Not Folding At Full Power On Restart?

[geokilla]

I turn off my computer at night. Often times when I turn on the computer, I find the GPU doesn't fold at full power until I manually apply my overclock I have set in EVGA Precision X1.I know it's not folding at full power because TPF is twice as much and the GPU isn't under load as the fan is spinning at only 20% compared to the usual 50% when folding. Is this a driver error? I'm on 512.77.

[aetch]

There's a lot to unpack there.

First, we recommend you don't overclock.

OK, caveat out of the way.

Overclocking Settings
The GPU does not save your overclocking settings, Precision X is doing that. Precision X should have a toggle to allow it to start when Windows starts. Try saving the overclock as the default profile, so it is the one that loads up when Precision X is loaded.

Full Power?
Your GPU will operate at full power when you load it.
There are tools to measure this.

Windows Task Manager -> Perfomance Tab -> GPU
You will see 4off graphs arranged in a 2x2 grid.
Depending on whether "Hardware-accelerated GPU scheduling is enabled you're looking for the following:-
Enabled - performance will show on graphs selected for 3D
Disabled - you need to change the graph dropdown to either CUDA or Compute_0

Precision X should have a graph and/or indicator to show GPU usage (I assume Precision X has this, MSI Afterburner does which is the one that I use)

CPUID HWMonitor also has an indicator for GPU utilization (I just realised I need to update my HWMonitor as it's not reporting this).

Precision X and HWMonitor also report temperature, when your GPU is loaded you will see it get hotter.

Fan Speed
The fan is programmed to maintain a certain temperature, roughly 70-80C. If it's only spinning at 20% then it means you have a very efficient heatsink and the fan does not have to spin very much to move the heat. When the GPU is sitting idle the fans may stop. If you want the fan to spin at 50% all the time, regardless of load, you can set it in Precision X.

TPF (time per fold)
There are many things which can affect this and they are usually red herrings:-
*). it's a work unit for a project your computer has not processed before, the initial ETA is set to the timeout which skews the tpf, it normally takes 2-3% for this to normalise and become more realistic.
*). the project itself. Each project is simulating different types of proteins for different lengths of time. I've seen work units with TPFs of seconds and other with TPFs of up to 30 mins.
*). your computer installs update or does other things in the background. To be fair, the silent updates can tank the performance of FAH and is worth a watch. The normal remedy is to reboot the computer.
The closest you'll get is comparing work units from the same project, even then you'll get some variation.

Driver Error
This is possible but it's more likely to manifest itself as errors or system crashes. Try some of the older drivers if you think this is the issue, I'm still using 511.79.

[ManiacJoe]

I am seeing similar problems with the lack of performance after updating to NVidia's latest studio driver.
512.x seemed fine, 516.59 seems to get virtually no work done.

Edit: On follow up, something caused a reboot of my win10 computer two nights ago. Since then the 516 driver has been cranking out work units as normal.

[Jimboc]

I am seeing similar problems with the lack of performance after updating to NVidia's latest studio driver.
512.x seemed fine, 516.59 seems to get virtually no work done.

Edit: On follow up, something caused a reboot of my win10 computer two nights ago. Since then the 516 driver has been cranking out work units as normal.

Hi ManiacJoe,

Thanks for your update and good news. Can you confirm which 516 driver you are using, please? 516.40 or 516.59? I'm on the older 512.96 driver and wanted to check this before I update.

Thanks again.

[ManiacJoe]

Sorry for the late response.
I am currently on the 516.94 studio drive. Probably on the 516.93 driver at the time of the original message.
Seems to hum right along.

Reading messages and watching my system, I think the original (and current) problem is that some of the WU are huge and seemingly are not correctly marked for time lengths. For example, the WU 17627 has a base credit of 500000 with a timeout of 2 days. My box will not complete it by then even with the driver seemingly running as expected. I had a similar thing happen with a 200000 credit WU.

[BobWilliams757]

Sorry for the late response.
I am currently on the 516.94 studio drive. Probably on the 516.93 driver at the time of the original message.
Seems to hum right along.

Reading messages and watching my system, I think the original (and current) problem is that some of the WU are huge and seemingly are not correctly marked for time lengths. For example, the WU 17627 has a base credit of 500000 with a timeout of 2 days. My box will not complete it by then even with the driver seemingly running as expected. I had a similar thing happen with a 200000 credit WU.

Is the problem that the work unit won't finish within allocated time, or that it won't finish within allocated time if you turn your computer off at night?

By my guesstimate of what has been reported your computer should finish that in no more than 15 to 18 hours. Even if you turn the system off at night it should be sufficient time. Or you could simply let it run into the night and use the finish command so it won't pick up another work unit.

It's not on common for work unit time per frame and points payouts to vary quite a bit with different projects. That is normal behavior and somewhat expected.

But if you are dumping or otherwise not finishing the work units simply due to the points payout, please don't. It does nothing other than slow down the science since it won't be reassigned until the deadline is reached.