Message boards :
Number crunching :
Smaller Work Units Needed
Message board moderation
Author | Message |
---|---|
Send message Joined: 26 Mar 09 Posts: 7 Credit: 702,781 RAC: 0 |
I need smaller work units. My computers are not able to complete the massive work units by the deadline, so a bunch of time is being wasted. Back when the work units were smaller my computers had no trouble whatsoever meeting the deadlines. Please help. |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
Are you running n_body or de_separation (the classical MW) tasks? I can't see any for your host now. I'm asking because the de_separation tasks are really well suited for ATI GPUs. From my point of view running them on any other hardware is wasting energy. Your CPU could probably of use for the n_body tasks, though. They're only for CPU and should thus have longer deadlines and / or less work in them. To put the issue with the de_separation tasks into numbers: your Athlon XP 2400+ has done ~70k credits here at MW. That's nice, but my tuned HD6950 is doing this much work in less than 7 hours! MrS Scanning for our furry friends since Jan 2002 |
Send message Joined: 19 Jul 10 Posts: 601 Credit: 19,025,629 RAC: 5,740 |
Is the current n_body app running without SSE2? It's not written anymore on the application page and since I'm not running Milkyway ATM, I don't know exactly what it needs, but an AthlonXP does not have SSE2, so it was not running on mine before too. As to shorter work units... my AthlonXP 2000+ needed before less than 11 hours for one 214Cr-WU with the old opt. app, so if the new apps are not awfully slow, than it should not be a problem to finish them in time, even if the machine is not running 24/7. |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
my AthlonXP 2000+ needed before less than 11 hours for one 214Cr-WU with the old opt. app, so if the new apps are not awfully slow, than it should not be a problem to finish them in time, even if the machine is not running 24/7. You're right - and you demonstrate nicely why I think it doesn't make any sense to run de_separation tasks on the CPU. Within 11 h the results will very probably long be obsolete anyway, so can not really contribute to the search. And good point about SSE2. Don't know either, sorry. MrS Scanning for our furry friends since Jan 2002 |
Send message Joined: 19 Jul 10 Posts: 601 Credit: 19,025,629 RAC: 5,740 |
Within 11 h the results will very probably long be obsolete anyway, so can not really contribute to the search. If that's the case, than I don't know why any tasks are send to CPUs. I mean, a "normal" user will run that probably only as a screensaver and not 24/7, so even with a faster CPU he will probably need longer than these 11 hours, more something like 1-2 days. |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
When the ATI app was introduced we had a heated debate over this and I brought up the idea of moving "ATI only". We also had a fellow who basically said "no matter how inefficient it may be, I want to crunch these WUs on my CPU and it's not your place to forbid me doing so". Well, can't argue with that. I never got an official answer from the project staff, but I'm sure they didn't want to alienate CPU crunchers, at least in case they'd need them again in the future (hint: n_body simulation). Why de_separation hasn't moved totally to the GPU today I can't tell. Might be nice to get an official answer. I do feel that the project is not the quickest to flexibly react to new situations, though (saying this without judgement - it's not neccessarily bad). MrS Scanning for our furry friends since Jan 2002 |
Send message Joined: 12 Nov 07 Posts: 2425 Credit: 524,164 RAC: 0 |
Certain calculations can't be done on GPUs so they must be done on CPUs. So both are needed/ or can be used by the project. Doesn't expecting the unexpected make the unexpected the expected? If it makes sense, DON'T do it. |
Send message Joined: 1 Sep 08 Posts: 204 Credit: 219,354,537 RAC: 0 |
Yeah.. it's just that de_separation is not part of that. MrS Scanning for our furry friends since Jan 2002 |
Send message Joined: 15 Jul 08 Posts: 383 Credit: 729,293,740 RAC: 0 |
It's been suggested over and over that CPUs be limited to N-body WUs. The separation WUs run so fast on ATI it seems wasteful to run them on anything else. But I suppose that's up to whoever is paying the electric bill :) |
Send message Joined: 19 Jul 10 Posts: 601 Credit: 19,025,629 RAC: 5,740 |
Hmm... I just tested one separation WU on my Pentium M laptop, it was one for 213.76 credits. The CPU time compared with the old v0.21 sse2 opt. app was about doubled, from ~31.000 (+/- 1.000) seconds to 61.090. If the slowdown on the AthlonXPs is about the same, that means 20-24 hours CPU time, depending on the usage of such machine and BOINC configuration it can be indeed hard to make it before the deadline. |
Send message Joined: 8 May 10 Posts: 576 Credit: 15,979,383 RAC: 0 |
It's been on my list of things to do for a long time to figure out some way for a CPU-Nbody GPU separation preference but I have no idea when I'll get to it. |
©2024 Astroinformatics Group