| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Beta Testing
Forum: Beta Test Support Forum
Thread: New Beta starting Aug 5, 2011 |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 94
|
|
| Author |
|
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges: 
|
It's my experience, too, that these betas don't like AMD Phenoms running Ubuntu. I have one Phenom II quad. It never gets invalids, except for one batch of DDDT2 and now these betas. I suspect it's something to do with AMD handling 64-bit better than Intel.
----------------------------------------Here are my result counts on these betas on 4 machines:
Unfortunately I aborted the WUs that my AMD Fusion laptop received from the 7/29 batch because they were causing just too much heat. (I run C4CW and HCC on it regularly.) My guess is that Fusion laptops are going to be very popular, so it would be good if WCG sciences run well on them.  |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
AMD better at 64 bit than Intel as a cause? WCG has/had for some sciences ''homogeneity groups" for CPU types, but the techs don't like it as it's manual labor to maintain the tables [maybe not with updated server software]. If AMD pairings would lead to valid results, then, well that could get tested. ;P
--//-- |
||
|
|
Bearcat
Master Cruncher USA Joined: Jan 6, 2007 Post Count: 2803 Status: Offline Project Badges:
|
Bearcat, How often does this system ask for work? I use .3 to connect and .7 to cache. Its a 2.93 xeon hex so takes about an hour to complete HCC. Guess I will have to monitor more closely.
Crunching for humanity since 2007!
 |
||
|
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges:
|
AMD better at 64 bit than Intel as a cause? WCG has/had for some sciences ''homogeneity groups" for CPU types, but the techs don't like it as it's manual labor to maintain the tables [maybe not with updated server software]. If AMD pairings would lead to valid results, then, well that could get tested. ;P --//-- Sorry -- "AMD better at 64 bit" as a cause wasn't a good way for me to phrase that. This is pure guesswork on my part, but I suspect what may be happening is that a 64-bit AMD processor may carry along more significant digits than Intel when going through a sequence of calculations, so when rounding to 32 bit for the final result, there may be a difference out in the last digit or two. If the inputs to the calculation were in 32-bit, then whatever significant digits beyond that AMD was carrying along would be meaningless, so the AMD calculation wouldn't necessarily be "better," but it also wouldn't necessarily be worse. If the validation comparison is made to a very tight precision, then "inconclusives" will happen when an Intel and an AMD machine are paired if these slight rounding differences are present. Since there are more Intels than AMDs out there, the 3rd wingman is likely to be an Intel. Thus the AMD WU is the one declared "invalid," whether it was actually "better" or "worse" to begin with. I'd love to see a test of AMD pairings if it didn't make extra work for the techs. |
||
|
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1320 Status: Offline Project Badges: 
|
AMD Athlon II
BETA_ BETA20_ ace_ 0000000_ 0480_ 0-- 1430776 Valid 05/08/11 18:08:11 05/08/11 22:49:58 4.47 88.6 / 102.7 AMD Athlon II BETA_ BETA20_ ace_ 0000000_ 0013_ 0-- 1430653 Pending Validation 05/08/11 18:07:30 05/08/11 22:29:35 3.38 68.0 / 0.0 AMD Phenom I BETA_ BETA20_ ace_ 0000000_ 0003_ 2-- 674419 Valid 06/08/11 02:41:30 06/08/11 10:57:25 6.46 90.1 / 97.7 |
||
|
|
kateiacy
Veteran Cruncher USA Joined: Jan 23, 2010 Post Count: 1027 Status: Offline Project Badges:
|
That's fascinating, CP! Are those machines running Windows or Linux?
---------------------------------------- |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
That's fascinating, CP! Are those machines running Windows or Linux? To add, what bit size, and what Windows or Linux (kernel version too). As for the occasional series of 4-5 that hit a signal 11 on the quad [right when one task finishes and a new one starts], middle of the night too, it's only valid valid valid that is delivered from 2.6.38.10 bld 47, 64 bit. Last time this happened was July 26, with 4 DDDT2 doing this within a few seconds and before that 4 on July 16, 3 CEP2+1 C4CW and before that June 25, same mix. Whilst we report all these observations, which is great for other volunteers to look out for when testing, the Techs do get all the logs and base technical system info, and detail quorum failure reasons, so if there are commonalities, then they'd be seeing them. Will wait for Monday, what happens next on the testing front. --//-- |
||
|
|
KerSamson
Master Cruncher Switzerland Joined: Jan 29, 2007 Post Count: 1672 Status: Offline Project Badges: 
|
Hi,
----------------------------------------As you can see below (AMD, Ubuntu 10.04 x64): BETA_ BETA20_ ace_ 0000000_ 0970_ 0-- enezeussa Invalid 8/5/11 18:13:24 8/6/11 06:08:32 2.94 89.3 / 72.1 In comparison (Intel, WinXP Pro SP3; excepted kerdiwi01, see above): BETA_ BETA20_ ace_ 0000000_ 0089_ 1-- kerdiwi01 Valid 8/5/11 18:09:27 8/6/11 06:15:18 3.39 88.7 / 92.9 My hosts computed much more Beta WUs end of July, but the results have been already purged. Cheers, Yves --- PS: Since the other BETA WUs are currently considered as inconclusive, I assume that they will also become invalid during the next hours. Regarding the lost efforts, it would be fine to handle the problem in a more reliable way. ---------------------------------------- [Edit 1 times, last edit by KerSamson at Aug 6, 2011 3:27:26 PM] |
||
|
|
KWSN - A Shrubbery
Master Cruncher Joined: Jan 8, 2006 Post Count: 1585 Status: Offline |
I had two of four go invalid on my Opteron system with Ubuntu 64 bit. If the techs are getting all the details, as you say SekeRob, then I'll let them gather that from the logs.
----------------------------------------Invalid results do seem to be a significant problem with this project, however.  Distributed computing volunteer since September 27, 2000 |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
After my previous posts I’d like to summarize my experience so far with this betas. First is important to keep in mind that all I’m going to say applies to this particular setup:
Pentium 4, HT-on, WinXP 32 bit, WCG BOINC client 6.10.58 set to use 100% processors and 50% CPU time. As I said in my previous posts, two of these betas in parallel ignores BOINC settings regarding CPU time limits, running 100% almost all the time, and not suspending immediately when selecting “snooze” in the client. Running one of the betas in parallel with another project is much better, but still not perfect. While running two betas in parallel, CPU time for both jobs almost doubles the elapsed time. This behavior disappears when running only one. But wait! Isn’t it always the other way around? (elapsed time should be >= CPU time). What’s going on here? Regarding Sek’s comments, I’ve tried TThrottle in the past and I didn’t like it very much. But even if that program could help my particular case, the majority of users would still be using the client throttle. This new project, as all the others, has to adhere perfectly to CPU time limits set in BOINC (or any other setting for that matter) regardless the combination of OS/CPU/BOINC. |
||
|
|
|