Message boards :
Number crunching :
Athlon 64 x2's and v2.10 of crunc3rs app
Message board moderation
Author | Message |
---|---|
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
I've just obtained a new PC running an AMD Athlon(tm) 64 X2 Dual Core Processor 4200+ under windows XP, this runs two seti programs at once, which should by my reckoning run at 50% cpu allocation each under task manager, which they do most of the time. ;-) But I've noticed that every now and then one of the seti programs starts consuming maybe 80% for 10 seconds or so leaving the other with only 20%. The wierd thing is that the CPU time recorded in Boinc stays as if the seti process was using 50%. so that in effect the task manager says the process has consumed say 80 minutes but Boinc says 60 minutes and so the creit claimed is 60 minutes. but the second process has been slowed down in real time as the cpu cycles has been stolen and lost by the other process I'm currently losing about 25% of the CPU seconds registered in task manager for the Seti processes compared to what BOINC shows So I did some experimentation using CPU Affinity After letting seti run for a while I noticed that the clients gradually slowed down to a halt as far boinc was concerned but were each consuming 50% CPU this didn't make a difference what method of setting cpu affinity (manually, SMPSeesaw or through the truXosft boinc client) so took the plunge and finnally worked out how to roll back to the unoptimised clients, and bingo now there's no cpu cycle loss whatsoever, so looks like if I want faster processing I have to put up with about 20-25% cpu cycle loss per WU. But I'm still better off than using unoptimised clients I'm using Crunch3rs Athlon 64 optimised client, which seems to be the "culprit". I know these things are let the buyer beware but has anyone seen this before or know of a workaround. |
mr.kjellen Send message Joined: 4 Jan 01 Posts: 195 Credit: 71,324,196 RAC: 0 |
Hmm, I don't see any of that going on with my machines. Just to check, you have set the preferences to allow work on 2 processors, right? They should both have 100% work all the time. /Anton |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
Yep, Seti's setup for 2 processors and the cpu % in task manager does add up to 100% in total, plus the green cpu graphs are running at 100% each. Its's just that the cpu % in task manager should be a constant 50% per process, ie the load is split equally 50% per core, and that the processes should really only stick to one core each. What happens is that the processes every now and then take cpu cycles from the other core and this cross over doesn't get recorded in boinc. hence my attempts with cpu affinity...... I'm using the SSE3 AMD64 crunch3r client, I've realised there are other AMD optimised clients out there, I might try those tonight. |
mr.kjellen Send message Joined: 4 Jan 01 Posts: 195 Credit: 71,324,196 RAC: 0 |
Ok. That's not anything I've seen here. But I'm not looking very carefully ;) I am using the 2.09s version of the SSE3 AMD64 optimizations, and I see you are up to version 2.10. /Anton |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
You can spot it quite easily if it happens. When the WU gets to say 95% processed check the CPU time fro rhe client in Boinc and then in task manager, they should be reasonably close say up to a minute difference max, mine are out by up to 20 minutes sometimes....... I'll try the 2.09 version. There's an older version setion on crunc3rs website, I'm assuming thats 2.09. Cheers Keith |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
Got it! tried 2.09s which is the "older version" from Crunch3rs website and everythings working smoothly, no need for cpu affinity or anything.... anyone know how I get hold of crunch3r to let him/her know that AMD Athlon(tm) 64 X2 Dual Core Processor based systems should use the old version of the Athlon 64 SSE3 optimised client? |
trlauer Send message Joined: 6 May 04 Posts: 106 Credit: 1,021,816 RAC: 0 |
Got it! He has an e-mail address listed on his website near the bottom of the page. It is a disguised e-mail address. Torrey Lauer Modern Travel Services / Rainbow Sky Travel moderntravel d.10t net / RainbowSkyTravel d.1.0t com |
miw Send message Joined: 31 Jan 01 Posts: 20 Credit: 8,552,848 RAC: 0 |
Yup. That fixed it! I was having the same problem with the latest Crunch3r app and an Athlon 64 X2 4800+. If no CPU affinity set, then I was losing about 50% of the cycles. In fact this machine had a lower RAC than my A64 3000+ with the same projects. If I set cpu_affinity, then SETI stopped altogether. Now with the older version and cpu_affinity set, it is crunching twice the WUs it was before, and so far I think the other projects are doing better as well (because cpu_affinity is set.) --Mark Got it! --miw |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
Hmmm.... still no update on Crunch3rs Website.......... anyone reading this who knows Crunch3r?, that could forward this information on to. I sent an email but it's probably been lost in the noise, as I imagine he get quite a few emails from users requesting support, and he assumed mine was just another one. Cheers Keith |
Daniel Schaalma Send message Joined: 28 May 99 Posts: 297 Credit: 16,953,703 RAC: 0 |
Hmmm.... still no update on Crunch3rs Website.......... Keith, As far as I know, Crunch3r is no longer working on the standard Seti apps at all. 100% of his time is being devoted to Seti Enhanced optimizations. With Enhanced soon to be released, there is no reason to try to further tweak the standard app. Crunch3r & co are working with the devs to make sure that the Enhanced app is free of bugs, and that the processor specific optimized versions are ready to go the minute Enhanced is released for production here. As long as the older version is working well for you, I would just keep using that until Enhanced is released. Happy crunching! Regards, Daniel. |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
As far as I know, Crunch3r is no longer working on the standard Seti apps at all Ah, that would explain it! Cheers |
BorisM Send message Joined: 31 Jul 03 Posts: 14 Credit: 1,888,005 RAC: 0 |
i had same problem on my x2 3800+...i was using crunch3rs opt. app. (upated on 25.12.2005 with sse3)...i was loosing about 30-40% of my cpu time with that app. Right now im using the older version of crunch3rs opp. app. (with sse3) and everything is working fine...give it a try ... http://www.bm-makler.de/setiathome/setiathome/old/Athlon64_SSE3.zip |
KWSN - Sir Brian - err sorry - wrong film! Send message Joined: 18 Feb 06 Posts: 11 Credit: 674,394 RAC: 0 |
Yep, that's it make sure the version is 2.09s and everything should work ok so far the behaviours been seen and corrected on a 3800 4200 and a 4800 so it looks like it specific to the x2. my gut feel is that it's L1/L2 cache specific. in that each core has a dedicated cache. |
JohnB175 Send message Joined: 15 Oct 03 Posts: 124 Credit: 321,769 RAC: 0 |
Anyone know if this also affects Pentium D machines? |
Babyface uk Send message Joined: 28 May 03 Posts: 86 Credit: 1,972,184 RAC: 0 |
You can add the same problem/solution for a 4400+. And no this don't affect any pentium machines. babyface UK |
Arnulf Send message Joined: 23 Oct 00 Posts: 9 Credit: 14,854,293 RAC: 16 |
I have the same problem with my new Opteron's. Time before upgrading the machine, with 2 x Opteron 242's at 1,6GHz each: 53 minutes on each CPU. After: about 140 minutes on each of the 4 cores on my new 265's at 1,8GHz. I'm currently running SeeSawPro, and at least each core uses about the same time and are not fluctating wildly. The computer: http://setiathome.berkeley.edu/show_host_detail.php?hostid=2316167 Running benchmark using crunch3r's 2,09 version: 2x242's ( two cores ) = around 5700 2x265's ( four cores ) = around 3500 ... Any tips on how to fix this are welcome! Edit: Editing the prefs and reducing the available CPU's to 2 made the benchmark look better - 7447. But I really would like to be able run all 4! Edit2: Weird things seems to be happening - check out this result: http://setiathome.berkeley.edu/result.php?resultid=282718294 ( std_err.txt seems to be repeating itself 6 times?! ) |
Arnulf Send message Joined: 23 Oct 00 Posts: 9 Credit: 14,854,293 RAC: 16 |
More info: SeeSawPro seems to loose the settings everytime a WU are finished and a new WU are started, so it's not very useful in the long run. I'm now testing Trux's calibrating BOINC with Crunch3r's client and the <cpu_affinity/> command. So far I have succeded to isolate the process on CPU 1 - both cores. Now I will have to wait to se if it's working. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.