That's the way I read the error as well but didn't want to assume -- also wanted to make sure it was reported in case there was an error in the Work Unit assignment algorithms that should have been able to map work to memory configuration of a specific GPU.
I'll stop processing work and monitor updates for any enhancements that will provide a better match between work units and GPU configuration.
Linux: Intel(R) Xeon(R) CPU X5472 @ 3.00GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1646 runtime 45000 - 61000 sec
Windows:Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1607 runtime 7100 - 8800 sec
hmm... .
I would expect ~ 8000sec for juniper radeon
After several failed attempts to get BOINC 7.0.2 (DEV) working under Ubuntu 11.04 and 11.10, I have returned an Albert OpenCL result under Win7-64.
The GPGPU was a HD 5770, latest AMD driver version available 11.12, BOINC client is development version 7.0.2
The standard 500 coblestones was awarded.
Run time was 30,744.65 seconds
CPU time was 1,304.54 seconds
So I make that just over 9 points per 10 minute period.
Barely better than a CPU but using bucket-loads more power.
The completion time is not quite 1 quarter of one of CPU-core's completion time on the same machine.
But the CPU has 6 cores and so can still beat the snot out of the OpenCL performance on the HD 5770.
Not really worth it then. Well, not with the current AMD Catalyst driver anyway.
Once I get BOINC 7.0.2 working under Linux_AMD64 I should be able to test the different Catalyst drivers. But until then, this is the best I can do.
I saw an intriguing trend in the ATI HD 5770 WUs.
the first finished at 33k+ seconds
the second finished at 28K+seconds
the third finished at 11k+ seconds
I'm waiting on the fourth to see where this is going.
Please, look at this
)
Please, look at this WU.
http://albertathome.org/workunit/12832
What's wrong? Why Ati open'cl app made invalid result?
ATI driver 11.8
I ran that one with Catalysts
)
I ran that one with Catalysts 11.7; also wonder why we didn't match.
iMac OS X 10.7.2/7.0.2 BOINC
)
iMac OS X 10.7.2/7.0.2 BOINC Client
all ATI WUs failing with "Error while computing" status.
Sample Stderr Output:
http://albertathome.org/task/57408
http://albertathome.org/task/51158
--
Jon
As the error messages
)
As the error messages indicate, your GPU doesn't have enough free memory. Sorry but we can't reliably check this in advance right now...
Oliver
That's the way I read the
)
That's the way I read the error as well but didn't want to assume -- also wanted to make sure it was reported in case there was an error in the Work Unit assignment algorithms that should have been able to map work to memory configuration of a specific GPU.
I'll stop processing work and monitor updates for any enhancements that will provide a better match between work units and GPU configuration.
Thanks for the feedback and explanation.
--
Jon
Same Codebase different
)
Same Codebase different platform / Compiler .. ?
Linux: Intel(R) Xeon(R) CPU X5472 @ 3.00GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1646 runtime 45000 - 61000 sec
Windows:Intel(R) Core(TM)2 Quad CPU Q6600 @ 2.40GHz [2] AMD ATI Radeon HD 5700 series (Juniper) (1024MB) driver: 1.4.1607 runtime 7100 - 8800 sec
hmm... .
I would expect ~ 8000sec for juniper radeon
The successors (cypress, cayman) looking good.
After several failed attempts
)
After several failed attempts to get BOINC 7.0.2 (DEV) working under Ubuntu 11.04 and 11.10, I have returned an Albert OpenCL result under Win7-64.
The GPGPU was a HD 5770, latest AMD driver version available 11.12, BOINC client is development version 7.0.2
The standard 500 coblestones was awarded.
Run time was 30,744.65 seconds
CPU time was 1,304.54 seconds
So I make that just over 9 points per 10 minute period.
Barely better than a CPU but using bucket-loads more power.
The completion time is not quite 1 quarter of one of CPU-core's completion time on the same machine.
But the CPU has 6 cores and so can still beat the snot out of the OpenCL performance on the HD 5770.
Not really worth it then. Well, not with the current AMD Catalyst driver anyway.
Once I get BOINC 7.0.2 working under Linux_AMD64 I should be able to test the different Catalyst drivers. But until then, this is the best I can do.
NEW 12.1 drivers fixed the
)
NEW 12.1 drivers fixed the HD6990 "high CPU usage" problem... work great!
http://support.amd.com/us/kbarticles/Pages/Catalyst121Previewdriver.aspx
9-)
I saw an intriguing trend in
)
I saw an intriguing trend in the ATI HD 5770 WUs.
the first finished at 33k+ seconds
the second finished at 28K+seconds
the third finished at 11k+ seconds
I'm waiting on the fourth to see where this is going.
RE: NEW 12.1 drivers fixed
)
This is great news, finally! Can others confirm that, too? I'll update our test rigs to check it myself...
Oliver