I have had 1617 tasks validated and 207 tasks pending in the last week or so via one of my Linux systems with a NVIDIA card. There was one task that failed to validate yesterday but otherwise, the new OpenCL application appears to be running stable overall with Linux/NVIDIA.
I'm experiencing a some very strange bug. I've two video cards and they are working very well in Einstein@Home project. However when I try to run FGPR Opencl WUs on them I can see very strange behavior - two WU's are running on the second card simultaneously and nothing is happening on the first card. Nevertheless boinc manage shows that WUs are being processed each one on the separate cards.
I'm experiencing a some very strange bug. I've two video cards and they are working very well in Einstein@Home project. However when I try to run FGPR Opencl WUs on them I can see very strange behavior - two WU's are running on the second card simultaneously and nothing is happening on the first card. Nevertheless boinc manage shows that WUs are being processed each one on the separate cards.
BOINC are having problems with their code (server, client, API - not sure which is most relevant) in that area - identifying which card is which 'device' in OpenCL terminology. Oliver might want to look at that report in connection with the current BOINC developer discussions (mainly being initiated by Charle Fenton).
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
The application produces with 1 cpu core and 1 gpu a little bit more computation output for FGRP as if i would run it on all of my 6 cores on the fx6100@3600Mhz
and that on a gpu fluctuation on my gtx 660 OC @1150Mhz at around 30%.
It looks like i could run 2 or 3 simultaneously if i would.
Yes i know actually for testing it´s not allowed.
Specifically 2 cpu cores and 0,5 gpu or 3 cpu cores 0,33 gpu.
But with 2GB of gpu memory it´s certainly possible?
I'm experiencing a some very strange bug. I've two video cards and they are working very well in Einstein@Home project. However when I try to run FGPR Opencl WUs on them I can see very strange behavior - two WU's are running on the second card simultaneously and nothing is happening on the first card. Nevertheless boinc manage shows that WUs are being processed each one on the separate cards.
BOINC are having problems with their code (server, client, API - not sure which is most relevant) in that area - identifying which card is which 'device' in OpenCL terminology. Oliver might want to look at that report in connection with the current BOINC developer discussions (mainly being initiated by Charle Fenton).
This is either a problem in the client or a problem communicating the device to run on from the client to the application.
AFAIK we're using the exact same code in the FGRP App than in the OpenCL BRP App, and I know of no such error being reported with the latter. My guess is that it's either a bug in the OpenCL implementation in the driver (in which case using a different driver version might help) or in the client (try a different version here, too).
When two such tasks are running, could you take a look at the command lines? There should be a small integer number following a "--device" option. If this number is identical for the two tasks, then this is definitely a problem in the Client. If it is different, it may be a problem in the App code or the driver.
Command line in slot 0 ends --device 1
Command line in slot 1 ends --device 0
But GPU-Z shows
MSI Afterburner shows significant usage on what it calls GPU 1, zero usage on GPU 2.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
When two such tasks are running, could you take a look at the command lines? There should be a small integer number following a "--device" option. If this number is identical for the two tasks, then this is definitely a problem in the Client. If it is different, it may be a problem in the App code or the driver.
BM
Sorry, wasn't very informative on my previous post.
The command lines look fine:
--device 0
--device 1
Boinc 7.0.28 and GTX660Ti as fisrt card and GTX560Ti as second.
NVIDIA driver 314.07
This is just to let you know that we are working on it. It definitely is more difficult than I first thought.
It might help if someone of you having this problem could send us (or post here, stripped of any personal info you like) the init_data.xml files from the two slots of the tasks that end up running on the same device.
It might help if someone of you having this problem could send us (or post here, stripped of any personal info you like) the init_data.xml files from the two slots of the tasks that end up running on the same device.
BM
Which of these apps need to be enabled?
Binary Radio Pulsar Search: no
Binary Radio Pulsar Search (single DM): no
Binary Radio Pulsar Search (Perseus Arm Survey): yes
Gravitational Wave S6 LineVeto search (extended): no
Gamma-ray pulsar search #2: yes
I have had 1617 tasks
I have had 1617 tasks validated and 207 tasks pending in the last week or so via one of my Linux systems with a NVIDIA card. There was one task that failed to validate yesterday but otherwise, the new OpenCL application appears to be running stable overall with Linux/NVIDIA.
I'm experiencing a some very
I'm experiencing a some very strange bug. I've two video cards and they are working very well in Einstein@Home project. However when I try to run FGPR Opencl WUs on them I can see very strange behavior - two WU's are running on the second card simultaneously and nothing is happening on the first card. Nevertheless boinc manage shows that WUs are being processed each one on the separate cards.
RE: I'm experiencing a some
BOINC are having problems with their code (server, client, API - not sure which is most relevant) in that area - identifying which card is which 'device' in OpenCL terminology. Oliver might want to look at that report in connection with the current BOINC developer discussions (mainly being initiated by Charle Fenton).
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg
The application produces with
The application produces with 1 cpu core and 1 gpu a little bit more computation output for FGRP as if i would run it on all of my 6 cores on the fx6100@3600Mhz
and that on a gpu fluctuation on my gtx 660 OC @1150Mhz at around 30%.
It looks like i could run 2 or 3 simultaneously if i would.
Yes i know actually for testing it´s not allowed.
Specifically 2 cpu cores and 0,5 gpu or 3 cpu cores 0,33 gpu.
But with 2GB of gpu memory it´s certainly possible?
RE: RE: I'm experiencing
This is either a problem in the client or a problem communicating the device to run on from the client to the application.
AFAIK we're using the exact same code in the FGRP App than in the OpenCL BRP App, and I know of no such error being reported with the latter. My guess is that it's either a bug in the OpenCL implementation in the driver (in which case using a different driver version might help) or in the client (try a different version here, too).
When two such tasks are running, could you take a look at the command lines? There should be a small integer number following a "--device" option. If this number is identical for the two tasks, then this is definitely a problem in the Client. If it is different, it may be a problem in the App code or the driver.
BM
BM
I think it's a app or driver
I think it's a app or driver problem.
Testing with host 5367: two identical GTX 670, BOINC v7.0.64, driver v314.22, Windows 7
Event log says:
Command line in slot 0 ends --device 1
Command line in slot 1 ends --device 0
But GPU-Z shows
MSI Afterburner shows significant usage on what it calls GPU 1, zero usage on GPU 2.
I didn't want to spam the boards with my stats - just milestone theads - but apparently signatures are no longer optional. Follow the link if you're interested.
http://www.boincsynergy.com/images/stats/comb-3475.jpg
Thanks, that's
Thanks, that's helpful.
That's as far as I could track it down, Oliver or HB should be able to pick that up after the weekend.
BM
BM
RE: When two such tasks
Sorry, wasn't very informative on my previous post.
The command lines look fine:
--device 0
--device 1
Boinc 7.0.28 and GTX660Ti as fisrt card and GTX560Ti as second.
NVIDIA driver 314.07
Thank you.
This is just to let you know
This is just to let you know that we are working on it. It definitely is more difficult than I first thought.
It might help if someone of you having this problem could send us (or post here, stripped of any personal info you like) the init_data.xml files from the two slots of the tasks that end up running on the same device.
BM
BM
RE: It might help if
Which of these apps need to be enabled?
Binary Radio Pulsar Search: no
Binary Radio Pulsar Search (single DM): no
Binary Radio Pulsar Search (Perseus Arm Survey): yes
Gravitational Wave S6 LineVeto search (extended): no
Gamma-ray pulsar search #2: yes