Folding Forum

Hi I am new to this and I have 2 x E5-2699 v3 and 1 x rtx 3070. it seems that only one of my CPU is getting used which is dissapointing as I have dedicated this server to this cause. Am i doing something wrong ?

thx

It’s common for the very first work unit to only use one cpu. Future work should use the number you set, or a number the client chooses based on power level.

What OS are you using?

calxalot wrote: ↑Wed Feb 22, 2023 11:44 pm It’s common for the very first work unit to only use one cpu. Future work should use the number you set, or a number the client chooses based on power level.

Hi thx for the response, I am using windows 10 and the web client. I have the interface set for full power and when I am working ?

Sounds good. If the GPU shows as disabled, you may need to install drivers from nvidia.

calxalot wrote: ↑Thu Feb 23, 2023 8:48 am Sounds good. If the GPU shows as disabled, you may need to install drivers from nvidia.

GPU is working fine, its the cpu that doesnt seem right, task manager shows only 1 NUMA * cpu* being used. I have 2 cpu with 18cores each and a total of 36 cores and 72 threads. it would seem that the app is using 32 threads instead of 32/36 cores?

The maximum threads used for a WU on Windows is 32.
I believe you can make a second CPU slot using FAHControl (Advanced Control in sys tray menu).

I think some people use Process Lasso to set affinity of each FahCore to its own cpu.

calxalot wrote: ↑Thu Feb 23, 2023 10:59 am The maximum threads used for a WU on Windows is 32.
I believe you can make a second CPU slot using FAHControl (Advanced Control in sys tray menu).

I think some people use Process Lasso to set affinity of each FahCore to its own cpu.

Thx for your support, I think its a limitation perhaps with windows.

with 64 views and only your support I have probably come as far as I can go.

Thx again

With recent versions of Windows it is less of an issue, but would require going back through the builds of the client and CPU folding core to make certain all of the correct options were selected during coding and when compiling the executables. There may be some additional OS settings that need to be made as well.

Part of this a holdover from earlier versions of Windows where specific license levels needed to be purchased to use more than a specific number of CPU threads by a single executable. So testing on multi-processor, high thread count systems has been limited. It has not been a problem in general for Linux or macOS as they never tied thread usage to license levels. About the only problem I recall has been setting appropriate NUMA values.

Besides possible license restrictions, the api used was only capable of seeing 32 logical cpus.

v8 has supposedly fixed this, but I haven’t heard of it being tested. But then, WU assignments in v8 have min and max cpus allowed. I think it’s effectively 32 for most WUs. V8 should automatically run multiple WUs to use all resources made available to the client.