AMD RX 7900 XT F@H Stats

[Johan45]

I see a few have been curious so I have started a session with the 7900 XT and so far it's right about where people assumed it would be 4-5 mil PPD or about half of the 4070Ti and it's using ~ 280W. these guys have some info https://folding.lar.systems/gpu_ppd/brands/amd/folding_profile/navi_31_radeon_rx_7900xtxtx

I'll post up some screen shots after it runs a bit longer.

 

[freeagent]

So a little slower than a 3070 Ti, Interesting. I can do about 5.5-6m a day with a tune on my card.

 

[JLK03F150]

That is not much faster than my RX 6800 XT & I expected more from the RX 7900 XT. Mine has been 3.3-4 mil PPD over that past 5 days. It was pushing 5 mil PPD before the last tree fell in my neighborhood.

Thanks for running the test! WTF!

 

[don256us]

Really appreciate you sharing. Thank you.

 

[EarthDog]

Big oof on the results, but good to know regardless.

 

[Johan45]

Ok here it is, it looks like we hover around 300 W+ and 4.8 Mil PPD over a 24 hr period. I also have F@H set for Parkinsons whether or not that makes a difference.

 

[EarthDog]

Ugly. These aren't too efficient compared to NV cards.

Is that a driver thing or has AMD always been that way for F@H?

 

[Johan45]

Ugly. These aren't too efficient compared to NV cards.

Is that a driver thing or has AMD always been that way for F@H?

I honestly don't know, I would have thought it would be better since it has double the FP32 capabilities of the 6900 series but performs only slightly better

 

[dfonda]

My thoughts were that AMD's code to run Cuda isn't nearly as efficient as Nvidia. I don't know about openCL/GL.

 

[Johan45]

My thoughts were that AMD's code to run Cuda isn't nearly as efficient as Nvidia. I don't know about openCL/GL.

AMD doesn't run CUDA, which is proprietary to Nvidia. AMD only uses OpenCL (open compute language) CUDA can be up to 30% faster in general. They both use FP32 computing, the 7900XT's theoretical FP32 performance is 51.6 TFLOPS, and the 4070 Ti is 40 TFLOPS. So with a 30% advantage, you would think they'd still be relatively close. Is it possible that F@H has been streamlined for CUDA?

 

[freeagent]

Is it possible that F@H has been streamlined for CUDA?

For sure. But some of it lies with AMDs code too. Maybe in time we will see some optimizing.

 

[JLK03F150]

Here's a note from the AMD Adrenalin 22.7.1 update. Clearly they still have some work to do. Current driver for the 6000 series is 22.11.2 & 22.12.1 for the 7000 series.

Fixed Issues​

• Lower than expected Folding@home™ compute performance with OpenCL™ API on some AMD Graphics Products such as the Radeon™ RX 6800.

 

[dfonda]

AMD doesn't run CUDA, which is proprietary to Nvidia. AMD only uses OpenCL (open compute language) CUDA can be up to 30% faster in general. They both use FP32 computing, the 7900XT's theoretical FP32 performance is 51.6 TFLOPS, and the 4070 Ti is 40 TFLOPS. So with a 30% advantage, you would think they'd still be relatively close. Is it possible that F@H has been streamlined for CUDA?

does amd use cuda - Google Search

www.google.com

 

[JLK03F150]

I wonder if the AMD Pro driver (compute driver) would function better than the Adrenalin driver. I might give that a try at some point just to see, but I will need to install the Adrenalin driver for gaming after the test.

 

[Johan45]

does amd use cuda - Google Search

www.google.com

The answer is still no; porting CUDA apps through a translator doesn't really count. Nvidia keeps their stuff close to the chest

 

[dfonda]

The answer is still no; porting CUDA apps through a translator doesn't really count. Nvidia keeps their stuff close to the chest

My point was that cuda creates higher performance in fah. AMD does not run it natively hence it can't match Nvidia's performance. You agreed with what I said....

I thought that 15-30% was improvement over what nvidia currently had done. Not that it was 30% faster than AMD.

Folding@Home gets 15-30% performance boost with NVIDIA CUDA - VideoCardz.com

TOGETHER WE ARE EVEN MORE POWERFUL: GPU FOLDING GETS A POWERUP WITH NVIDIA CUDA SUPPORT! CUDA support comes to Folding@home to give NVIDIA GPUs big boosts in speed, and you don’t have to do anything to activate it! GPU Folders make up a huge fraction of the number-crunching power power of...

videocardz.com

 

[Johan45]

CUDA is about 30% faster at everything in general compared to OpenCL, was my point. It's possible whoever designed the F@H core has boosted that even more