I have similar problems on win xp 32 bit. Same hardware just other OS.
With 1 CPU + 1 GPU = 50% + 22% the GPU load is at 63%.
I get a progress rate of ~ 1.3%/min.
With 1 CPU + 1 CPU + 1 GPU = 48% + 48% + 2% and the GPU load drops under 20% and even down to 12%.
Which giveas a progress rate between 0.3%/min and 0.6%/min
Any official comment? Is this problem known and/or being worked on?
Well.. it's almost Christmas. At my institute not much is going to happen for the next 3 weeks or so.. ;)
Only when I set the priority of the CUDA-task in the task-manager to realtime (Echtzeit) while running two CPU-tasks, it will use the 20% CPU-time and the GPU-load goes up to the normal 53% under vista64.
This is somewhat odd. To prevent this kind of performance degrading, BOINC is supposed to give the Cuda tasks a higher priority with the operating system so that for the little CPU work that needs to be done, the app is never starved.
BOINC is setting the priorities correctly, but it looks like Windows doesn't treat these priorities as expected. Under investigation...
My system received the first BRP3 WU I have seen this morning. The first one processed in 2061 seconds, which is considerably quicker than the previous generation CUDA work unit which took around 2.36 hours via the same GPU. GPU load on one GPU (580) was 35-45% and 60-65% on the other GPU (8800).
My system received the first BRP3 WU I have seen this morning. The first one processed in 2061 seconds, which is considerably quicker than the previous generation CUDA work unit which took around 2.36 hours via the same GPU. GPU load on one GPU (580) was 35-45% and 60-65% on the other GPU (8800).
This is the kind of performance that would be expected from previous tests performed on the new app.
I noticed you are using driver 262.99 . Is anybody experiencing the problem described here with this driver version?
Only when I set the priority of the CUDA-task in the task-manager to realtime while running two CPU-tasks, it will use the 20% CPU-time and the GPU-load goes up to the normal 53% under vista64.
I confirm, manual set priority to realtime helps.
I'm running 12 CPU WUs and 2 GPUs WUs simultaneously;
i7-980X, 2x460GTX, Windows XP x64, v260.99 Nvidia driver, Boinc client v6.12.8
GPUs usage without changing priorities ~5%
After set priority to RealTime, jumps to ~50%
My first BRP3 WU completed after 25000s (on GTX460)!!!
Next WU took 6000s to complete(priority changed in the middle of a task).
It should be better now. We will see.
I haven´t got any BRP3 wu the last days, but if I remember correctly the priority *is* changed from the lowest level (S5GC) to the second lowest level (BRP3).
@mikey RE: Cuda is
)
@mikey
hotze33 has 4 GPUs 2x2
@hotze33
I got the runtime down to 6000s.
So with 500 credits per task I could get around 7000 crredits/day.
@ Developers: Does the new
)
@ Developers: Does the new app benefit from the new gpus (fermi)?
I have similar problems on
)
I have similar problems on win xp 32 bit. Same hardware just other OS.
With 1 CPU + 1 GPU = 50% + 22% the GPU load is at 63%.
I get a progress rate of ~ 1.3%/min.
With 1 CPU + 1 CPU + 1 GPU = 48% + 48% + 2% and the GPU load drops under 20% and even down to 12%.
Which giveas a progress rate between 0.3%/min and 0.6%/min
Any official comment? Is this
)
Any official comment? Is this problem known and/or being worked on?
Well.. it's almost Christmas. At my institute not much is going to happen for the next 3 weeks or so.. ;)
MrS
Scanning for our furry friends since Jan 2002
Hi, one more test. Only
)
Hi, one more test.
Only when I set the priority of the CUDA-task in the task-manager to realtime (Echtzeit) while running two CPU-tasks, it will use the 20% CPU-time and the GPU-load goes up to the normal 53% under vista64.
RE: This is somewhat odd.
)
BOINC is setting the priorities correctly, but it looks like Windows doesn't treat these priorities as expected. Under investigation...
BM
BM
My system received the first
)
My system received the first BRP3 WU I have seen this morning. The first one processed in 2061 seconds, which is considerably quicker than the previous generation CUDA work unit which took around 2.36 hours via the same GPU. GPU load on one GPU (580) was 35-45% and 60-65% on the other GPU (8800).
RE: My system received the
)
This is the kind of performance that would be expected from previous tests performed on the new app.
I noticed you are using driver 262.99 . Is anybody experiencing the problem described here with this driver version?
CU
HBE
RE: Only when I set the
)
I confirm, manual set priority to realtime helps.
I'm running 12 CPU WUs and 2 GPUs WUs simultaneously;
i7-980X, 2x460GTX, Windows XP x64, v260.99 Nvidia driver, Boinc client v6.12.8
GPUs usage without changing priorities ~5%
After set priority to RealTime, jumps to ~50%
My first BRP3 WU completed after 25000s (on GTX460)!!!
Next WU took 6000s to complete(priority changed in the middle of a task).
It should be better now. We will see.
I haven´t got any BRP3 wu
)
I haven´t got any BRP3 wu the last days, but if I remember correctly the priority *is* changed from the lowest level (S5GC) to the second lowest level (BRP3).