Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
![]() |
World Community Grid Forums
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
No member browsing this thread |
Thread Status: Active Total posts in this thread: 55
|
![]() |
Author |
|
DrMason
Senior Cruncher Joined: Mar 16, 2007 Post Count: 153 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Got a bunch of beta WUs. I previously noted a severe linux penalty to run times based on two systems with the same internal configuration, but one was running linux and the other was running windows. I'm happy to report that my linux results seem to be vastly improved with these new beta units. Ryzen 5 3600 with Windows 10 finishes in 1.3 hours, and with Linux finishes in 1.5 hours. Massive improvement! Thanks for the hard work and congrats on the improved results!
----------------------------------------Work unit time scales well with clock speed, and seem to be otherwise well behaved. Will chime in if I get any that error out or behave unusually. ![]() |
||
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1330 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
On Windows the previous Beta-version needed 2.56 hours/task with 5 Beta's and 25 OPN1's running concurrently.
----------------------------------------Now with 30 Beta's running concurrently 3.35 hours is needed per task. With Linux OS on the same hardware the previous Beta-version needed 6.2 hours when running 10 Beta's and 4 OPN1's. This version needs 3.67 hours/task when running 14 Beta's concurrently. A very well improvement for Linux, but also showing when more MCM's are running concurrently the tasks slow down each other. [Edit 1 times, last edit by Crystal Pellet at Jul 22, 2020 8:56:47 AM] |
||
|
adriverhoef
Master Cruncher The Netherlands Joined: Apr 3, 2009 Post Count: 2217 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
With Linux OS on the same hardware the previous Beta-version needed 6.2 hours when running 10 Beta's and 4 OPN1's. With respect, is this a fair comparison?This version needs 3.67 hours/task when running 14 Beta's concurrently. We seem to have two specific types of MCM1 tasks, one type that runs in half the time of the other type: Result Name Status Sent Time Due / Return Time CPUh/Spent Claimed/Granted[Copied from Results Status, generated by wcgformat] The one 'halved' type running in almost half the time of the other, long type. Did the former MCM1-BETA have the long type and does the current one have the 'halved' type? Result Name Status Sent Time Due / Return Time CPUh/Spent Claimed/Granted[Copied from Results Status, generated by wcgformat] Looking at the results above, the current MCM1-BETA seems to carry the 'halved' type. Compare the following two MCM1-BETAs (from a different device), the first one from two weeks ago, the latter the current one: App CpuTime Elapsed Claimed Granted ModTime Exit Outc SentTime ReceivedTime Name Nevertheless, if the long type lasted 8 hours and the 'halved' type now lasts 3 hours (instead of 4, see below, 1 week ago), that would still be impressive. App CpuTime Elapsed Claimed Granted ModTime Exit Outc SentTime ReceivedTime Name |
||
|
Crystal Pellet
Veteran Cruncher Joined: May 21, 2008 Post Count: 1330 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Did the former MCM1-BETA have the long type and does the current one have the 'halved' type? Since the Windows-tasks took even longer (probably cause of more concurrent BETA-MCM1-tasks), I suppose all tasks of this BETA-batch are from the same long sized jobs as the previous BETA-batch (when not even exactly the same tasks, but with a higher batch-number). |
||
|
adriverhoef
Master Cruncher The Netherlands Joined: Apr 3, 2009 Post Count: 2217 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Seeing a considerably higher Working Set Size in the MCM1-BETA than in the 'normal' MCM1-workunits:
----------------------------------------$ wcgresults -NDTPSOVW1 On another device:Thu Jul 23 18:09:44 2020 8:04:08 99.8343% (R) 6 90 MB 89 MB MCM1_0165037_9619_1 MCM1: 70 MB à 90 MB BETA: 238 MB [Edit 2 times, last edit by adriverhoef at Jul 22, 2020 2:58:19 PM] |
||
|
Mike.Gibson
Ace Cruncher England Joined: Aug 23, 2007 Post Count: 12561 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
i5-M520@2.40GHz Win7Pro 3 Valid 1 PVal ave.CPU 4.905 ave. elapsed 6.21
i7-3770@3.40GHz Win7Pro 5 Valid 3 PVal ave,CPU 2.0775 ave. elapsed 2.1225 Mike |
||
|
Grumpy Swede
Master Cruncher Svíþjóð Joined: Apr 10, 2020 Post Count: 2277 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I got 4 on my Windows 8.1, Intel(R) Core(TM) i7-3630QM CPU @ 2.40GHz. They did run to completion with no problems, when I was sound asleep.
|
||
|
Pete Broad
Senior Cruncher Wales Joined: Jan 3, 2007 Post Count: 169 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Well, my Androids picked up quite a few. Unfortunately some of my tablets were overwhelmed with the numbers received despite having ample memory. I was unable to suspend them so I had to abort them. One of my Androids got 6 of them, locked the Android up despite having 32mb. Sorry about that, I hate aborting tasks and I'll look into limiting the numbers that I get. Those Androids with 3 units or less look to be O.K.
----------------------------------------Sadly, there doesn't seem to be a way of limiting beta units so I've left the beta team....for now. Pete ![]() [Edit 1 times, last edit by Pete Broad at Jul 22, 2020 7:10:57 PM] |
||
|
nanoprobe
Master Cruncher Classified Joined: Aug 29, 2008 Post Count: 2998 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
Well, my Androids picked up quite a few. Unfortunately some of my tablets were overwhelmed with the numbers received despite having ample memory. I was unable to suspend them so I had to abort them. One of my Androids got 6 of them, locked the Android up despite having 32mb. Sorry about that, I hate aborting tasks and I'll look into limiting the numbers that I get. Those Androids with 3 units or less look to be O.K. Sadly, there doesn't seem to be a way of limiting beta units so I've left the beta team....for now. Pete I had issues on android too. Had to cut my quad Pine 64s to run on 3 cores for it to work. On 4 the WUs would run for a few minutes then restart.
In 1969 I took an oath to defend and protect the U S Constitution against all enemies, both foreign and Domestic. There was no expiration date.
![]() ![]() |
||
|
alanb1951
Veteran Cruncher Joined: Jan 20, 2006 Post Count: 1061 Status: Offline Project Badges: ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
I was getting quite enthusiastic about the apparent performance improvement I was seeing in the second set of Beta test tasks here, until I realized that every single task I received in the second set uses VMethod=LOO whilst every one I checked in the first set used VMethod=NFCV.
----------------------------------------It has been my experience that pre-Beta MCM1 NFCV tasks for SARC1 take about 125% longer than LOO tasks on all my Intel machines, and about 25% longer on my Ryzen 3700. The differences I'm seeing between the first and second sets of Beta runs here reflect much the same differences (though, if anything, the first set (NFCV) were slightly worse on the Intel machines. Have I just been unlucky in my workload mix, or was all the first set NFCV and the second set LOO? If the former, I'd be interested to hear the experience of someone who got some NFCV in the second set; if the latter, I think a [re]test of some NFCV tasks might be in order to see if there really is an improvement or whether NFCV tasks are just inherently a lot slower (and whether it is to some extent a Windows v. Linux compilation thing...)! [Added] Or are we going to switch all processing to VMethod=LOO??? Cheers - Al. [Edit 1 times, last edit by alanb1951 at Jul 23, 2020 8:47:03 AM] |
||
|
|
![]() |