Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Raistmer on 20 Feb 2011, 05:00:44 am
-
After doing some amount of AP tasks my GPU fetched MB tasks again. And all were marked ~22-23 min of estimated time.
Even VLAR ones.
Definitely my GPU takes much longer to compute them. To this point it finished much more than 10 MB tasks so server estimates should be stable.
So, AP work still affects on MB estimates?
Or smth scewed again on server side?
Very probably I get -177 errors on all or most of just downloaded tasks... :(
-
After doing some amount of AP tasks my GPU fetched MB tasks again. And all were marked ~22-23 min of estimated time.
Even VLAR ones.
Definitely my GPU takes much longer to compute them. To this point it finished much more than 10 MB tasks so server estimates should be stable.
So, AP work still affects on MB estimates?
Or smth scewed again on server side?
Very probably I get -177 errors on all or most of just downloaded tasks... :(
Duration Correction Factor in the core client applies to (and reacts to) all tasks for the project. If it is 1 then the elapsed time limit is 10x the estimate. But if DCF is 10 the limit is the same as the estimate, or if DCF is 0.1 the limit is 100x the estimate.
Certainly it's also possible there's a server side problem, I noticed a couple of NC forum posts about -177 errors but without enough detail.
Joe
-
I noticed that my ATI WU's were estimated for 7 hours yet the CPU WU's were all around 2 hours. the 2 hour WU's are about right so what gives with the ati ones
-
After doing some amount of AP tasks my GPU fetched MB tasks again. And all were marked ~22-23 min of estimated time.
Even VLAR ones.
Definitely my GPU takes much longer to compute them. To this point it finished much more than 10 MB tasks so server estimates should be stable.
So, AP work still affects on MB estimates?
Or smth scewed again on server side?
Very probably I get -177 errors on all or most of just downloaded tasks... :(
Duration Correction Factor in the core client applies to (and reacts to) all tasks for the project. If it is 1 then the elapsed time limit is 10x the estimate. But if DCF is 10 the limit is the same as the estimate, or if DCF is 0.1 the limit is 100x the estimate.
Certainly it's also possible there's a server side problem, I noticed a couple of NC forum posts about -177 errors but without enough detail.
Joe
I running Jason's mod and AFAIK this mod allows setarate corrections for separate apps. This puzzles me even more...
-
I running Jason's mod and AFAIK this mod allows setarate corrections for separate apps. This puzzles me even more...
You are still working with/against server side dcf estimates.
Plus while Jason's correction formula has a higher down slope than the original (iirc) it still needs time to correct for large changes - and you don't have the panic up correction of the original, but the same asymptotic bahaviour as for downwards corrections.
So unless something messed up the adcf value stored for the app in question (or your setup has changed) it has to be server side.
I'd edit the acdf to something more sensible and then run Fred's rescheduler to correct for the possible -177s.
-
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for astropulse_v505 505 (ati13ati) windows_intelx86 to 1.417662
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for setiathome_enhanced 610 (ati13ati) windows_intelx86 to 0.432831
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for setiathome_enhanced 608 (cuda) windows_intelx86 to 0.922735
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for setiathome_enhanced 603 () windows_intelx86 to 1.115482
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for astropulse 506 (ati13ati) windows_intelx86 to 1.471724
22/02/2011 01:51:14 [aDCF] Client state set anonymous platform aDCF for astropulse 506 (ati13ati) windows_x86_64 to 1.471724
-
My DCF is completely off the chart, it is at 8.12 right now. GPU work all looks normal, but CPU work is predicted to run 85 hours.
-
My DCF is completely off the chart, it is at 8.12 right now. GPU work all looks normal, but CPU work is predicted to run 85 hours.
Because the predicted runtime is so high, every task CPU will easily beat that, it'll knock the dcf down, (in time)
so eventually it'll reach 1 again,
Claggy
-
Biggest problem is I have over 1800 WU on the machine, most of them are now ATI.
I have set NNT for the time being on both Milkyway and SETI.