Just tried to crunch some cuda tasks...bad luck, after some minutes i had a short blue screen ( seconds ) and then a reboot. tried it with different driver, always the same, reboot. Slowly i get the feeling that it is not worth to try crunching with a gt240.
Edith
this is only a comment, i stoped chrunching einstein on the gpu and the system is running well.
Just tried to crunch some cuda tasks...bad luck, after some minutes i had a short blue screen ( seconds ) and then a reboot. tried it with different driver, always the same, reboot. Slowly i get the feeling that it is not worth to try crunching with a gt240.
Edith
this is only a comment, i stoped chrunching einstein on the gpu and the system is running well.
I have a gt240 on one of my computers, running under Vista. Crunching BRP3CUDA needs 145 minutes with a GPU load of ~75%. Using an App_info file and running two BRP3CUDA tasks in parallel needs 260 minutes for this 2 tasks (~25 minutes faster as for crunching in sequence). GPU load is 88% - 99%. It works fine
Just tried to crunch some cuda tasks...bad luck, after some minutes i had a short blue screen ( seconds ) and then a reboot. tried it with different driver, always the same, reboot. Slowly i get the feeling that it is not worth to try crunching with a gt240.
Edith
this is only a comment, i stoped chrunching einstein on the gpu and the system is running well.
I have a machine with 3 of them installed. Take about 2 hours 10 mins on each. Working a treat with 3 going at a time (ie one per GPU). I'm running Win7 x64 and nvidia 260.99 drivers.
Do we already have some performance numbers for the new Einstein CUDA app? Something where we can deduce which cards are efficient and which can not use their entire power? So we could recommend people buying the right stuff.
So far my impression is that high end cards are fast, but not in proportion to the advantage they'd have in other projects over slower cards. However, the old G92-generation of cards seems to be doing fine (e.g. in contrast to GPU-Grid, where they've become practically useless).
Do we already have some performance numbers for the new Einstein CUDA app? Something where we can deduce which cards are efficient and which can not use their entire power? So we could recommend people buying the right stuff.
So far my impression is that high end cards are fast, but not in proportion to the advantage they'd have in other projects over slower cards. However, the old G92-generation of cards seems to be doing fine (e.g. in contrast to GPU-Grid, where they've become practically useless).
MrS
The older cards including 2xx and 8800 seem to get loaded down fairly well per GPU. I see 60-75% GPU load running one WU on a 295 and 8800GT. The Fermi cards are loaded down far less to around 35-45% and possibly a bit higher with the latest drivers. Running three work units at once on Fermi provided enough memory is available is optimal for these cards. This brings the GPU usage up to 70-80%.
I just have some observations concerning scaling. I have a GTX460 used in my and also in my i7 2600K @ 4.6GHz.
2600K (nHT): 2600s gpu time, 1000s cpu time 64% gpu usage
2600K (HT) : 3000s gpu time, 1300s cpu time 57% gpu usage
E8400 : 4200s gpu time, 1700s cpu time 66% gpu usage
Driver was the same (260.99). So why is the gpu usage higher on the slower system?
@ astro-marwil : Referring to
)
@ astro-marwil : Referring to a thread you post this reaction in seems quite useless to me.
If you are farther interested in this topic look for this thread.
Oh, sorry! Don´t know, how
)
Oh, sorry! Don´t know, how this came in.
Here and here are the threads.
Sorry again and kind regards
Martin
Just tried to crunch some
)
Just tried to crunch some cuda tasks...bad luck, after some minutes i had a short blue screen ( seconds ) and then a reboot. tried it with different driver, always the same, reboot. Slowly i get the feeling that it is not worth to try crunching with a gt240.
Edith
this is only a comment, i stoped chrunching einstein on the gpu and the system is running well.
RE: Just tried to crunch
)
I have a gt240 on one of my computers, running under Vista. Crunching BRP3CUDA needs 145 minutes with a GPU load of ~75%. Using an App_info file and running two BRP3CUDA tasks in parallel needs 260 minutes for this 2 tasks (~25 minutes faster as for crunching in sequence). GPU load is 88% - 99%. It works fine
RE: Just tried to crunch
)
I have a machine with 3 of them installed. Take about 2 hours 10 mins on each. Working a treat with 3 going at a time (ie one per GPU). I'm running Win7 x64 and nvidia 260.99 drivers.
BOINC blog
Do we already have some
)
Do we already have some performance numbers for the new Einstein CUDA app? Something where we can deduce which cards are efficient and which can not use their entire power? So we could recommend people buying the right stuff.
So far my impression is that high end cards are fast, but not in proportion to the advantage they'd have in other projects over slower cards. However, the old G92-generation of cards seems to be doing fine (e.g. in contrast to GPU-Grid, where they've become practically useless).
MrS
Scanning for our furry friends since Jan 2002
RE: Do we already have some
)
The older cards including 2xx and 8800 seem to get loaded down fairly well per GPU. I see 60-75% GPU load running one WU on a 295 and 8800GT. The Fermi cards are loaded down far less to around 35-45% and possibly a bit higher with the latest drivers. Running three work units at once on Fermi provided enough memory is available is optimal for these cards. This brings the GPU usage up to 70-80%.
"Completed, marked as
)
"Completed, marked as invalid" is what I got for several WU's running multiple WU's on one gpu.
Furthermore I got some 30 WU's with "Completed, control not conclusive".
Not sure this translation from Dutch to English is right.
Any info on this ?
See here for inconclusive
)
See here for inconclusive results:
http://einsteinathome.org/node/195567
CU
HBE
I just have some observations
)
I just have some observations concerning scaling. I have a GTX460 used in my and also in my i7 2600K @ 4.6GHz.
2600K (nHT): 2600s gpu time, 1000s cpu time 64% gpu usage
2600K (HT) : 3000s gpu time, 1300s cpu time 57% gpu usage
E8400 : 4200s gpu time, 1700s cpu time 66% gpu usage
Driver was the same (260.99). So why is the gpu usage higher on the slower system?