[HOST#12161179] MSG(high) Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
20 Dec 2016 4:39:20 UTC
Topic 203900
(moderation:
Linux PC has stopped running GPU jobs. PC is running a GTX 650 Ti Boost. It's been running GPU jobs for years and now has stopped. Has something changed in E@H that no long use my GPU type. Below is a portion of the logs.
Thanks,
Steve
2016-12-20 04:27:34.9568 [PID=19857] [version] Checking plan class 'FGRPopencl-ati' 2016-12-20 04:27:34.9568 [PID=19857] [version] parsed project prefs setting 'gpu_util_fgrp': 0.500000 2016-12-20 04:27:34.9568 [PID=19857] [version] No ATI devices found 2016-12-20 04:27:34.9568 [PID=19857] [version] Checking plan class 'FGRPopencl-nvidia' 2016-12-20 04:27:34.9568 [PID=19857] [version] parsed project prefs setting 'gpu_util_fgrp': 0.500000 2016-12-20 04:27:34.9569 [PID=19857] [version] OpenCL GPU RAM required min: 1071644672.000000, supplied: 1027211264 2016-12-20 04:27:34.9569 [PID=19857] [version] Checking plan class 'FGRPopencl-Beta-ati' 2016-12-20 04:27:34.9569 [PID=19857] [version] beta test app versions not allowed in project prefs. 2016-12-20 04:27:34.9569 [PID=19857] [version] Checking plan class 'FGRPopencl-Beta-nvidia' 2016-12-20 04:27:34.9569 [PID=19857] [version] beta test app versions not allowed in project prefs. 2016-12-20 04:27:34.9569 [PID=19857] [version] no app version available: APP#40 (hsgamma_FGRPB1G) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0 2016-12-20 04:27:34.9569 [PID=19857] [version] no app version available: APP#40 (hsgamma_FGRPB1G) PLATFORM#1 (i686-pc-linux-gnu) min_version 0 2016-12-20 04:27:34.9626 [PID=19857] [debug] [HOST#12161179] MSG(high) No work sent 2016-12-20 04:27:34.9626 [PID=19857] [debug] [HOST#12161179] MSG(high) Binary Radio Pulsar Search (Arecibo) is not available for your type of computer. 2016-12-20 04:27:34.9626 [PID=19857] [debug] [HOST#12161179] MSG(high) see scheduler log messages on https://einsteinathome.org/host/12161179/log 2016-12-20 04:27:34.9626 [PID=19857] Sending reply to [HOST#12161179]: 0 results, delay req 60.00 2016-12-20 04:27:34.9627 [PID=19857] Scheduler ran 0.046 seconds
Arecibo GPU tasks are not
)
Arecibo GPU tasks are not available anymore or not at the moment at least.
'Gamma-ray pulsar binary search #1 on GPUs' is replacing it.
https://einsteinathome.org/apps.php
https://einsteinathome.org/server_status.html
Problem seems to be your card is able to provide about 50MB less RAM than what this new application requires:
You could try if the newest beta version would run. For that you will need to allow "Run test applications" at the Project Preferences: https://einsteinathome.org/account/prefs/project
I updated the NVIDIA Linux
)
I updated the NVIDIA Linux 64-Bit Driver to Version 375.26, rebooted the PC and restarted the boinc-client and all is well, the PC is crunching GPU jobs again.
Thanks,
Steve
Host completed a couple of
)
Host completed a couple of jobs and is now back to idle with original problem. May have to leave E@H and find projects that can use my GPU.
Thanks,
Steve
SJC_Steve wrote:Host
)
Please realise that the "couple of jobs" were all BRP4G 'resends' - copies of BRP4G tasks that had failed on other computers that were given to your computer simply because it was available and asking at just the right time - a chance event.
You didn't solve the original problem - not quite enough GPU memory for the new FGRPB1G tasks. However, there might be some hope. There are two particular lines from the most recent server contact log and I'll list them here. Note that to improve readability I've trimmed some unnecessary stuff.
The first line shows that you have set 0.5 as the GPU utilization factor for FGRP type GPU apps. The second line shows the failure of the memory requirement. Looking at your computers list, it shows BOINC detecting 978MB of GPU RAM which I guess is why your GPU is failing the test. However, because of the lack of RAM, you shouldn't be trying to run 2 concurrent GPU tasks of the FGRPB1G variety and I wonder if you set the GPU utilization factor back to the default 1.0, you might be able to get work for your GPU. It's certainly worth a try.
In any case, I don't believe the project would be too keen to exclude 1GB GPUs. I would think it very likely that the RAM test would be tweaked at some point to allow GPUs like yours to contribute.
Cheers,
Gary.
Gary, Thanks for the ideas.
)
Gary,
Thanks for the ideas. I set my GPU utilization back to 1.00, unfortunately no change, E@H still claims that no work for my type of PC. I guess I'll have to wait until the project makes some changes to accept a bit less GPU ram. Strange that a couple of my other PCs with similar GPU memory are still crunching fine.
Thanks,
Steve
I have the same issue with a
)
I have the same issue with a Athlon 5350 APU running Linux 3.19 and a A4-4000 with Win10 x64, respectively with 707MB and 768MB shared memory.
I tested E@H on my main desktop PC (AMD R9 380 4GB, Win10 x64), it seems that FGRPB1G work units take endlessly to complete.
Any suggestion?
Eden_H _E.R._ wrote:I tested
)
The task you tested used version 1.15 of the FGRP GPU app and that was the first release for Windows and had very bad performance, most of the calculations was run in the CPU and only a small fraction used the GPU. We are now on version 1.17 and that version has dramatically improved performance. If you feel up to it you could test one more task and I'm sure you'll find it much better.
Very well, I'll try it as
)
Very well, I'll try it as soon as I can. My APU hosts, however, are yet incapable to receive WUs (due to <1GB VRAM, I guess)... Should I expect this issue to be fixed soon?
Thanks in advance :)