.com/forum/viewtopic.php?t=11797

Actually thats precicely what i am commenting on.