| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: GPUs have become unusable; disabling tasks |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 47
|
|
| Author |
|
|
ericinboston
Senior Cruncher Joined: Jan 12, 2010 Post Count: 265 Status: Offline Project Badges: 
|
1)Not to be rude, but I stated in my original post the exact error messages...you're not telling me anything new.
----------------------------------------2)This worked for 9 months...then magically in mid June these 2 GPU errors popped up and BOINC/WCG will not download any more work. 3)Device Manager states there are no problems...I already stated it's using the Standard Windows driver in a previous post. 4)I do not have physical access to this box...I have no way of telling what graphics card is actually in the box unless Windows 2008 tells me...and right now Windows is not telling me unless you know of another MS tool I can download that will identify the hardware. 5)Nowhere am I stating/telling BOINC/WCG to use GPUs...in fact, there isn't even a setting in the GUI to enable/disable GPU globally. 6)I've uninstalled and reinstalled BOINC and it was the same exact .exe installer...so there is no newer version of BOINC for me to try to cure this. Even if this is a Windows driver issue, BOINC should NOT be forcing me to use GPUs...that's BOINC's poor project/programming decision. I've been using BOINC since it came out and I've never been forced (until now it seems) to use GPU. I can't stress enough the fact that it worked for 9 months and then it just died. The only possible explanations for that are: a)an MS patch in June somehow messed up something that BOINC now dislikes or b)BOINC is somehow now doing something that it wasn't for the past 9 months...whether that be corrupt WU downloads or poor programming or now downloading GPU-only WUs that it never asked for before. It's quite an annoying problem.  |
||
|
|
ericinboston
Senior Cruncher Joined: Jan 12, 2010 Post Count: 265 Status: Offline Project Badges:
|
What happens if you tick the box to allow GPU use when active? Else, over to the HW guys. P.S. What's with 10% memory use when active? Is the GPU using the System RAM rather than having it's own memory? Just wonder if that 10% setting has impact on the whole i.e. if CPU tasks use the 10%, what's left for the GPU, which anyway is suspended during user input. Edit: This looks like 10% of 10%... 1% of 64GB it says the system has, or? 09-Jul-2012 18:10:26 [---] max memory usage when active: 655.25MB 09-Jul-2012 18:10:26 [---] max memory usage when idle: 6552.47MB 1)I've never played with the Memory settings...but it was odd it was set at 1% and 10%...I set them to 20% and 30% because the system has so much memory...and typically it crunches at night when nobody is using the box. 2)Yes, the box has 64GB RAM. 3)I have no idea if the GPU is using the system RAM but even if it was I can't imagine it using more than 1GB. 4)As I mentioned in #1, I set the memory to 20% when in use and 30% when idle and restarted BOINC...the error messages disappeared but it still won't grab new work. 5)After #4 I then enabled GPU while the machine is in use and restarted BOINC...error messages are still gone but it won't grab work. New lines from log file after #4 and 5 above: 09-Jul-2012 22:24:11 [---] General prefs: from http://boinc.bakerlab.org/rosetta/ (last modified 18-Sep-2007 22:28:15) 09-Jul-2012 22:24:11 [---] Host location: none 09-Jul-2012 22:24:11 [---] General prefs: using your defaults 09-Jul-2012 22:24:11 [---] Reading preferences override file 09-Jul-2012 22:24:11 [---] Preferences: 09-Jul-2012 22:24:11 [---] max memory usage when active: 13104.94MB 09-Jul-2012 22:24:11 [---] max memory usage when idle: 19657.41MB 09-Jul-2012 22:24:11 [---] max disk usage: 10.00GB 09-Jul-2012 22:24:11 [---] don't use GPU while active 09-Jul-2012 22:24:11 [---] suspend work if non-BOINC CPU load exceeds 25 % 09-Jul-2012 22:24:11 [---] (to change preferences, visit the web site of an attached project, or select Preferences in the Manager) 09-Jul-2012 22:24:27 [World Community Grid] update requested by user 09-Jul-2012 22:24:32 [World Community Grid] Sending scheduler request: Requested by user. 09-Jul-2012 22:24:32 [World Community Grid] Not reporting or requesting tasks 09-Jul-2012 22:24:33 [World Community Grid] Scheduler request completed 09-Jul-2012 22:25:01 [---] General prefs: from http://boinc.bakerlab.org/rosetta/ (last modified 18-Sep-2007 22:28:15) 09-Jul-2012 22:25:01 [---] Host location: none 09-Jul-2012 22:25:01 [---] General prefs: using your defaults 09-Jul-2012 22:25:01 [---] Reading preferences override file 09-Jul-2012 22:25:01 [---] Preferences: 09-Jul-2012 22:25:01 [---] max memory usage when active: 13104.94MB 09-Jul-2012 22:25:01 [---] max memory usage when idle: 19657.41MB 09-Jul-2012 22:25:01 [---] max disk usage: 10.00GB 09-Jul-2012 22:25:01 [---] suspend work if non-BOINC CPU load exceeds 25 % 09-Jul-2012 22:25:01 [---] (to change preferences, visit the web site of an attached project, or select Preferences in the Manager) 09-Jul-2012 22:25:12 [World Community Grid] update requested by user 09-Jul-2012 22:25:14 [World Community Grid] Sending scheduler request: Requested by user. 09-Jul-2012 22:25:14 [World Community Grid] Not reporting or requesting tasks 09-Jul-2012 22:25:15 [World Community Grid] Scheduler request completed |
||
|
|
ericinboston
Senior Cruncher Joined: Jan 12, 2010 Post Count: 265 Status: Offline Project Badges:
|
@all...I installed a tool called CPUID and it cannot determine what the video card is....so maybe the video card broke and is the root of why BOINC cannot do GPU anymore. Fine. I get it.
----------------------------------------But HOW DO I DISABLE BOINC FROM ATTEMPTING TO PROCESS GPU IN THE FIRST PLACE? If my video card can't do GPU, fine...I'd still like BOINC to do non-GPU work. BOINC really should make the whole GPU topic extremely easy to enable/disable in BOINC...and if some Project only does GPU work then BOINC should throw up in your face "hey, we can't give you any work because you can't do GPU work"...put it in laymen's terms and also in the log file BOINC writes. Thanks in advance if you know how to disable GPU work. [Edit 1 times, last edit by ericinboston at Jul 10, 2012 3:09:30 AM] |
||
|
|
ericinboston
Senior Cruncher Joined: Jan 12, 2010 Post Count: 265 Status: Offline Project Badges:
|
My concern is under device manager rather then BOINC. If your video card is not showing up as anything but Standard VGA Adapter, it isn't going to work with GPU. I'm glad it worked for you for 9 months, but please don't dismiss our suggestions as if hardware doesn't fail. I have never ignored the fact that the video card may be dead...I'm an IT guy and been working on computers since 1981. :) I would never ignore people's statements about checking for HW failures. But I am far away and there is no way for me to physically check. I can't even ask someone to go look. It's all Remote Desktop or nothing. :( Device Manager says everyone is happy. CPUID cannot determine the video card. My guess is the card is now dead but yet somehow still working partially. I rebooted the system twice over the weekend and those reboot messages would have been in the log file(s) I pasted in today. But as I mentioned in my last reply, BOINC should (or this is my feature request) have some way of really spelling out when GPU is enabled/disabled, what happens if GPU is disabled but a Project requires GPU, etc. The root of my problem may be a bad video card...but it's poor programming on BOINC's side for not telling me, the end user, what is wrong AND providing some kind of formal support for such error messages and/or options to avoid the errors in the first place (like disable/enable swtiches in the Preferences). Is it really that hard to have an option stating "Use GPU? Yes/No"? Seriously. Yet no such option exists. And if you set it to No and a project wants/demands GPU, a friendly "you need to enable GPU in order to use this Project" message comes up. Pretty straightforward if you ask me. :) [Edit 1 times, last edit by ericinboston at Jul 10, 2012 3:04:43 AM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
If you haven't attached to a GPU project through BOINC, and you haven't checked the machine's WCG profile option for GPU work, and if you haven't set the machine's WCG profile to allow beta testing, then you most certainly are NOT using whatever GPU is on this machine.
Additionally, my understanding of the behavior of the BOINC client is that every time a connection is made via MS RDP, [---] GPUs have become unusable; disabling tasks will be written to the BOINC event/message log. Every time for everybody [running Windows], not just you. This occurs because RDP unloads whatever video driver is in use and replaces it with a special driver to support the RDP connection. The computer's video device is temporarily disabled and the BOINC client detects this change. The message is generated regardless of the type of video device on the machine (or whether or not the device is a "usable" GPU). When the RDP connection is closed, the RDP driver is unloaded and the video device driver is re-loaded. The device becomes available again, and when the BOINC client detects this change, [---] GPUs have become usable; enabling tasks is written to the BOINC event/message log. Again, it doesn't matter what video hardware is installed on the machine. Bottom line: I don't believe your issue has anything to do with whatever graphics hardware/drivers may or may not be installed on the machine. Unless you have in fact attached to a GPU project (or have been participating in the recent WCG GPU betas), it's damn near impossible this is GPU related. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
OK, after a nights sleep.
1) There's a new option since a little while [2 weeks maybe] on the device profiles: Allow research to run on my CPU? Yes (Default) I'm hoping the option has gone somehow No. (Bug). Please alternate that option and save with yes. 2) Of course you've checked the Project tab of BOINC Manager, selected WCG line, and seen button on left that says "No New Work". If there's a button "Allow new work", hit it. On this, I've requested and got a mod in for the current latest test client, to actually print in the message log why work is not being fetched [there are now 10 conditional messages]. When a user suspended [on the client], it prints e.g.: 15035 World Community Grid 10-7-2012 8:43:53 update requested by user 15036 World Community Grid 10-7-2012 8:43:54 [sched_op] Starting scheduler request 15037 World Community Grid 10-7-2012 8:43:54 Sending scheduler request: Requested by user. 15038 World Community Grid 10-7-2012 8:43:54 Not requesting tasks: "no new tasks" requested via Manager 15039 World Community Grid 10-7-2012 8:43:54 [sched_op] CPU work request: 0.00 seconds; 0.00 devices 15040 World Community Grid 10-7-2012 8:43:58 Scheduler request completed 15041 World Community Grid 10-7-2012 8:43:58 [sched_op] Server version 700 3) Time fraction counters are messed up. Don't know what the normal "connect about every ..." and "Additional buffer" are set at, but propose you set first for the moment to 0.00 and second to 2.00 days 4) On 3) above, also open the client_state.xml and post the time keep section here as sample: <time_stats> <on_frac>0.994327</on_frac> <connected_frac>0.604166</connected_frac> <active_frac>0.999801</active_frac> <gpu_active_frac>0.999801</gpu_active_frac> <last_update>1341903122.957939</last_update> </time_stats> 5) To debug 3) above, there's is 2 log flags that can be added. a) <sched_op_debug>1</sched_op_debug> The one that I've got always on for enhanced log info. b) <work_fetch_debug>1</work_fetch_debug> which produces this: 15241 World Community Grid 10-7-2012 8:55:21 [work_fetch] CPU: fetch share 0.000 rsc backoff (dt 0.00, inc 0.00) Let us know |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
SMTB1963,
----------------------------------------No, dont have that "[---] GPUs have become unusable; disabling tasks" line. First time seen it in ericboston's problem report. ericinboston, In response to that capitalized line [I tend to ignore such text, thoroughly], it's done with the <no_gpus>1</no_gpus> in the cc_config.xml <options> section. You said you fixed it (with 0). [Edit 1 times, last edit by Former Member at Jul 10, 2012 7:16:39 AM] |
||
|
|
Coleslaw
Veteran Cruncher USA Joined: Mar 29, 2007 Post Count: 1343 Status: Offline Project Badges: 
|
ericinboston I agree that BOINC should be easier for disabling GPU use. Currently the option is to log into the website preferences for whatever project you support to disable receiving work or to possibly edit the cconfig file. As far as the card currently in the machine, nVidia has a link to try to auto detect the card you have if it is nVidia. I don't have time to check this AM, but I think AMD does as well. I am not familiar if there is problems with auto detect while RDM'ing though because I prefer Logmein.com. It doesn't have as much trouble with BOINC. I know you said standard Windows driver...but actually you said Windows Standard Graphics which you would obiously know is different because Windows also has standard drivers for specific cards. Maybe just miscommunication. From what you describe, it is not an identified card. So, I would try the auto detect to reinstall it or, just leave it alone and not computer with GPU at all. There are however settings in the WCG website preferences where you can disable this option. I hope this helps you further and please keep us updated.
----------------------------------------Edit: And before someone else jumps in, suggesting the GPU feature in BOINC wont get far here since WCG doesn't program BOINC. You would need to head over to Berkeley's support forums and give them a shout out. Thanks      [Edit 1 times, last edit by Coleslaw at Jul 10, 2012 11:35:10 AM] |
|||
|
||||
|
[GPU Force] Robert 7NBI
Cruncher Joined: Apr 25, 2011 Post Count: 17 Status: Offline |
ericinboston I agree that BOINC should be easier for disabling GPU use. Currently the option is to log into the website preferences for whatever project you support to disable receiving work or to possibly edit the cconfig file... Edit cc_config is the best choice, but also in BM/BT you can set "gpu activity" to "never".[Edit 1 times, last edit by [GPU Force] Robert 7NBI at Jul 10, 2012 12:01:27 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
SMTB1963, No, dont have that "[---] GPUs have become unusable; disabling tasks" line. First time seen it in ericboston's problem report. Well Seke, here you go: 7/10/2012 8:06:38 AM | GPUGRID | Started upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_3 7/10/2012 8:06:40 AM | GPUGRID | Finished upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_2 7/10/2012 8:06:40 AM | GPUGRID | Started upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_4 7/10/2012 8:06:55 AM | GPUGRID | Finished upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_3 7/10/2012 8:06:55 AM | GPUGRID | Started upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_7 7/10/2012 8:06:57 AM | GPUGRID | Finished upload of I4R4-NATHAN_RPS1120528-27-166-RND5865_1_7 7/10/2012 8:07:10 AM | | GPUs have become unusable; disabling tasks 7/10/2012 8:14:47 AM | | GPUs have become usable; enabling tasks 7/10/2012 8:14:55 AM | GPUGRID | task I2R163-NATHAN_RPS1120528-19-166-RND0832_1 resumed by user 7/10/2012 8:14:56 AM | GPUGRID | Starting task I2R163-NATHAN_RPS1120528-19-166-RND0832_1 using acemdlong version 616 (cuda31) in slot 8 Now you've seen it again!  All I did was connect/disconnect to my wife's machine via RDP. Her machine is Win7x64 running BOINC 7.0.28. I can assure you that the same thing would happen if it was running 6.10.58, because I experienced the same thing back when I used that client version. I could RDP into my other boxes and post the logs here to show that these messages occur on them as well, but I think I'll let you give it a try yourself.  I would note that "disabling tasks" in the log above doesn't mean the same thing as "computation suspended" (or anything of that nature). All the WCG CPU tasks continued to run on my wife's box during the RDP connection. [Edit 1 times, last edit by Former Member at Jul 10, 2012 1:38:48 PM] |
||
|
|
|