Message boards :
News :
BOINC Scheduler changes for multiple app version case
Message board moderation
Author | Message |
---|---|
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
BOINC Scheduler has had problems sending different app versions to clients when there's multiple possible versions for a platform. For example, this happens when there's both OpenCL and Stream/CUDA or both 32-bit and 64-bit CPU app version available. To hopefully fix this our scheduler has been changed to send each app version until it has enough host specific speed samples. Only exception is when that version has been failing. Please report any problems of getting work or having them fail more often in our forums. Thank you and happy crunching! |
Send message Joined: 6 May 11 Posts: 7 Credit: 1,719,644,730 RAC: 0 |
I have a host with two GTX1080s in http://moowrap.net/show_host_detail.php?hostid=798990 and up until a few hours ago they were both crunching the same "Distributed.net Client v1.03 (cuda31) windows_intelx86" tasks quite happily. Now I'm only getting "Distributed.net Client v1.04 (opencl_nvidia_101) windows_intelx86" tasks which all fail after about 20 seconds with "computation error" The project attempts to run one task per GPU but GPU-Z shows no load and no power used. It may have something to do with the message "[May 02 03:32:57 UTC] Automatic processor type detection did not recognize the processor (tag: "Intel(R) HD Graphics 530 ")" in the stderr file. No changes at my end, another hosts with the same GPUs is also getting the opencl tasks and running them although they are claiming to need over six hours (and increasing) to finish! I don't use an app_config. |
Send message Joined: 6 May 11 Posts: 7 Credit: 1,719,644,730 RAC: 0 |
I've let it run again and tasks now state: 17:45:15 (7524): wrapper v1.4 build 18 for nVidia OpenCL starting (BOINC Wrapper v7.5.26011) So apparently these "Distributed.net Client v1.04 (opencl_nvidia_101) windows_intelx86" tasks are running on the CPU?! |
Send message Joined: 2 Jun 17 Posts: 1 Credit: 79,989,864 RAC: 1,305 |
I have a host with two GTX1080s in http://moowrap.net/show_host_detail.php?hostid=798990 and up until a few hours ago they were both crunching the same "Distributed.net Client v1.03 (cuda31) windows_intelx86" tasks quite happily. For at least those people who use Norton Antivirus (as I do), Norton is treating the file as having a virus, blocks it and this results in a computation error when the task is run. Filename: dnetc520-win32-x86-opencl.exe Threat name: Heur.AdvML.CFull Path: d:\programdata\boinc\projects\moowrap.net\dnetc520-win32-x86-opencl.exe ____________________________ On computers as of 03/05/2018 at 09:20:11 Last Used 03/05/2018 at 09:20:11 Startup Item No Launched No Threat type: Heuristic Virus. Detection of a threat based on malware heuristics. ____________________________ dnetc520-win32-x86-opencl.exe Threat name: Heur.AdvML.C Locate Few Users Hundreds of users in the Norton Community have used this file. Mature This file was released 5 years 2 months ago. High This file risk is high. ____________________________ http://moowrap.net/download/dnetc520-win32-x86-opencl.exe Downloaded File from moowrap.net Source: External Media ____________________________ File Actions File: d:\programdata\boinc\projects\moowrap.net\ dnetc520-win32-x86-opencl.exe Blocked ____________________________ File Thumbprint - SHA: Not available File Thumbprint - MD5: Not available |
Send message Joined: 19 Dec 16 Posts: 1 Credit: 22,392,704 RAC: 0 |
Same problem here. I have Norton and it's spitting out a message that one of the downloaded files for the new version is a virus. dnetc520-win32-x86-opencl.exe I don't know if this is a false positive or not by Symantec. |
Send message Joined: 22 Jun 11 Posts: 2080 Credit: 1,844,407,912 RAC: 3,717 |
Same problem here. I have Norton and it's spitting out a message that one of the downloaded files for the new version is a virus. Set Norton to ignore the Boinc directories, any 'false positive' will be ignored and any real virus that tries to get out of the Boinc directories will be caught. 'False positives' are a problem for Boinc due to the way some projects send and receive data fairly often, mimicking a virus collecting data. |
Send message Joined: 6 May 11 Posts: 7 Credit: 1,719,644,730 RAC: 0 |
It's happened before but it's not my issue, I don't run any anti virus on my crunching machines. I have several hosts running Moo and the only ones having problems are the ones with GTX 1080 GPUs in and only with Opencl tasks, everything else runs without problems. Probably a driver problem but I'm not going to mess around with things at the moment. No matter, I'll move them to a different project. |
Send message Joined: 20 Apr 11 Posts: 388 Credit: 822,356,221 RAC: 0 |
It may have something to do with the message "[May 02 03:32:57 UTC] Automatic processor type detection did not recognize the processor (tag: "Intel(R) HD Graphics 530 ")" in the stderr file. Right, you seem to have internal Intel GPU as well and Dnet OpenCL app detects that and tries to use but fails to. There's two problems here, first that the Intel GPU fails to run the app and second that it gets detected by the Dnet app so the GPU numbering is out of sync with BOINC Client. BOINC Schduler should have given up after 10 failures on the OpenCL app but doesn't seem to be doing that. I'll try to figure out if this can be fixed (you should only be getting the cuda31 app until problems with Intel GPUs can be fixed). -w |
Send message Joined: 16 Mar 17 Posts: 1 Credit: 210,998,270 RAC: 0 |
On a GTX 1070, Windows 7 64 bit, driver 384.94, which runs various CUDA and OpenCL applications just fine, all "Distributed.net Client v1.04 (opencl_nvidia_101) windows_intelx86" tasks fail with "No OpenCL platforms available!". - edit - Nevermind. Something must have shut down uncleanly before this. I rebooted the computer, and now it's running just fine. |
Send message Joined: 5 Dec 17 Posts: 1 Credit: 66,514,992 RAC: 0 |
Same problem as Woodles, i7 CPU based GPU racks (with 1080/1080 Ti's) producing computational errors. I have only tried one rack so far with 3x 1080 but every WU failed. Our GPU racks only use the 1080/1080 Ti's for crunching, but never the IGPUs. (intel). This is the host: https://moowrap.net/show_host_detail.php?hostid=984556 I just tried running Moo on Xeon boxes and they run fine, under the same NVidia driver. So the problem does seem to be related to the i7 CPUs based Racks. |
Send message Joined: 6 May 11 Posts: 7 Credit: 1,719,644,730 RAC: 0 |
Right, you seem to have internal Intel GPU as well and Dnet OpenCL app detects that and tries to use but fails to. There's two problems here, first that the Intel GPU fails to run the app and second that it gets detected by the Dnet app so the GPU numbering is out of sync with BOINC Client. I've never used the Intel GPU for crunching so it's probably got none of the correct drivers. The task does indeed give up after ten goes of trying to use the Intel GPU ... and then moves onto the next task and repeats. I tried to download work again last night and got "191 Distributed.net Client v1.04 (opencl_nvidia_101) windows_intelx86" tasks. Some Cuda ones would be nice as the external GPU has no issues with them. |
Send message Joined: 14 Nov 17 Posts: 11 Credit: 1,961,103,615 RAC: 1,238,230 |
I don´t know if I have the same problems, but I´ve create a new thread because I haven´t seen this before. https://moowrap.net/forum_user_posts.php?userid=254811 |