Page 1 of 1

Best CPU slot settings for i7 6950x (10c20t)?

Posted: Mon Mar 23, 2020 7:59 am
by treckin
Hows it going! First post and all.

I have a 6950x CPU alongside a couple Pascal Titan X GPUs and I have been folding the past few days.

I have noticed that if I use the default "-1" value for thread count, FAH will spawn an 18 core slot.

I basically get zero work units with the 18 core slot, I have done a few for the 1175x Covid projects, but just a few. Usually idle filling the console with "no WU available for this configuration."

If I make 2 CPU slots, each with 8 threads, I get work units occasionally but generally idle.

If I make a 20 thread cpu slot manually I also get no work units.

Is this because there just arent that many jobs that need x86-64? has everything moved to openCL/GPU?

Also noticing that one of my GPUs is struggling to get WUs despite both cards being identical, on water cooling, etc etc.

Re: Best CPU slot settings for i7 6950x (10c20t)?

Posted: Mon Mar 23, 2020 4:38 pm
by Joe_H
Mostly because there is more demand than the servers can handle. There is work, perhaps not for some specific configuration, but the available servers can only handle so many connections at a time. With all of the new systems looking for work the servers are sending out around 100,000 WUs an hour, over 10 times the rate 2 weeks ago. But the pool of machines looking for work has grown even more. They are working on getting more servers set up, several were added last week, but it does take time.

As for settings, multiples of 2, 3 and sometimes 5 will work. Multiples of prime numbers 7 and larger do not. I know there are some larger CPU projects in the pipeline to be released that will be fine for 20 or 18 threads, but that is a few days off. And in general there is more CPU project work getting set up and prepared for release as server space becomes available.

Re: Best CPU slot settings for i7 6950x (10c20t)?

Posted: Mon Mar 23, 2020 5:33 pm
by treckin
Joe_H wrote:Mostly because there is more demand than the servers can handle. There is work, perhaps not for some specific configuration, but the available servers can only handle so many connections at a time. With all of the new systems looking for work the servers are sending out around 100,000 WUs an hour, over 10 times the rate 2 weeks ago. But the pool of machines looking for work has grown even more. They are working on getting more servers set up, several were added last week, but it does take time.

As for settings, multiples of 2, 3 and sometimes 5 will work. Multiples of prime numbers 7 and larger do not. I know there are some larger CPU projects in the pipeline to be released that will be fine for 20 or 18 threads, but that is a few days off. And in general there is more CPU project work getting set up and prepared for release as server space becomes available.
Thank you for approving and replying!

I got a cpu:18 job today for https://apps.foldingathome.org/project?p=14328. As you point out though, after it completed it went back to perma-searching for WUs.

I wonder if there is not a way for F@H to use the unassigned/underassigned resources from the swarm to help with allocation/post-processing as well as with simulation?

Seems like just another kind of WU, although I realize DC architecture doesn't just write itself.

Best,