Vega 56 only 85% utilization

[Juergen_Roland]

Hello,

I am folding using a Vega 56 which is producing roughly 1,000,000 PPD.

Looking at the gpu utilization in msi afterburner it doesn't get higer than 85%.

Has anyone a clue how to unlock the remaining 15%?

Thanks in advance,
Juergen

OS: Win 10
CPU: Ryzen 1700X
Ram: 16GB

[PaulTV]

Edit... 85% of what? My GPU is running on about 80% power draw, but GPU load is 97%.

There may be several reasons your card is running less than 100& utilized:
- The Folding@home core will not use all possible functionality of your card, but for what it can use, it'll use as much as possible. F@h certainly does not use (or need) all memory on your GPU card.
- As far as I understand: the specific job you have may not be able to utilize 100% because of the atom count - this could be the case on the latest and greatest cards.
- If the card runs too hot, it may slow down. You should be able to check that with GPUZ, if not visible in Afterburner.
- F@h needs a CPU core to keep the GPU running, so if the CPU is overloaded, the GPU will not run at it's top capacity.

Maybe there's more that I can't think of right now

[gunnarre]

I second that GPU-Z's "Sensors" tab is very useful, because it tells you if the processing is limited by the power limit, the voltage or the temperature. If none of those are triggering, it's likely that the atom count of the specific project isn't utilizing every compute unit in your GPU.

[toTOW]

If the WU has not enough atoms, it will fail to use all the GPU stream processors ... this is particularly true on AMD GPUs ...

[Juergen_Roland]

Thanks for the answers.

With gpu utilization I meant the gpu load. It shows 85% in MSI Afterburner, GPU-Z and the Task Manager. The Power limit is set in MSI Afterburner to +50%, the card is drawing aroung 150W.

The gpu is not getting too hot, around 64 degree Celsius. I am not running any heavy loads on the cpu while folding, so plenty of power left.

If the theory about the WU not having enough atoms is correct, should't I see different gpu loads with different projects?

[Neil-B]

If the theory about the WU not having enough atoms is correct, should't I see different gpu loads with different projects?

Yes

I use HWINFO64 for hardware monitoring (everyone has their favourite tool) and dependant upon the project my gpu core load, gpu emory controller load, gpu memory usage, total gpu power % of tdp all vary ... very useful for baselining projects and working out which projects perform well on which class of gpus ... There aren't that many projects that really allow the latest fast/wide gpus to perform to their fullest yet - the tough issue being that those projects are likely to take "ages" for wus to complete if on slower/thinner gpus and their is only so much one can play with assignments to get them to appropriate gpus - but things are improving (one project currently in test runs my RTX3070 at 100% core load with 85% memory controller load for 3+hrs) ... It isn't helped that the nVidia/AMD gpus are differently optimised - some projects have to be restricted away from AMD as they just throw too many errors - but again things are I believe/hope improving.

[Juergen_Roland]

Guess I have to be patient and satisfied with the work the gpu is doing. Well thanks for the opinions!

[psaam0001]

I myself have been trying to tweek a system that has been running an AMD iGPU (the CPU is a Ryzen 3 3200G), as for BIOS/UEFI settings. Given that I updated it a couple of weeks ago, and it has not been a smooth process (yes... I have dropped GPU WU's during this time).

So, I'm going to keep trying to fold those proteins...

Paul