r420 may beat nv40 in doom3 with anti-aliasing

[PatrickL]

Why would there be any increases for DX9.0c? This is just a tweak to enable the specific implementation of SM3.0 in GF6, its highly unlikely to affect anything to do with SM2.0 or before.

Ah, but you forget some guys are still expecting that mythical DX 9.1 that will increase performance of their 5900 by 60 % at least !

 

[THe_KELRaTH]

I was rather looking at 5-10% - something that the software developers and/or Nvidia considered worth the time to program alternative paths for.

 

[Dave Baumann]

It's been mentioned that in FarCry using SM3 path over SM2 will be just a performance boost for 6800u and that there will be no visual differences

And this is something that needs to be supported by the developer.

 

[Evildeus]

It's been mentioned that in FarCry using SM3 path over SM2 will be just a performance boost for 6800u and that there will be no visual differences

And this is something that needs to be supported by the developer.

And that is a thing that will, don't you think? as Nv and Ati will deliver SM3.0 parts

 

[Dave Baumann]

For the most part its not something that will "just happen" when DX9.0c runtime is installed.

 

[THe_KELRaTH]

I think that really takes us back to my original post regarding the small performance boost for 6800x using SM 3 path. Yes it may require specific support and I would expect this support in 3dMark03 via an update as it's a DX9 benchmark.

 

[Evildeus]

Sure. That being said, SM 3.0 will be supported by developpers, as Nv and Ati have/will have SM3.0 chips. So why are you escaping the answer?

 

[Dave Baumann]

So why are you escaping the answer?

I'm not, you're changing it. The original point was "DX9.0c may increase 3DMark03 scores for 6800" and this is not the case that this will happen just by installing the new runtime.

 

[Evildeus]

I'm not, you're changing it. The original point was "DX9.0c may increase 3DMark03 scores for 6800" and this is not the case that this will happen just by installing the new runtime.

No you are escaping, i didn't change anything. Look at the thread:

THe_KELRaTH wrote: It's been mentioned that in FarCry using SM3 path over SM2 will be just a performance boost for 6800u and that there will be no visual differences

To which you answered: And this is something that needs to be supported by the developer.

So i said: And that is a thing that will, don't you think?

And you answered: For the most part its not something that will "just happen" when DX9.0c runtime is installed.

Which is clearly escapîng the question. So, will developpers support SM3.0 as Ati and Nv will provide SM3.0 parts? I don't think it's much difficult to see the answer

 

[anaqer]

DX9.0c was brought up wrt performance increase before Dave mentioned it...

Another point of consideration is will the 6800u 3DMark03 performance increase with Dx9.0c?

 

[THe_KELRaTH]

My appologies, I didn't realise I had to explain my point to the smallest detail while conversing with others even more fluent in Video technologies than myself.
.....

Actually reading my post again I see your point. When I think of DX9.0c I think of SM 3 also as without either both are meaningless. (Plug n Socket).

 

[Dave Baumann]

Which is clearly escapîng the question.

Its a question that wasn't related to his initial statement. He initially stated "Another point of consideration is will the 6800u 3DMark03 performance increase with Dx9.0c?" which gives the clear indication that the belief was that just the use of DX9.0c runtime will provide a performance increase, which is not the case as it will require developer support to do it, something that is not going to happen for 3DMark03 - I've never said that there wouldn't be potential increases for other future applications or patches to current applications, but these will require support from some party (other than Microsoft) to do so.

 

[Evildeus]

I agree with the 9.0c part, and the second part. So 3dmark03 won't include SM3.0? Thanks for the precision. What about 3dMark04(5?)? 8)

 

[Dave Baumann]

Why would they change 3DMark03 when they have already stated its not their policy to alter the tests in a fashion that could alter the scores?

 

[Evildeus]

Well, why not? A DX8 part can be benchmark, a DX7 also (i think :? ), why not make some special SM3.0 test? The more the better won't you say? If it's clearly distinguished...
*edit* not that i mind, i don't use 3dMark

 

[THe_KELRaTH]

Would it be possible for Nvidia to force SM3 path for SM2 calls in drivers as they have previously done with 16bit?

 

[Colourless]

Forcing a SM2 compiled shader to run as SM3 will do nothing at all.

 

[Dave Baumann]

AFAIK there are cases where if you recompile an SM2.0 shader as a SM3.0 in HLSL there can be more instructions compiled out for the SM3.0 version.

 

[cristic]

AFAIK there are cases where if you recompile an SM2.0 shader as a SM3.0 in HLSL there can be more instructions compiled out for the SM3.0 version.

How come ? Are there cases when you have to emulate a SM2.0 instructions with more SM3.0 instrunctions because there is no equivalent ?

I though SM2.0 should produce the same output if compiled as a SM3.0 because SM2.0 is just a subset of SM3.0... (or is it not ?) :?

Hope it does make sense what I'm trying to say..., I'm new at this

 

[Xmas]

Maybe it's a matter of the ps_3_0 profile "preferring" other instruction sequences, or different weights for temp register usage vs. more instructions. Just like with ps_2_a.

But PS2.0 is a real subset of PS3.0, so you can just change the shader version of a PS2.0 assembly shader and run it as PS3.0, though there will be no difference whatsoever (assuming no bugs).