Intel Core 3 100U info

[darkf3d3]

Hi everyone,
I recently started using folding@home on a pc with Intel Core 3 100U cpu.
This cpu have 6 cores an 8 threads:
- 2 Cores, 4 Threads (Performance Cores)
- 4 Cores, 4 Threads (Efficient Cores)
with Intel Graphics.

Into the machine setting i can chose 1 to 8 cpu + the gpu.
First of all I noticed that despite I enabled gpu, from the main page i see GPUs set to 0. This mean that the gpu it's not used?
Then I can't find a way to decide which core to use. For example there's no way to set it to use 4 efficient cores only?

[Joe_H]

The Core 3 100U is a fairly recent mobile processor. As best as I can find with a quick search, it has an iGPU of a generation that Intel did not include support in hardware for double precision calculations. So it should show up as unsupported, but I don't have the specific PCI ID number for the iGPU to check.

For the CPU cores the client is only able to detect how many threads are supported by the processor. Use of which cores is left to the OS and its process scheduler. Third party software can be used to assign usage to threads associated with specific processor cores. In practice you usually want to be using the performance cores for folding. I am not certain how well folding will work on just the E-cores alone for example. For the P-cores folding usually works better just using one thread per core, contention for the shared FPU in the core can slow down processing.

[darkf3d3]

Thanks for the information.
I was able to find a windows setting that allows you to assign only certain cpu cores to a process (cpu affinity from task manager).
So I set folding@home to use all 8 cores and then chose the cores from the task manager.
Over the next few days, I will do a bit of testing to see which is the best setting.
This is a pc I use for work, but since I do basic use, I want to run folding@home in the background. Previously I had noticed that even if I assigned only 2 or 3 cores out of 8, the cpu always ran at 100%.

[muziqaz]

Your task manager affinity setting will reset when you finish a WU. Use Process Lasso to tie specific processes (fahcore_a8 and fahcore_a9) to specific cores

[darkf3d3]

Thank you, I thought affinity settings would be retained until power off or restart.
I'll try process lasso.

[Joe_H]

The Task Manager affinity settings are on the process. The client starts a new fahcore_a8 or fahcore_a9 process each time it starts processing a CPU WU. This also applies each time processing is paused and then resumed.

[darkf3d3]

Do you know If folding@home assigns projects according to the number of available cpus?
Let me explain... Previously, when I was using it limited to e.g. 4 cpus, I received projects that took an average of 6/9 hours.
Whereas now that I have activated all 8 cpus and then limited the cores using process lasso. I received a project that will take about 30 hours.
Could it be that I am getting more demanding projects because I have increased the number of cpus in the folding@home settings?

[muziqaz]

It is just coincidence, but CPU WU are assigned per core count.

[calxalot]

Yes, some work is constrained by cpu min or max counts. For example CPUs >= 8

[arisu]

Yes, some work is constrained by cpu min or max counts. For example CPUs >= 8

I noticed that too. I turned off SMT since it slows down my workload, but it's caused me to get WUs that take just 40 minutes to complete. I hope I'm not "stealing" easy work that are intended for older systems. It feels like cheating because I have higher PPD with these "easy" projects.

Perhaps fah-client should report physical core counts instead of logical core counts to the AS because it is a more accurate indicator of performance (or maximum logical core count whether or not SMT is on). Would that be worth a new Github issue?

[muziqaz]

You are thinking about it too much. You set up your PC the way you want it to run best. What FAHClient thinks of it, is not your problem or worry
Please don't create an issue out of nothing
P.S. AMD SMT gives performance boost

[arisu]

You are thinking about it too much. You set up your PC the way you want it to run best. What FAHClient thinks of it, is not your problem or worry
Please don't create an issue out of nothing
P.S. AMD SMT gives performance boost

I overthink everything.

AMD SMT reduces performance when the iGPU is in use, at least on my system. Pinning to logical cores 1-7 and 9-15 to leave the first physical core (0 and 8) free improves the situation somewhat, but not nearly as much as disabling SMT entirely. It's a difference between 50s TPF and 40s TPF (measured on the same work unit over 10 frames, restored from checkpoint to the same spot each time, no other competing activity, all IRQs redirected to core 0). I suspect it has to do with how the amdgpu driver handles DPM (dynamic power management). It seems like it's making suboptimal choices when SMT is on, leaving the shader clock high even when the graphics pipeline is being starved to under 70% (and idling shaders at a high clock rate is particularly inefficient).

Many iGPUs are terrible but this one is a species 4 so it gets decent WUs, which is why I leave it on. The CPU's PPD decrease (roughly 300k down to 200k) is more than made up for by the additional 250k PPD that this particular iGPU adds.

But if getting projects that I suppose were "intended" for 4c/8t systems rather than projects "intended" for slightly more powerful 8c/16t systems doesn't cause any harm, then there's no problem! I just didn't want it to feel like cheating.

[calxalot]

I think it’s fine.
Only time I felt guilty was quite a while ago when I thought my M1 mini might be stealing work meant for raspberry pis.

[muziqaz]

IGPU grabs some of the awaylable CPU TDP budget, plus brings it's own heat to the table. This is not SMT fault technically thinking. It is the same if we talk about the computer case without good air pressure. If CPU alone is loaded, you get specific performance from it. Add GPU workload to it, and CPU performance will drop because inside of the slcase you have much increased temperatures, which limit CPU boost.
CPU alone with SMT overall brings some performance increase compared to same CPU without SMT