Ryzen CPU folding power isn't changing PPD?

[Demmers]

Hi everyone,

I run FAH on just a CPU (AMD Ryzen 3400G, 4 core/8 thread, clocked at 4Ghz), no GPU. I've been folding on this PC (Linux) build since last November, most of the time with the CPU set manually to fold on 4 CPU threads. This gives me an average PPD of 50,000. Now from my understanding, and please correct me if i'm wrong, but doesn't 4 threads equal 2 cores dedicated to FAH? Reason I ask is because over the last couple of days, i've decided to fold at full power (8 threads), but have noticed there is no increase to the PPD, still averages 50,000. I wouldn't actually expect it to double, can't believe it's that simple, but I would have thought that utilising all 4 cores would "speed up" the process, thereby increasing the PPD?
I've already tried searching for answers to this on here but couldn't find anything this particular. Interestingly, I came across this website https://greenfoldingathome.com/2020/05/ ... f-threads/ which has a graph showing the change in PPD when increasing the amount of used threads for a Ryzen 3950x. Coincidentally, 4 threads matches around the 50,000 mark, 8 threads around 140,000! So maybe I should be expecting a doubling of performance?!
If it is that simple that the PPD should be increasing, i'm wondering if it is due to a BIOS setting related to CPU power saving? I have Performance Boost turned off (due to manually overclocking the chip), everything else is left as is.
Any feedback on this would be much appreciated. Thanks!

[Neil-B]

If you change thread count upwards mid WU it wont take effect until the next WU is downloaded ...You should see some increase but not double by any means ... Rysen 3950x has 16 cores 32 threads so upstep from 4 to 8 is still in true core territory and the PPD is over double probably due to significant impact from QRB.

[Demmers]

Hi Neil, thanks for that. I forgot to add that I am aware about when the changes take place. I've had multiple WU's since making the change. It's all good, i'm happy to let it continue at max power. FYI, even running at 4 threads, i've noticed i've worked on the largest WU's there are for CPU's, where the atom counts are in the 400,000's. I don't know if running at full power means I get sent more of them? Either way, this is all just an observation and wasn't sure what to expect. Thanks again.

[Neil-B]

Have you checked your temps/clocks? ... I am wondering if by increasing the threads which normally increases temps you are getting a bit less boost or some throttling that is nullifying the gain in ppd you might expect to see?

[bruce]

A lot depends on whether you're running close to the optimum temperature.

Adding more threads, increases the temperature which, if you were already throttling, lowers the processing rate to throttle at about the same power level. This may not apply to you, but if it does, adding more fans / better heatsinks might do more for you that you think.

[Demmers]

Fair points. Yes I have been monitoring both the CPU speed and temps. My PC is a custom build from scratch last October (so all last year components), and have since added better cooling with the use of a Noctua tower cooler and exhaust fan. My case comes with a 200mm intake fan. The tower/exhaust fans are PWM controlled, and run at max at 60 C. As i'm typing, FAH is running all 8 threads. CPU is averaging 3990Mhz across all threads (remember i've clocked it in BIOS to 4Ghz), temp is at 65 C. For those interested, Vcore is set to 1.325V. I would have thought 65 C could be a lot worse for CPU's in general, so i've been pretty happy with that.
When the CPU has been running at 4 threads, it was still able to average 4Ghz, temp between 58-62. But either way, once the PC has been left alone with only FAH running, it's been in the high 40K/low 50's, at both 4 and 8 thread power, hence me scratching my head lol.
Also to add, RAM is at 16GB - 3200Mhz, 650W PSU (for when I eventually add a next-gen GPU later this year). Motherboard - Gigabyte B450 mini-ITX. OS is Linux Mint.

[MeeLee]

You could possibly try to fold on the IGP of the Ryzen?
It's a bit slow, but it is more efficient than the CPU.

[Demmers]

You could possibly try to fold on the IGP of the Ryzen?
It's a bit slow, but it is more efficient than the CPU.

Tried getting this to work this morning out of interest, but turns out to be impossible: The amdgpu driver is installed as part of the kernel, but AMD don't add opencl as part of it, and opencl is what FAH uses for graphics cards. To install opencl, I would need to install the amdgpu-pro driver (this includes opencl), however, AMD don't have an up-to-date version for Ubuntu 20.04, which my latest version of Mint uses. Tried installing the previous driver version, but this failed due to the OS's not matching. Oh well.
I've also got my CPU down to stock speed of 3.7Ghz, spent long enough at 4. Would rather have less PPD than to use more electricity than necessary.
Going back to original point of topic, I've read that some WU's can make the PPD count fluctuate, so i'll continue using all 8 threads for at least the next week to get a good average, then compare with 4 threads again.

[MeeLee]

You could possibly try to fold on the IGP of the Ryzen?
It's a bit slow, but it is more efficient than the CPU.

Tried getting this to work this morning out of interest, but turns out to be impossible: The amdgpu driver is installed as part of the kernel, but AMD don't add opencl as part of it, and opencl is what FAH uses for graphics cards. To install opencl, I would need to install the amdgpu-pro driver (this includes opencl), however, AMD don't have an up-to-date version for Ubuntu 20.04, which my latest version of Mint uses. Tried installing the previous driver version, but this failed due to the OS's not matching. Oh well.
I've also got my CPU down to stock speed of 3.7Ghz, spent long enough at 4. Would rather have less PPD than to use more electricity than necessary.
Going back to original point of topic, I've read that some WU's can make the PPD count fluctuate, so i'll continue using all 8 threads for at least the next week to get a good average, then compare with 4 threads again.

You can always install the opencl developers drivers,
sudo apt install ocl-icd-opencl-dev

[Demmers]

You could possibly try to fold on the IGP of the Ryzen?
It's a bit slow, but it is more efficient than the CPU.

Tried getting this to work this morning out of interest, but turns out to be impossible: The amdgpu driver is installed as part of the kernel, but AMD don't add opencl as part of it, and opencl is what FAH uses for graphics cards. To install opencl, I would need to install the amdgpu-pro driver (this includes opencl), however, AMD don't have an up-to-date version for Ubuntu 20.04, which my latest version of Mint uses. Tried installing the previous driver version, but this failed due to the OS's not matching. Oh well.
I've also got my CPU down to stock speed of 3.7Ghz, spent long enough at 4. Would rather have less PPD than to use more electricity than necessary.
Going back to original point of topic, I've read that some WU's can make the PPD count fluctuate, so i'll continue using all 8 threads for at least the next week to get a good average, then compare with 4 threads again.

You can always install the opencl developers drivers,
sudo apt install ocl-icd-opencl-dev

I've already installed that, due to the research I did this morning, but "clinfo" in Terminal displays "Number of platforms - 0", hence why I thought the extra driver needed installing.
I did try to run the gpu: Added an extra slot and from what I read, changed "opencl-index" to 0. Left gpu-index at -1. I also set gpu to True in Expert. First time, it took ages for it to eventually download a WU, but would then fail to start it. I can't remember the exact error, but it mentioned opencl. I checked the GPU.txt file in fahclient, the chips GPU is listed: [Radeon RX Vega gfx902].