Folding Forum

Hello everyone i don't know where to post this but am having big issue with my 13900k.

If am folding on my 6800XT on full power.
I can't use my cpu for BOINC or F@H. If i use 7 thread or 32 the same thing happen ppd drop from 3-4m to like 15-400k ppd.

I tried playing with high priority without success. Am on Windows 11 so i suspect and Intel Thread Director issue.

Am thinking of selling my 13900k for a ryzen 7950x

You have to get all work done on the P-cores; if E-cores are involved, it'll just slow down work for FAH. That processor has 8 P-cores, which can run 16 threads. Set the number of threads (in FAH Control: go to Configure -> Slots -> cpy slot -> CPUs) to 16. Then use Process Lasso to bind the process FahCore A8 / A9 to the P cores. I don't need Process Lasso myself, so how to do that exactly, I can't tell you.

The CPU core processes are either A8 or A9 (A9 is fairly new, and won't pop up very often yet). Chances are very low that you'd still get A7. The GPU core process is something like FahCore 22 - that can happily run on an E-core if you want to fold with your GPU.

When you increase the number of threads (CPUs), that is only applied to the next job. When you decrease the number of threads, it's applied immediately. There probably is some good design reason for that, but it confused the hell out of me when I started on this, so just mentioning

Optimally, I would not expect (much) more than 400k PPD.

The thing with Intel 12th and 13th gen is that big.LITTLE architecture. Windows 11 is supposed to be able to deal with that, but still seems to have issues with it, and then applications aren't always tuned to that either. FAH at least is not yet.

PaulTV wrote: ↑Wed Dec 28, 2022 9:27 am You have to get all work done on the P-cores; if E-cores are involved, it'll just slow down work for FAH. That processor has 8 P-cores, which can run 16 threads. Set the number of threads (in FAH Control: go to Configure -> Slots -> cpy slot -> CPUs) to 16. Then use Process Lasso to bind the process FahCore A8 / A9 to the P cores. I don't need Process Lasso myself, so how to do that exactly, I can't tell you.

The CPU core processes are either A8 or A9 (A9 is fairly new, and won't pop up very often yet). Chances are very low that you'd still get A7. The GPU core process is something like FahCore 22 - that can happily run on an E-core if you want to fold with your GPU.

When you increase the number of threads (CPUs), that is only applied to the next job. When you decrease the number of threads, it's applied immediately. There probably is some good design reason for that, but it confused the hell out of me when I started on this, so just mentioning

Optimally, I would not expect (much) more than 400k PPD.

Windows reduce priority automaticly with windows 11 on intel processor with big little. if you put the software in the background. Even if i do this with like only 17 thread on BOINC / SIDOCK@ the ppd thank to like 200k for my 6800xt even though task manager report 60% cpu usage.

Process lasso seemed to have no effect on windows 11.

The gpu ppd tank even if i run just 7 thread on F@H CPU.
Wonder if E-Core are just really bad at feeding gpu or something...

That big.LITTLE architecture is one of the reasons for me to go for AMD as well, but that was for a new machine, I didn't have to switch. But at AMD side things can get complicated as well. I have processors with only one compute chiplet, which runs just dandy. If you have 2 compute chiplets (like 79x0x), it's not twice as fast as half the cores on a single chiplet. I've read of at least one other folder who uses process lasso to bind the folding process to one compute die (limited to 12 threads on a 5900x).

Oh, you should always reserve 1 thread on the CPU for feeding the GPU, and personally I also reserve 1 for the OS processes. So my 5600x with 6 cores folds on 10 threads.

PaulTV, have you checked your temperatures? I use HWMonitor. My Intel i9-10850K runs about 70-73C at 4.9GHz with a CoolerMaster AIO MasterLiquid ML360 cooler but runs 84C at 5.1GHz.

You may find some help at Intel. Click around the articles.

Here's a review that may convince you to go to AMD.

I would test it with Ubuntu 22.10 before selling it.

Vester wrote: ↑Wed Dec 28, 2022 1:24 pm PaulTV, have you checked your temperatures? I use HWMonitor. My Intel i9-10850K runs about 70-73C at 4.9GHz with a CoolerMaster AIO MasterLiquid ML360 cooler but runs 84C at 5.1GHz.

You may find some help at Intel. Click around the articles.

Here's a review that may convince you to go to AMD.



I would test it with Ubuntu 22.10 before selling it.

PaulTV wrote: ↑Wed Dec 28, 2022 9:54 am That big.LITTLE architecture is one of the reasons for me to go for AMD as well, but that was for a new machine, I didn't have to switch. But at AMD side things can get complicated as well. I have processors with only one compute chiplet, which runs just dandy. If you have 2 compute chiplets (like 79x0x), it's not twice as fast as half the cores on a single chiplet. I've read of at least one other folder who uses process lasso to bind the folding process to one compute die (limited to 12 threads on a 5900x).

Oh, you should always reserve 1 thread on the CPU for feeding the GPU, and personally I also reserve 1 for the OS processes. So my 5600x with 6 cores folds on 10 threads.

No issue with temp am on a 360mm aio and vrm are Actively cooled. Cpu do throttle but it 1 core in a while for a a second max.
May go amd



Godamn all those cpu are garbage these days haha.


Am thinking of trying to run it in a virtual machine to reduce priority in a more extreme ways. Will report back how it go.
Even running only 7 thread the ppd tank. Seem like something is causing a performance crash.
My 13900k wreck in a game engine benchmark the 7950x so it do have a good use of the ECORE when running a single task.

Here are some numbers from Rosetta@home. Relative Folding@home performance should be about the same for your processor.

No, Rosetta is a single thread application. FAH is multi-threaded and much more optimised.

But yes, big.LITTLE architectures are a bad idea when it comes to FAH ...

Did you try to leave one CPU P-core free to feed to GPU ?

toTOW wrote: ↑Wed Dec 28, 2022 6:38 pm No, Rosetta is a single thread application. FAH is multi-threaded and much more optimised.

But yes, big.LITTLE architectures are a bad idea when it comes to FAH ...

Did you try to leave one CPU P-core free to feed to GPU ?

I left over 18 core once and still had big loss. Even browsing the desktop cause loss if i move alot of thing fast for some reason.

Am thinking the F@H software is the issue.

With hwinfo64 i figured if i drop the software in the background ( VIRTUAL BOX with windows 11 pro workstation with PRIME95 running 16 thread )

The p core are idle and the e core fully loaded almost. BUT i think the ppd still drop really bad but i need to wait to see if it go back up.

Well there nothing to do. Either windows as a big issue or F@H software.

Here you can see with 16 core loaded gpu show about 30% usage then jump to 85% when i stop the vbox using 16 thread.

Even at idle on the desktop F@H Jump up and down from 30-95% for no reason and seem to have a hard time running.

Do you know where the GPU Fahcore (Fahcore_22.exe) runs ?

toTOW wrote: ↑Thu Dec 29, 2022 1:11 pm Do you know where the GPU Fahcore (Fahcore_22.exe) runs ?

You mean the process? I can retry increasing priority to realtime.
Kinda out of trick to try haha. Seem like big little is only good for phone

Yes, is the process running on a P-core or an E-core ? Do you have a tool that can tell you that ?

Disabling E-cores or getting an AMD CPU will save you a lot of headaches

toTOW wrote: ↑Thu Dec 29, 2022 3:37 pm Yes, is the process running on a P-core or an E-core ? Do you have a tool that can tell you that ?

Disabling E-cores or getting an AMD CPU will save you a lot of headaches

Hwinfo64 show per core usage. If i put my CPU intensive program in the taskbar windows force it to use the e-core automaticly. the P-Core where Idle

Will try disabling E core to see or maybe HT when i have some time next weekend.

I regret upgrading from 12700k to 13900k.

Anwyays to force f@h to override windows Priority and be like 100% real time? I mean there should be a ways to force it. Taskmanager and core 22 to realtime change nothing for some reason.

FIXED SOMEWHAT!

Disabling Hyperthreading and putting FAHCORE22 in REALTIME seem to work. With hyperthreading it would not work even in REALTIME. Will Retest to be sure in a few hour/days.

But am using 80% + of my cpu in vbox with p95 and my GPU USAGE stay over 80% Using both PCORE and ECORE.