Timeouts - Where are they heading, and should we report them
Posted: Thu Aug 26, 2021 8:16 pm
With the influx of the higher end GPUs in the last couple of years, it seems some of the timeouts are tightening up quite a bit. Though this is understandable, my question is essentially, should we report when any certain GPU/iGPU is running short of making the timeouts?
For the record, not a complaint, and I really don't care about points. I've seen posts concerning the time it takes to classify GPU's, and why lesser GPU's in a class might not make the timeout while others make it just fine. I also understand that the people making things happen have limited time for all these various projects and the support it takes. I'd just rather not risk WU's being issued more than once on a regular basis if it can be avoided. Though I have a meager iGPU that averages probably in the 70-75K range for the last 16 months or so, it seems the rate of missing timeouts has gone up greatly in the last couple months. Of those 16 or so months, I've only dumped 1 WU, in accordance with the rules and policies. But there are series of WU's that are being reassigned and causing duplicate work to be done on a somewhat regular basis in some cases.
I'd gladly report those instances if it can help, but don't want to report it if it just frustrates those already spread thin.
It would seem to me that there are so many newer and very fast GPU's in use that the researchers could put stricter limits on GPU species to ensure that work isn't done more than once, and more importantly the science advances more quickly. But I also understand that if the ID/species/specs make that difficult than it's just another project for someone already busy.
I do get it. Time marches on and technology advances. But it seems I'm seeing more on the forum here, and more often when checking WU status, that more and more hardware is not capable of the timeouts lately. If it comes to that, I'll just CPU fold here and there until I can find the right GPU for my situation. But more than anything, I was wondering if the tightening timeouts should be reported at all, or if the trend is moving towards speeding all work up and making more hardware obsolete for folding sooner rather than later.
Another slightly related question.... while checking WU status I obviously saw where WU's had been assigned to a second user if I exceeded timeout, as expected. But I came across a couple of instances of WU's being assigned to a second user well before the timeout... in some cases another user picked up the same WU about halfway or so into the timeout time I initially received. I've never noticed this in the past, and curious what might cause it.
And to end on a positive, I saw the update about measures being taken to prevent the cherry picking of WU's and support it 100%. I hope that trend continues.
For the record, not a complaint, and I really don't care about points. I've seen posts concerning the time it takes to classify GPU's, and why lesser GPU's in a class might not make the timeout while others make it just fine. I also understand that the people making things happen have limited time for all these various projects and the support it takes. I'd just rather not risk WU's being issued more than once on a regular basis if it can be avoided. Though I have a meager iGPU that averages probably in the 70-75K range for the last 16 months or so, it seems the rate of missing timeouts has gone up greatly in the last couple months. Of those 16 or so months, I've only dumped 1 WU, in accordance with the rules and policies. But there are series of WU's that are being reassigned and causing duplicate work to be done on a somewhat regular basis in some cases.
I'd gladly report those instances if it can help, but don't want to report it if it just frustrates those already spread thin.
It would seem to me that there are so many newer and very fast GPU's in use that the researchers could put stricter limits on GPU species to ensure that work isn't done more than once, and more importantly the science advances more quickly. But I also understand that if the ID/species/specs make that difficult than it's just another project for someone already busy.
I do get it. Time marches on and technology advances. But it seems I'm seeing more on the forum here, and more often when checking WU status, that more and more hardware is not capable of the timeouts lately. If it comes to that, I'll just CPU fold here and there until I can find the right GPU for my situation. But more than anything, I was wondering if the tightening timeouts should be reported at all, or if the trend is moving towards speeding all work up and making more hardware obsolete for folding sooner rather than later.
Another slightly related question.... while checking WU status I obviously saw where WU's had been assigned to a second user if I exceeded timeout, as expected. But I came across a couple of instances of WU's being assigned to a second user well before the timeout... in some cases another user picked up the same WU about halfway or so into the timeout time I initially received. I've never noticed this in the past, and curious what might cause it.
And to end on a positive, I saw the update about measures being taken to prevent the cherry picking of WU's and support it 100%. I hope that trend continues.