Pause CPU separate/independent of GPU

[jchang6]

I have several systems with supported GPU. There times I would like (nearly) full CPU for other tasks, that should not put much if any load on the GPU. I notice that on setting Folding Power to Light, the CPU uses half of the logical processor, and the FaH pauses activity on GPU whenever there is even the slightest screen activity?
I am a developer, so I occasionally need full CPU for my work, but rarely generate even the tiniest blip on the GPU.
I am running FaH 7.6.13, Windows 10 2004, Intel Xeon E3/E5 (various gen from Haswell to Skylake), and a mix of AMD Radeon RX5600 and nVidia GTX 2080 Supers.

[bikeaddict]

In FAHControl, the CPU and GPU slots can be paused independently. If you right click on the cpu row in the Folding Slots section at upper left, there is a menu item to pause the slot. It can also be set to On Idle. Then it can be restarted by right clicking and selecting Fold to resume or Finish to finish the work unit and pause.

[jchang6]

thanks, I have confirmed this works. This might need a different thread. Its getting hot now in Massachusetts. I just came across MSI Afterburner, which purportedly also allows lowering GPU power/temp? I am wondering if I can get a significant power reduction without giving up proportionate folding power? the 2080 Super is 2M PPD at 200W+ , what could I get for around 100W?

[bikeaddict]

thanks, I have confirmed this works. This might need a different thread. Its getting hot now in Massachusetts. I just came across MSI Afterburner, which purportedly also allows lowering GPU power/temp? I am wondering if I can get a significant power reduction without giving up proportionate folding power? the 2080 Super is 2M PPD at 200W+ , what could I get for around 100W?

Limiting power to improve efficiency has been discussed here before. Paragon, who sometimes posts here, has some experience with this and runs Green Folding@Home (https://greenfoldingathome.com/2019/02/ ... wer-limit/).

[debs3759]

the 2080 Super is 2M PPD at 200W+ , what could I get for around 100W?

To reduce the power consumption that much, even on the best card, you would need to significantly reduce the GPU speed. I don't know how much, but in MSI Afterburner, you would need to reduce the clock speed and the power used, gradually, and use a tool like Furmark to make sure the card is stable before folding. I would reduce the clock speed by about a third if looking for that sort of power saving, and reduce the power a bit at a time until you find the lowest setting that is stable. Kind of defeats the object of buying a high end card though

After each adjustment + test, you need to save the setting in Afterburner, so that you don't have to go through the same rigmarole each time you turn the computer on. You can always reset to default settings, then reset back to your underclock, with just a click of a button, if you need to change it for anything.

I can't comment on the effect on ppd, as it's not linear.

[MeeLee]

A 2080 you can't lower below 125W. And at that it'll work really slow.
I would recommend you to lower it to between 130-150W for testing purposes, and overclock the CPU a bit.
You could also overclock the RAM to 15 Gbits (7500Mhz), as the memory modules on most RTX GPUs are downclocked to ~13.5-14.5 Gbits from the factory.

You try to keep the GPU frequency as high as possible (lower temps increase GPU frequency, but power capping reduces it).
There is a balance somewhere between the 130-150W area.
If you drop below 130W, the GPU frequency will drop more, and the heat delta will be small.
If you increase above 150W (137W in my case on an open bench with fans set to 100%), the GPU frequency delta will become smaller; as the increase in watts make it run faster, but the increase in temps will do the opposite, and play catch up.
I prefer to run my 2080 around 1875-1935Mhz.

There is a maximum frequency that your GPU can fold on, and there's a maximum overclock your GPU can work stable with.
When you power cap your GPU, the first one you don't need to worry about (except for very small atom count WUs).

Depending from brand and model, an RTX overclocks between +35 for the lousy ones, +60Mhz for the cheap ones, +90Mhz for average ones (like evga budget models), +120Mhz for good ones, +170Mhz for A-binned models, and some go as high as +210Mhz overclock.
You can only figure out over time which one you have, by adjusting the OC whenever you see a 'fail' or 'is your system overclocked' notification in your log; and lower the OC, until you can run a week without errors.
You'll probably need a notepad or some record of your OC settings.
Also know that when temps increase, OC needs to be lowered (by as much as 10Mhz for a 30F ambient temperature delta)