| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Completed Research
Forum: The Clean Energy Project Forum
Thread: The Clean Energy Project & CPU temps |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 20
|
|
| Author |
|
|
BE04642
Advanced Cruncher Belgium Joined: Dec 14, 2004 Post Count: 59 Status: Offline Project Badges: 
|
Aren't we looking for a better, fatter, Nutricious Rice
---------------------------------------- Hmm, so far the trick seems to be working, 3 rice meals on 1 energy gives me a huge performance boost.  RICE, now that's a light diet. Switched the machine to 2 CEP + 2 HCC and things improved again, so there is a seeming serious resource contention. Yes, the techs have a job to do to get the efficiency up for the multi-cores.  |
||
|
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
Well, I'm glad we've figured that one out. Will report later if the memory pref tweaks keeps the CEP to 1 running per device.
----------------------------------------
WCG
Please help to make the Forums an enjoyable experience for All! |
||
|
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges: 
|
The performance degradation when running CEP on multiple cores is probably due to all the page faults overloading the operating system and causing a bottleneck there.
Further, some CEP WUs are terminating early, and for some unknown reason the Validator is letting these through. Kaiden's low-points WU (E000009_315A_000126013_1 9.28 200.9 / 57.3) is probably one of these. (Kaiden: Check the other guy's claimed/awarded. He will have terminated early & got what he claimed. You completed the job and were mistaken for an overclaimer.) I suspect that these early terminations are more likely to happen when multiple CEP WUs are running simultaneously on a multi-core machine. The page faults may not only be bottlenecking the o/s but also swamping it and causing errors. Would your observations support this postulation? See also: 2 WUs stopped early, CEP is experiencing a huge PF Delta rate . |
||
|
|
nasher
Veteran Cruncher USA Joined: Dec 2, 2005 Post Count: 1423 Status: Offline Project Badges: 
|
well only 1 of my machines multi processes (yep old computers)
----------------------------------------but it sounds to me like there are lots of issues with this project right at the moment i am going to deselect this project for the moment and keep up to date on it and probaly in a week be reconnected again.. of corse thats after i crunch the work i curnelty have   |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I have a 2.8GHz P4 HT running 2 CEP WUs. One of my WU's is going on 50 hours although at 95% so it's almost finished. I hear you guys talking about only running 1 project per process(core). I know I only have a HyperThread processor but is there a way for me to tell my BOINC client to run one CEP and one Rice? Right now it is running 2 CEPs.
|
|||
|
||||
|
Sekerob
Ace Cruncher Joined: Jul 24, 2005 Post Count: 20043 Status: Offline |
I think the P4Ht in this situation is the worst to run 2 jobs simultaneous. 2 RICE may be okay but not heard of anyone actually gaining anything more than a marginal throughput. Thing with RICE is, they run 10 hours irrespective, so it's hard to see other than by comparing credits running 1 thread or 2 threads.
----------------------------------------The only way to control is micro-managing. Fetch work and suspend tasks to the desired combo run. Note that suspending tasks stops work fetch, so you could end up with an idle device.
WCG
----------------------------------------Please help to make the Forums an enjoyable experience for All! [Edit 1 times, last edit by Sekerob at Dec 15, 2008 1:33:12 PM] |
||
|
|
nasher
Veteran Cruncher USA Joined: Dec 2, 2005 Post Count: 1423 Status: Offline Project Badges:
|
there is no easy way to have 1 on cep and the other on rice
----------------------------------------the only way i can see it is manualy you grab a few of each work unit and manualy suspend untill the right combination of projects is running... remember to unsuspend them though the reason i wouldnt go for this is the fact as soon as one of the work units complete you are at it again shifting suspend and restart again and again to do that... its the only way i see but mabey there is a simpler way good luck |
||
|
|
Kaiden
Cruncher Joined: Oct 27, 2008 Post Count: 7 Status: Offline |
The performance degradation when running CEP on multiple cores is probably due to all the page faults overloading the operating system and causing a bottleneck there. Further, some CEP WUs are terminating early, and for some unknown reason the Validator is letting these through. Kaiden's low-points WU (E000009_315A_000126013_1 9.28 200.9 / 57.3) is probably one of these. (Kaiden: Check the other guy's claimed/awarded. He will have terminated early & got what he claimed. You completed the job and were mistaken for an overclaimer.) I suspect that these early terminations are more likely to happen when multiple CEP WUs are running simultaneously on a multi-core machine. The page faults may not only be bottlenecking the o/s but also swamping it and causing errors. Would your observations support this postulation? See also: 2 WUs stopped early, CEP is experiencing a huge PF Delta rate . Yes, the WU that gave me a ridiculously low score gave me the same score it gave the original user who apparently canceled the processing after 4.5 hrs. I processed for the rest. They got 57.3 and so did I. I've previously read both the links you referenced. I agree with both and have seen the same exact issues. I haven't encountered any error messages like some have reported but there are definitely insane levels of PFs and the PF delta is wild. During the WUs that really drove up my CPU temps I had 4 WUs going with each in the area of 900 million PFs and the PF delta for almost all WUs that I checked on fluctuate wildly from 0 to 6000 to 9000, 17000 and even 21000+. The concept of this project is a good one. CEP is a great addition however it would definitely be nice if the WUs were pretty standardized where they all took roughly the same amount of time. At this point CEP WUs just seem to be random and haphazard. I'm happy to take part in distributed computer projects but this one has me a bit bothered about the temp issues. I've used numerous computers and never encountered a similar jump in CPU temps. I've been doing this awhile. I never take part in the beta projects. [Edit 2 times, last edit by Kaiden at Dec 16, 2008 6:12:19 AM] |
||
|
|
BE04642
Advanced Cruncher Belgium Joined: Dec 14, 2004 Post Count: 59 Status: Offline Project Badges:
|
P4 HT performance is a bit off topic, but I can assure you that it really helps doing 2 jobs simultaniously.
----------------------------------------Had until recently a P4 641 Cedar Mill 3.2GHz online, O'clocked to 4.27GHz and at first, for 200 days or so had Boinc set to use 1 (logical) CPU and got an average RAC of 300+. Then I said, what the heck, let's try to fry the CPU. Well, it never did, but it's RAC jumped to 500+ on 32bit XP with only Rice and FAH jobs. Boinc FP Mips was about 2300, can't remember INT Mips but it sure was a nice little bugger If you can, go for XP 64 bit (INT Mips are roughly doubled) and run HCC - mostly integer - and FAH - floating point - or Rice jobs together. Should give a nice RAC in the 6-700 range depending on your overclocking abilities  Go for 2 CPU's on your HT P4, download ProcesXP/XP64 from the sysinternals site. Set anything to idle except the system proces -> Normal, Boinc and Boincmgr and WCG processes to Above Normal, the HCC to Realtime and the FAH or Rice jobs to High. Beware! No 2 realtime jobs at once, the system will hang for awhile and 1 of the jobs will be set to idle... All to be monitored and done manually ofcourse  Voila, your set   I have a 2.8GHz P4 HT running 2 CEP WUs. One of my WU's is going on 50 hours although at 95% so it's almost finished. I hear you guys talking about only running 1 project per process(core). I know I only have a HyperThread processor but is there a way for me to tell my BOINC client to run one CEP and one Rice? Right now it is running 2 CEPs. |
||
|
|
Rickjb
Veteran Cruncher Australia Joined: Sep 17, 2006 Post Count: 666 Status: Offline Project Badges:
|
Sekerob: A late thought re. your post " ... suddenly the fast machine is a severe underclaimer... hard time believing that
----------------------------------------E000030_ 671A_ 00042500h_ 1-- Valid 12/14/08 19:10:11 12/15/08 06:15:55 4.74 78.4 / 132.1" That looks to me like one of these early-terminated WUs that got thru the validator, but lucky for you, points were awarded based on TOG's claim. Is 4.74hrs inside the normal range of run-times for CEP on the machine? Does the stderr file show the "copying wcgrestart.rst" error message? How many CEP WUs would have been running when this one ended? I expect CEP v6.23, currently in beta testing, may detect these as errors. /off-topic/ Interesting observations re HT on the P4, and very relevant to the new i7 CPUs - Nehalem,One cpu to rule them all! [Edit 1 times, last edit by Rickjb at Dec 16, 2008 4:45:28 AM] |
||
|
|
|