Folding Forum

[This is just my 2 cents, I am not 'official']

You have manually messed with files that are supposed to be updated by the software.
If it was me, I would uninstall F@H, and if asked, I would delete everything.
Then reboot and reinstall it all over again, doing the express install and not changing anything.

Best of luck.

[End of paranoid advice]

I can relate if something else happens during the update. Will try that next, but would be nice to confirm if the GPUs are on the master (server) before going further.
I'm using the official link; http://fah.stanford.edu/file-releases/public/GPUs.txt

Since the only thing that was edited was the one text file and that text file doesn't exist in a fresh install, just deleting the file and restarting the client causes it to download a fresh copy in the same way a fresh install would. I'm pretty sure you'd have to restart the client to get it to pick up the new GPUs anyway. A reinstall shouldn't be necessary unless something unrelated to this is going on.

Edit: deleted the GPUs.txt file from my gimpy client (since there's no GPU work for it anyway) and let the client download a fresh copy. The copy the client downloads still doesn't have the 2070 in it so that would *seem* to suggest the copy the server has is older.

Edit 2: interesting that I just deleted GPUs.txt while the client was still running and it pulled down a new copy a minute or two later without a restart. Convenient. I assume that this will (when the issues are resolved) be the easiest and least disruptive way to force an update. Also, this most recent update downloaded a copy of GPUs.txt with the new (incorrect) entries for the TU106 cards. Every refresh before this one, I got the old version without them. I wonder if that means the file is propagating across distributed servers or what. Hm.

The issues with adding GPUs are fixed and I confirm that we're supporting both variants of the 2070 :

1f02 TU106 [GeForce RTX 2070]
1f07 TU106 [GeForce RTX 2070 Rev. A]

This description isn't quite right. I have the 1f02 and it is a Rev A as well.

Edit: It's entirely possible this is a cosmetic name and has no bearing on anything important. If so, I'm fine with that. I suspect it will cause a number of confused/concerned inquiries but as long as the folding is proceeding as it should then that's all that really matters.

Edit 2: Deleting and letting the GPUs.txt file update automatically got another old copy so I grabbed the new file from the link Ryan posted above and saved over the existing copy. After that, I bounced the client and within a few minutes had started folding at over one million points per day. In the past, my NVidia GPU only showed 100% utilization on Compute_0 and nothing for any other aspect of the card. Now, I see fluctuating usage of around 30% for 3D, a steady 5% utilization on Copy, and Compute_0 hovers pretty stable at 90% but won't go over.

I'll let it run for a day or so and then try with the CPU manually set to eight threads instead of allowing it to go to the default of 10. We'll see if that impacts performance. Is there anything else I should do to test/improve performance/utilization?

I can confirm working 2070 here as well now. Thank you!

1.09m PPD on 11726 In windows 10, FAH 7.5.1, 8xPCIe @ 94% utilization, 2nd GPU is 100% utilization (doing other tasks), no CPU utilization. I think it can hit 1080ti numbers in Linux. Will test that next.

doune wrote:This description isn't quite right. I have the 1f02 and it is a Rev A as well.

It's entirely possible this is a cosmetic name and has no bearing on anything important. If so, I'm fine with that. I suspect it will cause a number of confused/concerned inquiries but as long as the folding is proceeding as it should then that's all that really matters.

My understanding is that production chips are binned into -A (which supports manufacturer OC systems) and non-A (which works on the Founders Edition.

Deleting and letting the GPUs.txt file update automatically got another old copy so I...

You probably forgot to clear your internet cache.

I'll let it run for a day or so and then try with the CPU manually set to eight threads instead of allowing it to go to the default of 10. We'll see if that impacts performance. Is there anything else I should do to test/improve performance/utilization?

Nine should be a good number, too

From what I can tell, all of them (so far) are 400-A1 or 400A-A1. Both mine and Ryan's are A-A1 even though his is 1f07 and mine is 1f02. I wouldn't hazard a guess at what that means. I suspect it's not meaningful to folding.

I actually switched to 9 (since I remembered reading that it just can't be a prime) and saw things improve in general.

I only have one remaining question. Is there a way to use more than one vCPU to feed the GPU? Since I have 12 vCPUs, seeing the one feeding the GPU pegged at 8.3% and the GPU itself capped at 90% makes me think that committing more CPU to it could increase GPU utilization to 100%, which would be a much better return on CPU resources than the CPU itself doing the folding.

it's the 400- vs. 400A- that matters ... not the -A1.

bruce wrote:it's the 400- vs. 400A- that matters ... not the -A1.

Yes, that's why I pointed out that both Ryan's 1f07 and my 1f02 are 400A.

In the GPUs.txt file, the 1f07 is listed as Rev. A and the 1f02 is not even though they are both 400A.

I'm assuming the stuff in brackets is just a cosmetic, friendly name so the distinction in the GPUs.txt file is irrelevant.

And so is the 1e87 ???

If that's so, where is the -400-a1 ?

Sure, it's all cosmetic, but I don't trust that we have accurate information. Everything after the 4th : is just for the human reader.