How can I use F@H for stress testing?

[REMISCS]

Hello,

I am a computer system builder and I would like to become more active with folding, that is why I have the thought to use it as a stress test form on the systems I build. Is there a way to optimize so that I only pick up work units that would take no more than 2hrs to complete? I would try to work this into the last test I use to make sure the computer should be stable at that point.

The lowest power CPU I use would be a laptop i7 9750H 6c/12t and go up to 18+ core CPU. Typically I have a GPU in my systems, lowest end being the RTX 2060 SUPER at this point unless it is a straight office build. Even my laptops have 2070's in them.

I am sorry if there is a search term I should have looked up, I just know I needed to get the ball rolling on this now before I get side tracked again on other projects.

Thanks,
Anthony

[bruce]

Unfortunately, scientific research doesn't break itself easily into 2 hr segments.
Various proteins that are being studied come in a wide variety of complexities and although some effort goes into increasing the time simulated for proteins with relatively few atoms, it's not something that is pursued very rigorously. If you look at https://apps.foldingathome.org/psummary and you'll see variations in atom count and in base credit within the category of FAHCore_a7 (for CPUs) and the category of OpenMM_21 (for GPUs). Moreover, the selections of what to assign are generally not controlled by settings you can make.

My I suggest https://fahbench.github.io/ which will you can manage. It runs the same type of GPU calculations without actually reporting scientific results back to the research departments, which also avoids disrupting the deadlines imposed on actual research assignments.

It should be noted that if a research WU is downloaded, it is costly to the actual research if it is not completed in a timely manner.

[JimboPalmer]

So far as I know, the first CPU Work Unit on each PC only uses one thread. After that it can use as many threads as your CPU supports.* (12 threads in your example) As a result, the first WU is not much of a stress test and is slow.

The first GPU thread is fully enabled. The program for the very latest AMD Navi GPUs is not yet out of beta.

*if you have both a GPU and a CPU, one CPU thread is reserved to 'feed' each GPU. This can be a problem as the CPU slot 'hates' large primes and their multiples.
So a 12 core CPU would reserve one thread for the GPU and have 11 left, but 11 is prime. So only 10 might be in use. Less a stress test that you planned.

[REMISCS]

Thanks for the information. I tried it on a new device and the first workload it took in was about 40mins to perform on an intel i9 9980HK (8c/16t). It used about 70% of the CPU through the process. If all the units tend to pull a similar work load then this could work out well. My personal system sometimes gets 3hr work units on an overclocked 6c/12t intel cpu so I was actually just trying to see if I could narrow it down to 2hr or less only for the client systems.

I should rephrase what I am looking for. I would like for the system to perform a meaningful task during burn in rather than a meaningless synthetic task. By the time I get to that phase of testing, all the actual difficult stress testing has been completed. Not trying to disrupt any projects and my systems are all pretty high end so I figure over time it could add up.

[MeeLee]

There is fah bench, which you can use to benchmark your system, and catch any overclocking issues early on.
From there it's just persistence testing I would guess.

As far as GPUs go, 2 hour WUs, you'll need something like an RTX 2070 super, or greater. But even those sometimes end up with 4 or even 6 hours WUs.

You could, with time, try to see what Wu's are associated with what branch, and eliminate the branch in FAHControl (eg: no Alzheimer's, or only cancer, or whatever gets you smaller Wu's).
But even then, each branch of projects might have fast or slow Wu's.

[bruce]

I appreciate your desire to do meaningful work while you're stress testing. My point is that meaningful work needs to be completed once it is started. (and unfortunately you can't really guarantee it will finish within your 2hr testing window.) Use the FINISH setting and then continue testing until the results are uploaded.

[REMISCS]

Thank you everyone, I will be looking at how I can make this work.

[JimboPalmer]

There is a button in Advanced Control marked Finish, which sets F@H to finish current WUs but not download new ones.

That would allow a clean ending to stress testing.

[HayesK]

Small SSD or HD with linux client. Connect the linux drive to the test machine using the internal sata or an external USB adapter. Sounds like your default setup would have nvidia driver with a CPU slot and GPU slot. You could have copies of FAH cfg files with various scenarios, multi boot or several different test drives.

Be sure to set the client to "finish", allow the work in progress to complete the upload, then pause client before shutting down. Otherwise,the client will download new work when rebooted, before you have time to make any desired cfg changes for the next test. Disabling the internet connection prevents new work too.

[REMISCS]

@HayesK

Ahhh I like that, so basically just keeping the OS and passing the install over to another machine and letting another machine finish it. That is clever!

[HayesK]

Set up linux and gpu driver on a dedicated disk.
install FAH and configure. (initial cpu work unit will run on a single thread, unless a FAH cfg file is already existing in the correct location).
Then use the portable disk for testing on the other machines.

The work units need to assigned, downloaded, run, completed, and results returned from the same hardware. If a hardware change is detected by the client software while a work unit is being processed, the work unit will fail if no compatible hardware is found. Pretty much needs to be identical. Best to finish all work in progress prior to making any hardware or software changes.

[MeeLee]

Do be aware that if you change systems, and the new system has a different type GPU, or less CPU cores/threads than the first one, the WU will be dumped.
For a transfer to be possible, you'll have to transfer between nearly identical systems (most RTX GPUs are recognized if they are the same type, eg: all 2060s, or 2070s; regardless of what brand. However a WU for a higher end GPU (eg a 2070) will not resume on a lower GPU, and a WU of a lower GPU might not continue on a higher GPU; especially not if the performance between the two cards is too great).

[foldy]

the work unit will fail if no compatible hardware is found. Pretty much needs to be identical.

Identical means a CPU with same core count for FAH CPU slot and a GPU from same vendor nvidia vs. AMD?

[toTOW]

Yes.

[HayesK]

A CPU wu will run on the same or less thread on intel CPU. Suspect that also applies to AMD CPU. Do not know if CPU wu started on intel will run on AMD.

A GPU wu will run on the same GPU model made by different suppliers, as long as FAH software identifies them as same.

Best to finish all work in progress before making any hardware or software changes.