Folding Forum

on multi-core systems does the "folding power" slider relate to multiple cores?

I've got a few 12 core and some 8 core machines (Xeons in general as well as some i7's) - but can't seem to see a logical correlation of the 'light, medium, Full" settings...

Full will use all available cores, Medium uses one less, and Light uses half of the cores. An older description of the client setup included more detail on how the number was calculated, and rounded, but that is no longer there.

The choices were made when most machines had 4-6 cores, and HT was not a source for virtual cores in many cases. So Medium would use about 3/4 of a system's CPU resources.

"Available cores" is the number left after reserving a CPU core or thread for potential or actual use by a GPU folding slot. In the case of your OS X machines the installer should not reserve any threads for GPU's, but might under some circumstances.

ADVWedgeman wrote:on multi-core systems does the "folding power" slider relate to multiple cores?

The slider provides only a limited number of choices. For precise control, use the FAHControl program and after allocating one thread per GPU, the remainder of GPUs can be controled precisely. Unfortunately, certain prime numbers don't work well. They'll be excluded automatically, but you can choose to create a second slot to use the remainder of your threads if that's what you choose to do.

Joe_H wrote:Full will use all available cores, Medium uses one less, and Light uses half of the cores. An older description of the client setup included more detail on how the number was calculated, and rounded, but that is no longer there.
The choices were made when most machines had 4-6 cores, and HT was not a source for virtual cores in many cases. So Medium would use about 3/4 of a system's CPU resources.
"Available cores" is the number left after reserving a CPU core or thread for potential or actual use by a GPU folding slot. In the case of your OS X machines the installer should not reserve any threads for GPU's, but might under some circumstances.

VERY helpful..

many thanks!
it makes a lot more sense now - my 12 core units show 24 active threads, but dropping to "Medium" only brought 1/2 of them down a tiny fraction - which is the perfect solution for servers running other minimal data processes.. this gives the server ample power to accomplish daily tasks while still running 20+ threads at peak performance.

There are some other limitations on thread counts due to the underlying Gromacs code used to create the folding cores. In general thread counts using primes 7 or larger or multiples of those prime numbers are not used. There are some projects where thread counts using 5 or its multiples also are not used.

So in practice, with some CPU threads kept free for other processing, you will see WU's running using 18 or 20 threads. On occasion you might see a WU assigned to use fewer if ones that could use more were temporarily unavailable.

On multi cores, you'll need to do the same as with GPUs.
Set 1 theead aside for feeding all CPU threads, and preferably 1 to 2 threads (1 CPU core) to the operating system.
On my 10 core 20 thread Xeon, I only configured 16-17 threads, and still saw 100% CPU usage. Running all available threads is overkill for your system, and if it's a modern CPU, it'll just throttle CPU frequency, and your overall work (though more threads) will end up slower than assigning 2 to 3 threads less for folding.

I'm a total newbie here; I've got 2 multicore Mac Pros (one has 12 cores, the other 8) that are not being used much now. I would like to enable FAH to get the most use out of them. Is it good enough to just leave the slider at full, or would it be better to configure FAHControl in some more advanced way?

When I occasionally have to use my computers for my audio work I will either turn the slider down to "low" or pause the program.

Should be fine to leave the slider at Full. The Mac Pros are less likely to run into thermal regulation down clocking the processors using the HT threads. On my MacBook Pro I find I get basically the same throughput with just using the 2 main CPU threads on its i7 as when I set it to Full and use all 4, and the laptop runs slightly cooler.

You can experiment a bit, run at Full using all 8 or 12, and compare to running on 4 or 6 respectively. You can check the time to do a few percent of a WU each way in the logs and decide what works best for you.

Just one note, a WU downloaded at a specific setting can have it usage of CPU threads decreased and returned to the same value. But not increased to a number higher than set at the time of download.

In addition to what has been stated above, do note that across Windows/Linux/macOS, the priority of CPU folding is either idle or low. Which means that virtually, all other processes would be given preference to use the CPU over folding. Thus, in theory you can leave all CPUs folding (if there's no GPU folding) and continue to work as normal in 99% of cases. There can be some cases where pausing CPU folding would result in a marginal gain but that's up to you