If this app's still heavily cpu limited with a highend card, has anyone done any testing from the other side to see how far down the GPU foodchain you can go before it becomes teh bottleneck?
People at GPUGrid are reporting errors with the GTX260 and 190.38 drivers (one even narrowed it down to the 192 version, has a 216 which works OK). Adding W7 and x64 to the mix adds to the probability that it's a driver issue.
Hi Richard!
I have had no problem running CUDAs concurrently: gpugrid, seti and aqua, all under 190.38. At one time I thought aqua and seti could not co-exist but that somehow got solved after a reboot.
I did have a problem installing 190.38 on windows 7 (9800gtx+) and Vista-64 (gtx280) with the "System has not been modified" symptom as described here
I do not have a gtx260 so I can only speculate on that problem. I have 1 gpu in each of three 64 bit PCIe systems and have adequate cooling with no game playing. Some of the complaints I read about seem to be systems with overclocking and overheating (2 or 3 gpu's per system)
It would appear that E@H does not yet have a CUDA version that runs on a 64 bit OS. Also, if the E@H program is named blah-blah-blah-intelx86.zip I would think it would be designed only for Intel and not AMD much like any other optimized 3rd party app nameing scheme. (my 2c)
I am willing to run some E@H 64bit cuda tasks if any are available but right now I am back to seti and gpugrid. aqua cuda does not pay enough and I never got E@H to work.
Ok, just slotted in this system for testing. Specs:
Intel i7 ( 920 @ 2.67GHz )
6GB DDR3 Corsair
Windows 7 ( release candidate, 64 bit )
NVidia GForce 9800GT with 1GB RAM on PCI EX ( driver version 8.15.11.8618 )
Now has 8 tasks running ie. claiming a virtual core each. All ABP 3.08, fine so far.
And the graphics are fine too.
Cheers, Mike.
( edit ) Extra GForce driver detail, post here so I don't lose it, as identified by BOINC Manager - driver version 18618, compute capability 1.1, 1024 MB, est. 60GFlops
( edit ) Ignore the 8 aborts, I bumped the wrong button! :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Now has 8 tasks running ie. claiming a virtual core each. All ABP 3.08, fine so far.
So you're not actually testing/using the CUDA card at all? ;-)
If you use that hybrid app_info.xml file I posted a while back, you could replace one of the 3.08s with a 3.07 - might shave an hour off the total time for all eight. (Early days, small steps.....)
Now has 8 tasks running ie. claiming a virtual core each. All ABP 3.08, fine so far.
So you're not actually testing/using the CUDA card at all? ;-)
If you use that hybrid app_info.xml file I posted a while back, you could replace one of the 3.08s with a 3.07 - might shave an hour off the total time for all eight. (Early days, small steps.....)
Sorry, confused ( a simple skill ) .... that's the app_info.xml in this post?
.
.
.
Ok, got that one in place now ......
So, I now have 2 x einsteinbinary_ABP 3.07 (cuda) running
some downloading .....
and PLOP PLOP, the cuda's have just fallen over .....
and the server just told me I've reached my daily quota of 40 jobs. Wot wif all dis stuffin' about .... :-)
Cheers, Mike.
( edit ) Fell over due to 'Error during CUDA host->device data transfer (unknown error)' as reported in these tasks here and here. ( Sorry I forgot to mention BOINC 6.6.36 )
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Sorry, confused ( a simple skill ) .... that's the app_info.xml in this post?
Yes, that's the one.
Quote:
and PLOP, the cuda's have just fallen over .....
Now you're testing!
Card, driver, BOINC version reports all look good. But you shouldn't have been running two CUDA tasks at the same time on that card - were they truly concurrent, or did the second start when the first failed? (May need to check BOINC log).
Most likely problem is getting a decent CUDA driver for W7_RC_64. During W7_Beta, people were having best success with drivers from Windows Update: but I stick to boring old XP_32 here, so best wait until one of the adventurous types piches in to advise.
If this app's still heavily
)
If this app's still heavily cpu limited with a highend card, has anyone done any testing from the other side to see how far down the GPU foodchain you can go before it becomes teh bottleneck?
RE: Richard Haselgrove
)
Hi Richard!
I have had no problem running CUDAs concurrently: gpugrid, seti and aqua, all under 190.38. At one time I thought aqua and seti could not co-exist but that somehow got solved after a reboot.
I did have a problem installing 190.38 on windows 7 (9800gtx+) and Vista-64 (gtx280) with the "System has not been modified" symptom as described here
I do not have a gtx260 so I can only speculate on that problem. I have 1 gpu in each of three 64 bit PCIe systems and have adequate cooling with no game playing. Some of the complaints I read about seem to be systems with overclocking and overheating (2 or 3 gpu's per system)
It would appear that E@H does not yet have a CUDA version that runs on a 64 bit OS. Also, if the E@H program is named blah-blah-blah-intelx86.zip I would think it would be designed only for Intel and not AMD much like any other optimized 3rd party app nameing scheme. (my 2c)
I am willing to run some E@H 64bit cuda tasks if any are available but right now I am back to seti and gpugrid. aqua cuda does not pay enough and I never got E@H to work.
Is it just me, or does the
)
Is it just me, or does the 'this thread' link in the 'MS Windows CUDA' section on this page point to the Linux CUDA thread? :-)
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Is it just me, or does
)
It looks OK to me.
Tullio
No, you're right. Sorry.
Ok, just slotted in this
)
Ok, just slotted in this system for testing. Specs:
Intel i7 ( 920 @ 2.67GHz )
6GB DDR3 Corsair
Windows 7 ( release candidate, 64 bit )
NVidia GForce 9800GT with 1GB RAM on PCI EX ( driver version 8.15.11.8618 )
Now has 8 tasks running ie. claiming a virtual core each. All ABP 3.08, fine so far.
And the graphics are fine too.
Cheers, Mike.
( edit ) Extra GForce driver detail, post here so I don't lose it, as identified by BOINC Manager - driver version 18618, compute capability 1.1, 1024 MB, est. 60GFlops
( edit ) Ignore the 8 aborts, I bumped the wrong button! :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Is it just me, or does
)
My fault. Fixed.
BM
BM
RE: Now has 8 tasks running
)
So you're not actually testing/using the CUDA card at all? ;-)
If you use that hybrid app_info.xml file I posted a while back, you could replace one of the 3.08s with a 3.07 - might shave an hour off the total time for all eight. (Early days, small steps.....)
RE: RE: Now has 8 tasks
)
Sorry, confused ( a simple skill ) .... that's the app_info.xml in this post?
.
.
.
Ok, got that one in place now ......
So, I now have 2 x einsteinbinary_ABP 3.07 (cuda) running
some downloading .....
and PLOP PLOP, the cuda's have just fallen over .....
and the server just told me I've reached my daily quota of 40 jobs. Wot wif all dis stuffin' about .... :-)
Cheers, Mike.
( edit ) Fell over due to 'Error during CUDA host->device data transfer (unknown error)' as reported in these tasks here and here. ( Sorry I forgot to mention BOINC 6.6.36 )
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: Sorry, confused ( a
)
Yes, that's the one.
Now you're testing!
Card, driver, BOINC version reports all look good. But you shouldn't have been running two CUDA tasks at the same time on that card - were they truly concurrent, or did the second start when the first failed? (May need to check BOINC log).
Most likely problem is getting a decent CUDA driver for W7_RC_64. During W7_Beta, people were having best success with drivers from Windows Update: but I stick to boring old XP_32 here, so best wait until one of the adventurous types piches in to advise.
OK ...... BOINC
)
OK ...... BOINC log
So, that sez ( ?? )
- started blurb_451_1
- started blurb_476_1
- blurb_476_1 falls over
- blurb_451_1 falls over
..... concurrent.
Cheers, Mike.
( edit ) Went to Win Update and got a new Nvidia driver, now version 8.15.11.8635. Will wait for the the server to forgive me now ..... :-)
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal