Folding Forum

ipkh wrote:That would be correct only if every distributed WU came back with an hour. Most take at least 2 hours to complete so there will be a delay in the return stats.

Only in the short run. If there is a sustained period of X WUs being assigned per hour, we would expect to see the same number coming back, on average.

I have no clue - speculation on my part - but if 100k WUs issued, of which 30k were reissues after timeout reached for various reasons (which might not be unfeasible given the last month) but because of the way issues are counted amassed towards the total, might the returns only be counting multiple returns of the same WU as one return having been collated differently - the chart may simply have been created to show completed WUs and not for comparison against the server stats?

Not asking as such just putting forward a hypothesis that might explain this seeming peculiarity

I am experiencing prolonged (hours-long) gaps in between WUs much less than I used to. Many thanks to the good people working hard to get these servers up and running!

Everytime I think things are getting better ... well ... they aren't ...

indeed, i've got all my cards consistently getting work at present but my awarded points are at best 50% or so of what they should be.

The stats are slow again, but they will catch up eventually. You can see here: https://apps.foldingathome.org/bonus how many WUs (but not the corresponding points, though) you have effectively delivered, in real time.
I just checked: there a 23 WUs gap for me right now.

EDIT: Well, it's not real time any longer, either....

It hardly matters, honestly. The science is still getting done. There's nothing to suggest that any results are getting lost. It's just taking some time to count them in the stats system and credit points.

I think there is a problem but I'm not sure where.

Frogging101 wrote:Only in the short run. If there is a sustained period of X WUs being assigned per hour, we would expect to see the same number coming back, on average.

Right. If the servers actually ISSUED 100 000 WUs and they all came back, (even if only 70 000 came back during the first day) there still would be 30 000 pending credits somewhere. that would have to come back the next day plus the 70 000 from that day.

The other discrepancy that I can think of is that the 100 000 is the number of requests that reached the AS, not the number of WUs that were actually issued ... or the number actually completed. Certainly there are several percent that get lost somewhere -- but the numbers still don't make sense.

I think it's likely that the 100 000 counts some of the Exception: Could not get an assignment events are being counted but not all of them. That's probably because there are several different reasons you can get the message No WUs available for this configuration

I am now getting 548 k PPD on P16418 on the first work unit on my Ryzen 3950X (Ubuntu 18.04.4, 30 out of 32 cores).
That is not far off the 850 k PPD I am getting on my GTX 1070 (P 11778).

Considering the power difference, this is the first time I have ever seen a CPU approach a GPU in efficiency.

EDIT: But I did have to retire my RTX 2060 due to lack of work, or more accurately switch it to GPUGrid.
Otherwise, I have been getting work consistently. I am now on my second CPU work unit, P14616 at 554 k PPD.

JimF wrote:Otherwise, I have been getting work consistently. I am now on my second CPU work unit, P14616 at 554 k PPD.

>500k from CPU alone?
Think I gotta retire my poor i7-2600k after all, I can't even get there with CPU and GPU combined.

uyaem wrote:Think I gotta retire my poor i7-2600k after all, I can't even get there with CPU and GPU combined.

You won't do badly with a Ryzen 3700X. I was getting 275 k PPD there, and it is a lot more practical. I just needed to upgrade one of my machines anyway.

A suggestion.... once things are stable again, it would be really great to see a list of who contributed extra server hardware to the cause. Even just a kudos article or something from the F@H team. I know Linus Tech Tips has a server. It looks like Oracle and maybe Avast as well? Looks like more and more come online every day.

AmDD wrote:A suggestion.... once things are stable again, it would be really great to see a list of who contributed extra server hardware to the cause. Even just a kudos article or something from the F@H team. I know Linus Tech Tips has a server. It looks like Oracle and maybe Avast as well? Looks like more and more come online every day.

Alongside Oracle's and Avast's servers, KTH Royal Institute of Technology seems to have fired up one, too. Great seeing some engagement in the Nordic countries as well

JimF wrote:

uyaem wrote:Think I gotta retire my poor i7-2600k after all, I can't even get there with CPU and GPU combined.

You won't do badly with a Ryzen 3700X. I was getting 275 k PPD there, and it is a lot more practical. I just needed to upgrade one of my machines anyway.

Do you use all 16 threads of the 3700X to get that PPD? Or do you also have a GPU slot running?

iceman1992 wrote:Do you use all 16 threads of the 3700X to get that PPD? Or do you also have a GPU slot running?

That was with all 16 threads. You would take a small hit by using one for a GPU, but it would be worth it.
It is just that my GTX 980 on that machine is too much this time of year. I will fire it up next fall. I have other machines with better GPUs anyway.