[dan2097]

I wish I still had my r300, I guess I could try Unwinders AD for 3dmark2001SE with my R200 and see if those optimisations are still there and just how much they affect performance

Well I can confirm that ATI has had optimizations for the r300 throughout the whole 3.x series and atleast up to the 4.3s as is shown by the fact that nature scores have changed negligably in this period.

My loss in frame rate in nature as I said is comparable to the level of optimization Nvidia is currently using on the benchmark. TBH if I was Nvidia I'd have left the aggressive optimizations in, if your going to cheat why not go the full distance and go for the win :P

EDIT: It would be interesting to see results from an r200, I expect there are still optimizations in the latest drivers, what would be most interesting is to use 3d analyse on old ati and nvidia drivers using the gf3/gf4 and r8500 to find who introduced detection of 3d mark 2001 first :P

[ChrisRay]

I remember early benches of the r200 in 3dmark2001SE where the card was only getting like 25 FPS in NAture, then later after the SE patch and new drivers it was getting 50 FPS, ATI said this was because it wasnt optimised for Shader 1.4,

But AFAIK 3dmark2001SE Game Test 4 has always Used Pixel Shader 1.0

[dan2097]

Originally Posted by ChrisRay

I remember early benches of the r200 in 3dmark2001SE where the card was only getting like 25 FPS in NAture, then later after the SE patch and new drivers it was getting 50 FPS, ATI said this was because it wasnt optimised for Shader 1.4,

But AFAIK 3dmark2001SE Game Test 4 has always Used Pixel Shader 1.0

That is correct Using ps1.4 instead of ps1.0 would definitly be a dodgy thing to do, especially as we can see now ps1.4 had poor adoption.

The thing I remember from the previews most was that 3d mark 2001 scores seemed high like as high as the gf3 ti 500 (although the 8500 did end up with this kind of performance) but its performance in games was intially more comparable to the gf3 ti 200

[StealthHawk]

Originally Posted by dan2097

That is correct Using ps1.4 instead of ps1.0 would definitly be a dodgy thing to do, especially as we can see now ps1.4 had poor adoption.

If you run an equivalent PS1.4 shader better than PS1.0/1 shader, it would be smart to transform all PS1.0/1 shaders to PS1.4(via your shader compiler, of course, not via app detection) as you don't lose anything in going from PS1.0 -> PS1.4, PS1.0 being a subset of PS1.4.

[dan2097]

Originally Posted by StealthHawk

If you run an equivalent PS1.4 shader better than PS1.0/1 shader, it would be smart to transform all PS1.0/1 shaders to PS1.4(via your shader compiler, of course, not via app detection) as you don't lose anything in going from PS1.0 -> PS1.4, PS1.0 being a subset of PS1.4.

Ah yes I see what you mean, In fact I was under the impression that is how the 8500 runs ps1.0-1.3 and that the 9700 does something similar to get 1.1-1.4 working, maybe. However unless you changed the shader to take advantage of ps1.4s abilitys to do the same shading in less passes that wouldnt help would it in terms of performance?

Admitedly I dont know/cant remember how the r8500 performs in terms of pure fill rate in ps1.1 and 1.4 but the 9700 just performs exactly the same in 1.1/1.4/2.0.

[I.S.T.]

Originally Posted by digitalwanderer

Yeah, people are actually going to be able to get the nV40 that nVidia is launching next week. (I'm implying it's going to be another loooooong interval between when nVidia "launches" the nV40 and when people can actually buy them and play them..I think that won't happen until July)

I remember now. Thanks, dw.

[ChrisRay]

*edit* Sorry for Double Post.

[ChrisRay]

Originally Posted by dan2097

Ah yes I see what you mean, In fact I was under the impression that is how the 8500 runs ps1.0-1.3 and that the 9700 does something similar to get 1.1-1.4 working, maybe. However unless you changed the shader to take advantage of ps1.4s abilitys to do the same shading in less passes that wouldnt help would it in terms of performance?

Admitedly I dont know/cant remember how the r8500 performs in terms of pure fill rate in ps1.1 and 1.4 but the 9700 just performs exactly the same in 1.1/1.4/2.0.

Well my 8500 All in wonder DV, In My experience, has ran Slower in shader 1.4 than Shader 1.0/1.1 (I know this seems strange) Tryen to remember that Test Neeyik gave me. it was strange becase my FX 5900 Was doing Shader 1.4 significantly faster in it than Shader 1.1 (of course its synthetic and not sure how to take it)

But this is the same test that showed my r200 dying under shader 1.4

[StealthHawk]

Originally Posted by dan2097

Ah yes I see what you mean, In fact I was under the impression that is how the 8500 runs ps1.0-1.3 and that the 9700 does something similar to get 1.1-1.4 working, maybe. However unless you changed the shader to take advantage of ps1.4s abilitys to do the same shading in less passes that wouldnt help would it in terms of performance?

Admitedly I dont know/cant remember how the r8500 performs in terms of pure fill rate in ps1.1 and 1.4 but the 9700 just performs exactly the same in 1.1/1.4/2.0.

Well, I've never done any graphics coding of any kind, so maybe it's not possible, but I would assume there are at least some things you can safely do to turn a PS1.0/1 shader into a faster PS1.4 shader. Possibly. Maybe.