Pixel/Vertex Shader performance in 5x.xx NVIDIA Drivers versus the previous 4x.xx

[PreservedSwine]

Just came across this at Digit-Life, looks like a good read.


http://www.digit-life.com/articles2/40vs50/index.html


hope this isn't a re-post!

[Reaper106]

Originally posted by PreservedSwine
Just came across this at Digit-Life, looks like a good read.


http://www.digit-life.com/articles2/40vs50/index.html


hope this isn't a re-post!

Thx Bro

[aapo]

Originally posted by PreservedSwine
Just came across this at Digit-Life, looks like a good read.

Hmm, the 5x.xx PS2.0 performance seems bad. Those drivers certainly aren't "the Wonder Dets" everyone seemed to be waiting for. The author himself states as a conclusion:

"HOwever, the drivers are not a cure-all for NVIDIA, and the real solution of the problem can be found only in hardware, be it NV38 or, more likely, NV40."

[Hanners]

Originally posted by aapo
Hmm, the 5x.xx PS2.0 performance seems bad. Those drivers certainly aren't "the Wonder Dets" everyone seemed to be waiting for. The author himself states as a conclusion:

"HOwever, the drivers are not a cure-all for NVIDIA, and the real solution of the problem can be found only in hardware, be it NV38 or, more likely, NV40."

I think it's interesting to compare the improvements in shader performance here against the improvement in widely-used synthetic benchmarks and games...

[Ruined]

I wonder why Rightmark showed less of an improvement than Shadermark 2.0? AFAIK both came out after the latest dets

[reever2]

Originally posted by Ruined
I wonder why Rightmark showed less of an improvement than Shadermark 2.0? AFAIK both came out after the latest dets

It all matters what kind of shading they are doing

[Hanners]

Originally posted by Ruined
I wonder why Rightmark showed less of an improvement than Shadermark 2.0? AFAIK both came out after the latest dets

Wasn't a lot of RightMark written using Cg? If so, the code would already have been pretty well optimised for the NV3x architecture, so I doubt there was much performance left to wring out of it.

[aapo]

Originally posted by Hanners
Wasn't a lot of RightMark written using Cg? If so, the code would already have been pretty well optimised for the NV3x architecture, so I doubt there was much performance left to wring out of it.

Probably so. The DX9 PS2.0 shading performance in Shadermark has gone up relatively much, and it seems to be legit optimizations this time. It's nothing to write home about, but anyway, FX-series drivers offering better shading performance without cheats???

Mebbe nVidia is finally on track again - or maybe not considering the AF filtering "anomalies".

[Skuzzy]

Speculation: Could be tossing one thing out to improve another? As AF is done in the drivers, simply short circuit the amount of AF work done when a shader needs to be applied. Also stop using tri-linear filtering might gain some registers for other operations.
Just a guess.

Originally posted by Skuzzy
Speculation: Could be tossing one thing out to improve another? As AF is done in the drivers, simply short circuit the amount of AF work done when a shader needs to be applied. Also stop using tri-linear filtering might gain some registers for other operations.
Just a guess.

I would think that Shadermark would be more shader limited than fillrate limited. Of course, I don't know that for a fact...or even for a educated guess.

[Drumphil]

correct me if im wrong, but isn't the main practical evidence that shader mark is shader limited rather than fillrate limited, the fact that the 9600 cards do better than half the performance of the 9800 cards with less than half the memory bandwidth and fillrate?