Folding Forum

Hello all. I have a AMD 3900x cpu with 12/24 count.

I also fold on a GTX 1070 and a RTX 2080Super. Should I leave the CPU @ -1 and let the system choose or should I put in fixed number of threads, and if so, how many is optimal so the GPU get's is juice of threads?

F@H by default reserves a CPU for each Graphics card to feed the PCIE bus with data, so you have 22 of 24 for folding

If you decide to over ride the -1, there is a quirk of F@H on CPUs. It hates Large prime numbers and multiples of them. 7 is always large. 5 is rarely large, 3 is never large.

So 2. 3. 4. 6. 8, 9, 12, 16, 18 are always safe. 5, 10, 15 and 20 are usually safe.

One haair brained but safe solution would be one cpu Slot of 18 and one cpu slot of 4

See this for a bit of explanation: viewtopic.php?f=16&t=32752

Your GPUs will need 1 thread each, so you're down to 22 usable for folding at that point. I'd recommend creating one CPU slot with 8 and one CPU slot with 12.

I could certainly be wrong too.

So what's the best course of action with a 4 core/8 thread CPU. Mine seem to select 7 threads on their own, but that sounds like a bad idea. Is six better and just leave one idle? 2 slots with 3/4? 6/1?

Lockheed_Tvr wrote:So what's the best course of action with a 4 core/8 thread CPU. Mine seem to select 7 threads on their own, but that sounds like a bad idea. Is six better and just leave one idle? 2 slots with 3/4? 6/1?

Just leave it at the default setting unless overheating or you are on Windows and need to leave a thread free for the GPU.
Setting are on Advanced Control> Configure> Slots> CPU

kiore wrote:

Lockheed_Tvr wrote:So what's the best course of action with a 4 core/8 thread CPU. Mine seem to select 7 threads on their own, but that sounds like a bad idea. Is six better and just leave one idle? 2 slots with 3/4? 6/1?

Just leave it at the default setting unless overheating or you are on Windows and need to leave a thread free for the GPU.
Setting are on Advanced Control> Configure> Slots> CPU

I'm not overheating but I am on Windows. Is there a better configuration for Windows?

No. with 8 and a GPU, 6 is as good as you will do.

I have a similar question.

I have a Lenovo Legion Y540 notebook with an Intel i7-9750H (6 core / 12 threads), nvidia RTX 2060 Mobile and 32GB of RAM (Windows 10 1909).

If I leave the Folding@Home client at it's default settings (-1 for CPU and GPU) it will get 11 cores for the CPU (the GPU is being properly detected and used).

Since 11 is a prime, it would be a better idea to force a CPU slot with a 10 thread count?

I think I am getting a really low PPD (around 80k). GPU is always 90%+ used.

Thanks in advance.

RMCholewa wrote:I think I am getting a really low PPD (around 80k). GPU is always 90%+ used..

Do you have a passkey set : https://foldingathome.org/support/faq/points/passkey/ ?

No, I asked for one this morning (6 hours ago or so) but received no email so far...

Did you look in your spam folder ?

toTOW wrote:Did you look in your spam folder ?

Yes, nothing there...

Maybe this information is relevant: I am receiving forum post alerts without problems.

JimboPalmer wrote:One haair brained but safe solution would be one cpu Slot of 18 and one cpu slot of 4

18 is a probably a very good idea because it factors to 2x2x3. The smaller the factors the better because the cells are larger.

Ian Cutress at AnandTech has been producing great charts showing per core power consumption for high core count CPUs.

Source: https://www.anandtech.com/show/14605/th ... the-bar/19

All cores are fighting for a piece of the power budget. One slot with 18 threads, rather than 18 + 4, may actually produce more points.
The idle cores/threads will donate their budget to the active cores, which will clock higher.

Theoretically, 18 cores active @ 3300 MHz = 59400 vs. 22 cores active @ 3000 MHz = 66000 would mean the 22 core configuration is capable of more FLOPS.
Will the higher QRB on the 18 configuration compensate compared to more science done by 18 + 4? For the project as a whole right now, having one less slot vying for a limited supply of work might be beneficial as well.