7900 XTX under utilized in F@H?

[Livin]

I just got my 7900 XTX replaced and back in my rig and wanted to use it for the remainder of the folding month.  Everything testing in benchmarks and games seems to be working well, but F@H seems to only be using the card at ~60% and power is sitting between 100w - 150w (benchmarks will typically push it above 400w).  I've had it running overnight to see if anything changes, but still seems to be the same with production of around 1.5M ppd (less than half the expected output of the card).  Anyone have any ideas on what I can do?  Is it just a matter of what projects I'm getting assigned?

[Zando_]

You've got CPU cores free to feed it? It could also just have little work to do, IIRC they have trouble trying to issue enough WUs during the big folding events. Though my 2060 Super at home isn't having any issues running the usual 1.5-1.8M PPD it usually does, so I would expect the 7900 XTX to run much higher. 

[Mihle]

Based on this:
https://folding.lar.systems/gpu_ppd/overall_ranks

Yes your GPU is underpeforming by a lot.
My RX 5700 XT is doing 1.3 million PPD or so if I had run it all the time.

I do not know why tho.
You dont starve your GPU by running it on few PCIe lanes or something else running in the background?

[Livin]

Very strange, I get home from work today and it's actually working harder. 

 

Average is still pretty tanked, but current looks much more like normal.

 

7 hours ago, Zando_ said:

You've got CPU cores free to feed it?

Yup, 13700k idling along when both screenshots were taken.
 

2 hours ago, Mihle said:

You dont starve your GPU by running it on few PCIe lanes or something else running in the background?

Nope, 16x link speed and as above, system is otherwise idle during both screenshots.

very confused, I guess we'll just see how it goes.

[cbigfoot]

sadly AMD is way under utilized on version 7 at least...im tempted to test v8...i have noticed SOME WUs push my 5700xtx harder than others. my 2060 is amost 1.5-2x the ppd. and is TECHNICALLY lowerend...

 

funny i go to look at my utilization and it makes a Lier out of me....17641 (0, 70, 48) My WU

 

NO i don't fold on my cpu due to temps in the room (if highs get below 60 i may spin up some cpu... Specs:

- Calculon (MainTower): Fractal Design Meshify S2 White | Win 11 Pro | AMD Ryzen 3900x | 32GB DDR4-3200 TridentZ RGB | Noctua NH-D15 | 1x SN750SE | AMD 5700XT XFX Rx-57XT8OFF6 | Gigabyte X570 WiFi Pro | Seasonic FOCUS Plus 750 Gold SSR-750FX | D3218HN & Dell U2719W & Dell S3422DWG

[Livin]

On 11/10/2023 at 7:14 PM, cbigfoot said:

i have noticed SOME WUs push my 5700xtx harder than others.

Yeah, let it run for the day yesterday and the utilization varied wildly depending on the WU. It's a shame.,the most I'm seeing is around 85% utilization and average realistically around 40%. The 3060 Ti I was using before had no problems averaging 80%+ over the course of a day. 

 

Well, I have another system I can get the 3060 Ti re-setup in for the remainder of the event so I'll probably do that. Disappointed the 7900 XTX couldn't help more

[TatamiMatt]

Yeah, some WUs vary pretty drastically, my 7800xt nitro+ that ive overclocked myself further to over 2.6GHz for both memory and cores can run anywhere between 1M and 4M depending on the WU it is folding at the time

[HoldSquat]

Wow that's weird, I am getting about 2-2.5M PPD running a 5700 and a 6700XT at 100% utilization. Wonder if its something with windows.I am running linux but I have it really cut down on what's running in the background.

 

https://folding.extremeoverclocking.com/user_summary.php?s=&u=1332538

 

 

[TatamiMatt]

1 hour ago, HoldSquat said:

Wow that's weird, I am getting about 2-2.5M PPD running a 5700 and a 6700XT at 100% utilization. Wonder if its something with windows.I am running linux but I have it really cut down on what's running in the background.

 

https://folding.extremeoverclocking.com/user_summary.php?s=&u=1332538

 

 

Linux is the recommended OS for folding I believe. And yes im running windows without cutting down on too much background also, fan control, openhardwaremonitor, afterburner etc all running.

 

So the difference in OS is possibly the difference in score. Im also only running my dGPU, not utilising either the CPU or iGPU as i dont like my system running loud and the majority of the CPU taken up

[HoldSquat]

2 hours ago, TatamiMatt said:

Linux is the recommended OS for folding I believe. And yes im running windows without cutting down on too much background also, fan control, openhardwaremonitor, afterburner etc all running.

 

So the difference in OS is possibly the difference in score. Im also only running my dGPU, not utilising either the CPU or iGPU as i dont like my system running loud and the majority of the CPU taken up

Oh yea I am def the opposite, I have it 100% all day and all night unless I am gaming, then I might pause it for a couple of hours. If my water pump is going 100% anyways might as well use it for folding

[Livin]

20 hours ago, TatamiMatt said:

Yeah, some WUs vary pretty drastically, my 7800xt nitro+ that ive overclocked myself further to over 2.6GHz for both memory and cores can run anywhere between 1M and 4M depending on the WU it is folding at the time

This is exactly what I'm seeing.  It's (sort of) good to know I'm not the only one.

[Gorgon]

On 11/13/2023 at 6:18 AM, TatamiMatt said:

Linux is the recommended OS for folding I believe. And yes im running windows without cutting down on too much background also, fan control, openhardwaremonitor, afterburner etc all running.

 

So the difference in OS is possibly the difference in score. Im also only running my dGPU, not utilising either the CPU or iGPU as i dont like my system running loud and the majority of the CPU taken up

Linux used to have a larger lead over Windows when it came to Folding but changes in Core 21 changed all that and dramatically reduced the difference. Now there is still a performance bump with Linux over Windows but it it is typically only about an advantage of 5% or less depending on the Work Unit (WU). Mostly down to Linux having better (more efficient) mechanisms for PCIe Bus arbitration.

 

Under the hood Folding uses OpenMM for modeling Protein Folding. On GPUs both OpenCL and CUDA are supported for OpenMM. AMD supports only OpenCL whereas NVidia supports CUDA and OpenCL. CUDA always enjoyed a 10-15% performance increase over OpenCL for Folding but during the Pandemic the CUDA Libraries were optimized further by NVidia and now CUDA out-performs OpenCL significantly.

 

The net result is, though on paper the Dual-Precision Floating Point (FP32) throughput of a GPU (TFLOPs) should be a good indicator of the yield of a GPU while Folding, in practice NVidia GPUs are almost double the yield of a AMD GPU with comparable Dual-Precision throughput.

 

So while AMD GPUs may be the value king for rasterization in Gaming, Nvidia GPUs end up being a better value for Folding despite their expense.

 

AMD could likely improve the Folding Performance of their GPUs by implementing ROCm on their Consumer hardware but, for some reason they have decided to only support ROCm on their Enterprise "accelerators"