- Joined
- Jul 24, 2006
Hayes, I'll have to investigate further. All I know is that the GUI threw an exception when I tried to set a slot to 40 cores. It was happy with 32. It defaulted to 28.
-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
T32 EOC WIR 7Sep18: Tiger's blood and winning.
- Thread starter don256us
- Start date
- Joined
- Dec 6, 2010
The client by default will assign 32 cores per slot, to use more than 32 if you have more than 32, you have to go in and set it manually. When I setup version 7.1.52 of the client on my 4P with Win 10 Pro for Workstation, I had to set the core count to 64.
p.s. if anyone wants 7.1.52 just for CPU folding, I can email it to you or set you up to download it from my NAS.
- Joined
- Jul 24, 2006
I must admit I'm not super familiar with Windows licensing anymore, is this "Windows for Workstation" different than a standard 10 Pro N license? Cause that's what I'm running. Procs are dual 2680v4s.
And also, WhEQ: I also have custom PNY drivers as I'm running quad P2000s in this box.
And also, WhEQ: I also have custom PNY drivers as I'm running quad P2000s in this box.
- Joined
- Dec 6, 2010
If FAH / OS only sees 32 cores and there are 4xGPU, then FAH would default to 28 ( 32 less 4 for the GPU). Current version of FAH will not allow a slot to be set for more cores than exist, but extra cpu slots can be created, even though the sum of the cores exceed the total cores in system. I had a host with a 14 core/28 thread cpu and 2 gpus setup for testing in Win10 pro with cpu slots for 24, 16, 12, 10, 8, 6, 4, 3, 2 cores.
- Joined
- Jul 24, 2006
Makes sense. So if I setup an additional 40 core CPU slot, I won't be starving the GPUs, correct? Or 32...whatever works for Win10, don't have time to reinstall in the middle of this contest.
How many CPU cores/threads does windows see, how many GPUs, and what are the GPUs?
General rule of thumb is a thread for each GPU and and extra one or two threads for the OS.
My preference is to run FAH on the AVX cores, which I think would be 40 for your system, but your options may be limited by the OS.
Why not create CPU slots for 28, 24, 16, 12, 8, 6 and 4 CPU. That would allow use of 40 cores with the 28+12 slots or the 24+16 slots. 28 cores with the 24+4 or 16+12 or 16+8+4.
Try starting the GPU slots and observing the TPF (time per frame) and estimated PPD after things stabilize, then start one of the CPU slots and observe if there is any impact on the GPUs. If no impact, start another CPU slot and observe the impact on the GPUs. If no impact, start another CPU slot and so on. If a CPU slot has impact on the GPU, it is possible to pause a CPU slot, reduce the CPUs, and restart the CPU slot. FAH does not allow the CPUs to be increased after the project is assigned to a slot, but usually no problem reducing the CPU count, as long as that number of CPU number is not excluded for that project.
General rule of thumb is a thread for each GPU and and extra one or two threads for the OS.
My preference is to run FAH on the AVX cores, which I think would be 40 for your system, but your options may be limited by the OS.
Why not create CPU slots for 28, 24, 16, 12, 8, 6 and 4 CPU. That would allow use of 40 cores with the 28+12 slots or the 24+16 slots. 28 cores with the 24+4 or 16+12 or 16+8+4.
Try starting the GPU slots and observing the TPF (time per frame) and estimated PPD after things stabilize, then start one of the CPU slots and observe if there is any impact on the GPUs. If no impact, start another CPU slot and observe the impact on the GPUs. If no impact, start another CPU slot and so on. If a CPU slot has impact on the GPU, it is possible to pause a CPU slot, reduce the CPUs, and restart the CPU slot. FAH does not allow the CPUs to be increased after the project is assigned to a slot, but usually no problem reducing the CPU count, as long as that number of CPU number is not excluded for that project.
- Joined
- Jul 24, 2006
I'm with you for the most part but NUMA begins to creep in. These 4 GPUs are spread over 2 CPUs. Lots of bifurcation complications that are neither here nor there unless we were in a serious optimization mode.
edit to say: I can't really baby sit this machine to do the tuning you like to do, Hayes. Its in the office and available but I don't have the wherewithal yet to come in and tune over the weekend, as much as I'd like to. I may also lose access to it at any time. (Cross your fingers, Don!)
- Joined
- Dec 6, 2010
Don't need to reinstall, if your running Win 10 Pro version 1803, the all you need is a new key to upgrade it to Win 10 for Workstation, if the version number is less than 1803, just let it update to 1803. Just goto Microsoft and buy the key. *
Also from what I have read in the forums, is that the off the shelf or OEM Win 10 Pro only sees 2 CPU's and up to 256 cores (if you had 2 CPU's with 128 cores), but if you have 2 CPU's with 24 cores each then max core count is 48.
* Note: This new Win 10 Pro is just for quad CPU MB's.
OP
- Joined
- Jul 17, 2003
- Thread Starter
- #30
I wish you no ill will but if you fall, I will be there. You're getting me by about 200k ppd and I need to be up by about 500k to beat you. Not looking good for me right now. I'm still working on what I can do to mitigate this issue.
Based on your experience with Rosetta on the host, you should have pretty good idea how many cpu/theads can be allocated without impacting the GPUs. I have not dealt with any multi-socket hosts, but perhaps running two CPU-16 slots would use most of the AVX cores and keep the tasks allocated to a single node. Also, Rosetta does not use AVX instructions, so FAH projects using A7 core with AVX will draw more power and generate more heat for same resource allocation.
