Folding Forum

I've read here a few times that FAH runs faster on Linux than on Windows. On the other hand, Windows 10 offers the possibility to run Linux in a windows, without reboot. Would it be a good idea to install WSL, then a Linux distro, and to run FAH in such a setup? Did someone already try?

The reason that current FahCores produce higher PPD in Linux is because there's less overhead in Linux (this may or may not change in the next version of FahCore). From my understanding, the implementation of WSL 2.0 uses a lightweight virtual machine via sub-features of Hyper-V. Thus, the question is how is will the following compare:
1) Native Windows
2) Native Linux
3) WSL 2.0

I am curious to see the results if anyone is keen to do it.

PantherX wrote:The reason that current FahCores produce higher PPD in Linux is because there's less overhead in Linux (this may or may not change in the next version of FahCore). From my understanding, the implementation of WSL 2.0 uses a lightweight virtual machine via sub-features of Hyper-V. Thus, the question is how is will the following compare:
1) Native Windows
2) Native Linux
3) WSL 2.0

I am curious to see the results if anyone is keen to do it.

Or 4) Linux VM under Windows host. If I remember correctly my own experiment using VirtualBox many years ago, Linux VM gave a 10% PPD increase than just native Windows. I'll probably try WSL when I have time

That would entail:
1) Install Win 10 2004 (with WSL 2)
2) Install WSL 2 and a Linux distro there
3) Install the same distro on a partition
4) Run FAH on NW, NL, and WSL
5) Compare the results

What would be a good comparison, though?
Note down for each WU a) Base Credit b) Estimated Credit? Or a) Base Credit b) effective Credit?
Then search for the same Base Credit in NW, NL and WSL and see if there is a significant and constant difference with the corresponding Estimated/Effective Credits?

What I would do is to note down the PRCG and the TPF average for that WU in each of the environment. If you get WUs from the same Project, that's great and it would be even better if the system was dedicated to folding during this experiment.

EDIT: Differences in PPD can be calculated from the various TPF as long as we have the PRCG of the WU.

Where do I find the TPF Average?
I prepared this: https://docs.google.com/spreadsheets/d/ ... edit#gid=0
Would that be ok?

I can't access the sheet without a Google account Could you maybe make it read-only for public and to modify require sign-in?

The Average TPF would be, assuming that you have done a fresh install in each instance, would be the TPF that is shown in FAHControl as it averages out.

done. The sheet should be accessible now, sorry!

The idea would be to run FAH in Windows for a number of days, say 7, then in Linux Native, and in Linux via WSL.

That's amazing as you have a CPU, Nvidia GPU and AMD GPU... looking forward to the results! If you have a bit of time, and would like something else, see if you can do the fourth option mentioned by iceman1992 which is a Linux VM in VirtualBox on Windows

OK! Yes, why not. I'll add VirtualBox, then.

Is there a difference between WSL1 and WSL2? I don't have Windows 10 2004 yet

Yes, MS is promising it will be faster: https://www.youtube.com/watch?v=MrZolfGm8Zk
You can get 2004 as "MS Insider".

NOTE: I'll let the system run 24/7, but I'll note down only the WUs that I do see during the day. It should be enough for the comparison and I can thus spare me the search in the logs. Plus: I don't see the TPF in the logs.

ajm wrote:...I don't see the TPF in the logs.

The TPF can be calculated manually by looking at the log and calculating the time taken for 1% to complete. 1% is one frame.

There is a Bonus Calculator that someone provided - http://www.linuxforge.net/bonuscalc2.php. I use an average over 10 frames as some projects and folding cores will vary from frame to frame. The author posted info about it in the forum section for 3rd Party Contributed Software a while ago, at least several years back.

Thank you Joe_H! But the TPF is a rather fleeting apparition. I have to be in front of the computer during the folding to see it, and it changes constantly. Whereas the points appear in the log and on https://apps.foldingathome.org/cpu
As the machine run 24/7, I would prefer using a value that I can more easily ascertain after the fact. Is the TPF that crucial for such a test?