Message boards :
Number crunching :
Long time workunits
Message board moderation
Previous · 1 · 2
Author | Message |
---|---|
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Thanks. Now Julie has the AMD links she needs for her Hosts. Now, if she will just install them :-) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Oh Noes, it did it again. Everything was going great for a couple days, then I found the ATI AP sitting at 46% complete with an elapsed time of 10 hours. Suspending/resuming did nothing. I stopped BOINC 7.0.44 and changed the plan class on that task to nvidia. After relaunching BOINC, the task now says 46% complete, elapsed time 2:57 minutes. That would be about normal for an ATI AP task taking 6.5 hours. The nVidia MB task and the AP CPU task are unaffected, it's just the ATI task that stalled. This is starting to remind me of the DELL P4 XP machine I have that has BOINC LOCK up after 2 days when running just CPU tasks. I fixed the DELL by switching to Linux. Unfortunately, you can't run a nVidia MB, and an ATI AP at the same time in Linux. At least not with Ubuntu, AFAIK. Hmmmm... |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
So far so good. I decided to go back to what worked the last time the 4670 was working with XP. From around Oct to Feb it worked fine in my other Dual core Host with XP. Then from Feb to Aug it worked fine with Windows 8 in the same other Host. Right now it's using BOINC 7.0.36 and AstroPulse r1316. Seems there is less screen lag with BOINC 7.0.36 on that host. For some reason both BOINC 7.0.44 & 7.0.64 invoke quite a bit of Kernel Activity on that Host, and there is quite a bit of redraw flickering in the BOINC Manager. Because of that, I usually kept the Manager minimized, which caused the Kernel Activity to be reduced. In fact, the last couple times an AP stalled, it was overnight with the Manager minimized. I think it was just a coincidence the AP stalled when minimized. If it makes it another day without stalling, I'll try it when minimized :-) |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
It's not a coincidence. The only ATI Stall suffered in the last few days occurred overnight when the BOINC Manager was minimized. The CPU & nVidia tasks were not affected. As soon as the Manager was Maximized, the ATI Progress resumed. This doesn't happen on my other Hosts. Strange things when viewing CPU activity in SIV. The 'heavy' boincmgr Kernel Activity is back. When Maximized, SIV shows boincmgr as using 20-40% 'Kernel' (RED). If you just scroll and hide the 3 active tasks, the Kernel Activity drops to almost nothing. Restarting BOINC does nothing. Whut? |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
It's not a coincidence. The only ATI Stall suffered in the last few days occurred overnight when the BOINC Manager was minimized. The CPU & nVidia tasks were not affected. As soon as the Manager was Maximized, the ATI Progress resumed. This doesn't happen on my other Hosts. Which BOINC version was that particular experience with? Charlie Fenton (BOINC developer with primary responsibility for the User Interface - i.e. the Manager - for all versions), has been upgrading wxWidgets (the control responsible for displaying the grid for the task list) - and he tried some techniques for reducing the flicker, but backed out of them again. My v7.2.28 (x64) shows wxWidgets Version 2.8.10 - could you post comparative figures from your Manager, please? (from Help|About). If there's a problem, we need to catch it quickly before this build is made 'recommended' - and I've delayed it three times already. I'm going to have to be very, very sure of my ground before I try that again. Edit - having read your second-to-last post again, I see you're using an older Alpha manager at the moment. So, maybe testing the release candidate v7.2.28 would be a good idea? You can find the links here. I'd still like to see the wxWidgets version number for the Manager you're using, please. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
I've experienced ATI stalls and Heavy Kernel activity with BOINC 7.0.36, 7.0.44, and 7.0.64. I'll update to 7.0.64 and play with SIV. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
After a fresh system restart, BOINC 7.0.64 shows around 10-35% Kernel, the Apps show less than 1% Kernel apiece. I've seen boincmgr higher with 7.0.64, probably take a couple days as with 7.0.44/36. Hiding the active tasks reduces boincmgr Kernel to around 5%. Total CPU usage in the BOINC Status window is around 55%...with the tasks hidden. I'll give the candidate a try. |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
I have noticed the cpu V7 multibeam going to 5 hours per wu compared to 2 hours or 2.5 maximum time. Since yesterday I have noticed this increase in time and cpu usage is half of what it was before. It was 95% - 100% loaded now 50% is what they are using. Long time for these units. |
Michael W.F. Miles Send message Joined: 24 Mar 07 Posts: 268 Credit: 34,410,870 RAC: 0 |
Well after thinking it was 7.2.27 I downloaded 7.2.28 and upon restarting Boinc I had 100% loading on all cores. 10 minutes later while on the same wu the load dropped to 50% Weird long run times |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
...having read your second-to-last post again, I see you're using an older Alpha manager at the moment. So, maybe testing the release candidate v7.2.28 would be a good idea? You can find the links here. I'd still like to see the wxWidgets version number for the Manager you're using, please. 7.2.28, wxWidgets 2.8.10, appears to be about the same as 7.0.64. The average Kernel use in SIV is around 15% then drops to around 5% when I hide the three active tasks. It drops to just about nothing when minimized. It might be awhile to see if the ATI task stalls when the Manager is minimized. I changed the ATI AP App back to r1843. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
The first task using 7.2.28 finished normally. The next ones will be run with the Manager Minimized. I just checked my Windows 8.1 Host running BOINC 7.0.64, SIV shows the boincmgr Kernel usage at around 1 to 2% while displaying basically the same three type of tasks as the XP Host. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Still no ATI AP stalls using BOINC 7.2.28. I must admit though, since determining that it only stalls when the BOINC Manager is Minimized, I haven't been minimizing it that much. It did make it through a couple of nights minimized. So, for anyone having this problem I would recommend trying BOINC 7.2.28. The GPU Driver updates mentioned earlier wouldn't hurt either. I've also noticed this Host, Computer 7102466 is still needlessly trashing APs and depriving the Raccoons of much needed points. A properly functioning 5870 should produce around 30,000 RAC a day running APs correctly. Also, after posting about the DELL P4, I realized I had never run it with SETI v7 tasks. Back on the 5th, I set it up to run the Lunatics v0.41 SSE2 CPU App. Other than the accumulated XP updates and updating to BOINC 7.0.64, that was the only change made since turning it off back in March. It is now approaching 6 days of 24/7 operation and the BOINC Manager still hasn't crashed. I think I will let it run a couple more days then give up on it. |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Oh well, just as soon as you post about it, it happens again. Same scenario. Minimize BOINC Manager, go to bed, then wake up to the ATI App showing much longer elapsed time than it should. It had just started the AP, almost 7 hours later it was showing 55% complete, it should have finished at 6.5 hours. After maximizing the Manager, it was showing normal progress. The nVidia CUDA App and CPU AP App were not affected. The CUDA App shows normal progress during the same period. The main Monitor is running off the ATI card with a s-video adapter connected to the nVidia card. The ATI App works fine as long as you keep the BOINC Manager Maximized. There wasn't any Blanking on the ATI AP. Restarting the BOINC Manager didn't reset the Elapsed time. It's working on another AP now, this one does have some blanking. Based on past times, I'd say it will finished the partially blanked AP in about 7 hours, much better than the 9.5 it just took on the Un-blanked one. Just keep it Maximized.... |
Urs Echternacht Send message Joined: 15 May 99 Posts: 692 Credit: 135,197,781 RAC: 211 |
... Did you try to set <no_priority_change>1</no_priority_change> in cc_config.xml already ? _\|/_ U r s |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
... The ATI card is set to -HP in the CL file, the nVidia card is set to Normal in the mbcuda file, you can see it in the Stderr output. I'd rather not raise the priority of the CPU task. What's strange is the inconsistency. I've been noticing a few longer than normal tasks since I moved the card to this host. You can see the OP host also has inconsistency, Error tasks for computer 7061796... ? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well, it's been 5 days since the last 'Stall'. The previous Stalls were occurring about every 2 or 3 days. Since leaving the BOINC Manager Maximized, the Stalls have disappeared. The un-blanked APs are all completing in a pretty consistent run-time, AstroPulse v6 tasks for computer 6979629 So, what's up with that? |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
It appears I have found the solution on MY Host, Computer 6979629 Leaving the BOINC Manager Maximized worked for the then existing setup, the HD4670 in the 16x slot driving the monitor and the GTS250 in the 1x slot connected to a S-video adapter. Then I changed the setup to having an 8800GT in the 16x slot running the monitor and a HD6770 in the 1x slot connected to the S-Video connector. The 6770 began going idle, sometimes within 30 minutes. The Task Time Elapsed kept right on going even though SIV showed the GPU had ceased output, down-clocked, and cooled down. Again, the nVidia card wasn't affected. WTH? After some time, I determined it must be the ATI Driver with Catalyst 12.1 since the nVidia card with driver 266.58 wasn't affected. Then I changed the Windows XP Power Setting from putting the monitor asleep after 2 hours to NEVER. So far, no more idle moments for the 6770. Apparently, the ATI driver thought that setting meant it was fine to kill the GPU whenever it wanted, as long as there wasn't any movement on the screen. Now to see what happens overnight after I turn the monitor Off... |
TBar Send message Joined: 22 May 99 Posts: 5204 Credit: 840,779,836 RAC: 2,768 |
Well, that worked for about a day.... I went back to what worked before...Just Leave It Maximized. Since I had the XP ATI desktop to the right of the NV desktop, I just extended the Event Log window so part of it was on the ATI desktop. That works, but, sometimes it's a few minutes before there is activity in the Log and the ATI card will Sleep for those few minutes. I needed something small and simple to create motion on the ATI desktop constantly. A Simple Fix, that works; http://softadvice.informer.com/Free_Desktop_Clock_For_Windows_Xp.html ClocX, It Works. Install it, right click on the tray icon and chose Options.../Appearance, and check 'Draw second hand'. Then drag it to the ATI Desktop. The second hand is all the motion you need to keep the ATI card from Sleeping. Look at the Consistent Runtimes for the tasks with a CPU time of around 200 seconds. They are all within a few seconds since using the clock, because, The ATI card is Not Sleeping. All AstroPulse v6 tasks for computer 6979629 Works for me. |
©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.