GPU not maxed out (Radeon RX 5700 XT)

[ajm]

Slow day, so I took a closer look at how F@H is using my PC.
The CPU is always maxed out, that is, FahCore_a7 is seemingly using every free bit, dynamically, so that the total of all cores stays at 100% (according to the Task Manager). But not so for the GPU, which fluctuates between 80 and 93% (according to GPU-Z and Task Manager / Compute 1), and from time to time drops to almost 0% for a little while.

Is that normal or is there something I could/should do to optimise the GPU load?
The thermals are quite good (CPU <70C, GPU <70C and Hot Spot ~80C).

[JimboPalmer]

Welcome to Folding@Home!

Ideally, one CPU thread per GPU is devoted to feeding data to the GPU across the PCIE bus
If you are really using all 28 threads on CPU tasks you may wish to step down by 1.

IF you read the log, when a Core_a7 task begins, look for parameters like -np or -nt, the number after that is the number of CPU threads a7 is using.

No matter what you do, some time will be waiting on the Bus, not actually computing. GPU-Z may well show how busy the Bus Interface is. (The faster your card computes, the more time waiting on the PCIE bus)

Ideally you would have a PCIE x16 slot for the card as that is the fastest.

[MrFrizzy]

I have a 5700 XT and have been folding for several hundred projects now. I typically see 90-95% usage on average with dips well below that and some spikes higher. Some projects average closer to 90%, others closer to 95%, but it is very rare to see my GPU at 99% usage for more than a second or two. I do notice that if I am folding on my CPU, the overall GPU usage does go down a little. I have a 6C processor so if I were to let the system run with 10 threads, my GPU usage can drop 5% or more overall whereas if I am only using 6 threads, it may not even be a 1% drop in GPU usage.

So it comes down to do you want to max out the points you are getting (and therefore maximize on the research), or try to max out the utilization on your GPU? If it is the former, let the client choose how many CPU threads to use (-1 in the settings) and let the GPU do its thing. If it is the latter, perhaps drop your CPU threads to half or less and let your GPU get pushed as much as the client can.

[ajm]

Thank you JimboPalmer!

Actually, I allocated only 24 cores to F@H (and i do find -np 24 in the log), but during a job, all cores are seemingly stuck at 100% (which is fine).
Yes, that card is on a PCIE X16 slot.

OK, then, I'll carry on like that.
Albeit, one more question: I intend to also install a 1080 ti, that is collecting dust, in this PC, maybe tomorrow. Anything I should be aware of, because of the mix AMD/NVIDIA?

[JimboPalmer]

I have no first hand experience with it, but I am told you may need to hand alter your GPU indexes so the right driver is used but the right card.

Best of luck! More work means a better chance of getting a cure!

[ajm]

So it comes down to do you want to max out the points you are getting (and therefore maximize on the research), or try to max out the utilization on your GPU? If it is the former, let the client choose how many CPU threads to use (-1 in the settings) and let the GPU do its thing. If it is the latter, perhaps drop your CPU threads to half or less and let your GPU get pushed as much as the client can.

Thank you MrFrizzy!
Makes sense. I'll try to lower the number of cores, then, because the GPU are obviously much better at crunching those WUs!

Best of luck! More work means a better chance of getting a cure!

Indeed! Thank you, you and the whole F@H team - you're are doing a wonderful job here!

[JimboPalmer]

Indeed! Thank you, you and the whole F@H team - you're are doing a wonderful job here!

The folks whose name is a bold dark blue are on the F@H team, the rest of us are just volunteer know-it-alls. And I am the bottom of that list! <G>

[Joe_H]

Albeit, one more question: I intend to also install a 1080 ti, that is collecting dust, in this PC, maybe tomorrow. Anything I should be aware of, because of the mix AMD/NVIDIA?

The mixed configuration can work, a number of people have done it since this version of the client was released. But it can be tricky. Best results have come after a uninstall and reinstall of the client software after installing the new hardware and drivers.

The uninstall should include data to remove configuration settings that would point to the old hardware setup. The installer does a better job of scanning for hardware than trying to update the settings on an already installed client.

One thing to be aware of, others have run into problems on Windows after updating video drivers. Sometimes the update will change the indexes used by the client to determine where a specific GPU is and its connection to a GPU folding slot. So a folding slot for the AMD card might then be pointing to the nVidia, and the GPU folding core will not work and error out.

[ajm]

Great, thank you Joe_H, very useful!

I'll take the long road, then. Uninstall F@H thoroughly (with Revo Uninstaller, so I'm sure everything is gone), then install the new card, update the drivers and software and then reinstall F@H. I'll just keep a copy of config.xml somewhere for reference.

One thing that came to my mind: I'll probably won't have enough PCI lanes for both cards to run on PCIE 16x. Would that hinder F@H from working properly or at full capacity?

[JimboPalmer]

It will spend more time sending data back and forth over the PCIE bus, so the GPU utilizaation will go down, but not by much, if both slots can do x8.

[ajm]

Oh, OK, thank you!
I was almost hoping that it would give me an excuse for upgrading ahead of schedule.

[ajm]

For the record, it works, flawlessly. It took several hours for me to be really sure, that is, to have a job on both GPUs, albeit not simultaneously yet. Quite nice!