Does configuring BOINC to run on an NVIDIA GPU exclude running S5xxx tasks?
As long as you keep
Use CPU yes
(enforced by 6.10+ clients)
in your Einstein@Home preferences, I see no way to exclude S5 tasks.
Gruß,
Gundolf
[edit]You could try to find hints in the server logs by clicking the "Last contact" link in the Your computers list. I can't, because your computers are hidden.[/edit]
Computer sind nicht alles im Leben. (Kleiner Scherz)
Does configuring BOINC to run on an NVIDIA GPU exclude running S5xxx tasks?
As long as you keep
Use CPU yes
(enforced by 6.10+ clients)
in your Einstein@Home preferences, I see no way to exclude S5 tasks.
Gruß,
Gundolf
[edit]You could try to find hints in the server logs by clicking the "Last contact" link in the Your computers list. I can't, because your computers are hidden.[/edit]
When I attempt to access the "Last Contact" link - All I get is:
Oops! This link appears to be broken.
There are only 10 kind of people in the world, those that understand binary and those that don't!
. . . Also, understand that "CPU utilization" and "GPU utilization" are VERY different concepts.
The CPU utilization number is the percentage of time the CPU is working on that task. 50% CPU utilization means that 50% of the time the CPU is working on that task and 50% of the time it's either doing something else, or sitting idle.
. . .
I was wondering about The CPU needs for the CUDA application.
Collatz shows a small bit of CPU with the GPU allowing for a
quad to do 4 tasks, plus the Collatz GPU task.
It seems that Einstein needs 100% of a CPU with the GPU, So I can only
run 3 other tasks while doing an Einstein CPU/GPU task.
Also, it seems that with Einstein CUDA is running, Collatz will not download
any new tasks. Yet, when Collatz CUDA is running Einstein will download their
CUDA tasks. That is how my system switched from running Collatz GPU to
Einstein's CUDA. I did not remember that Einstein had any CUDA applications,
or if it did, I would have to manually install the GPU app, like the others
I have looked into.
Now, tell me, how long would one of these task run, normally?
On my AMD Phenom 9650 quad running 2.3GHz and a GT220 card, the
manager estimated 4.5 - 4.75 hours to run. Is that good, bad,
so-so?
Never ran their CUDA apps before, so it would be nice to get an
idea on how my system is doing.
I was wondering about The CPU needs for the CUDA application.
Collatz shows a small bit of CPU with the GPU allowing for a
quad to do 4 tasks, plus the Collatz GPU task.
It seems that Einstein needs 100% of a CPU with the GPU, So I can only
run 3 other tasks while doing an Einstein CPU/GPU task.
Correct. The next version of the app will put more load on the GPU and less on the CPU.
Quote:
Also, it seems that with Einstein CUDA is running, Collatz will not download
any new tasks. Yet, when Collatz CUDA is running Einstein will download their
CUDA tasks.
That must be a coincidence. The E@H app has no direct say in the decision what tasks are downloaded for later execution. This is managed by the BOINC core client. Maybe if your pipeline length is small, BOINC will decide not to download additional CUDA tasks if it already has a E@H task.
Quote:
Now, tell me, how long would one of these task run, normally?
On my AMD Phenom 9650 quad running 2.3GHz and a GT220 card, the
manager estimated 4.5 - 4.75 hours to run. Is that good, bad,
so-so?
Never ran their CUDA apps before, so it would be nice to get an
idea on how my system is doing.
What BOINC estimates and what it finally takes in reality to complete a task are not necessarily the same. I'd expect that on your system, ABP2 tasks should take quite a bit less than 3 hours. For example, on my AMD X2 4850e with a rather mid-range 9800 GT Eco card, it takes about 3hr 15 minutes to complete a CUDA task (6 hours when run on the CPU only). The difference between CPU and CUDA runtime is less drastic on Intel CPUs, btw.
REmember GPU tasks are always done FIFO ... so, if BOINC thinks it has enough tasks it will never download more GPU tasks ...
I have reported this bug, posted the logs with the flags on and everything where BOINC will run the buffer dry, download some GPU tasks, run dry, download some tasks, run dry ... ad inf ... no interest by UCB ...
There are other issues with Strict FIFO, but you are getting bit by one of those that UCB is not interested in looking at ...
Just an observation from last night. Since the web update I can see run time vs. cpu time.
Up to this morning the run time equals cpu time (6,791.43 vs 6,717.98). Now the run time is smaller, than the cpu time (5,644.96 vs 6,716.27).
Is there something going on or is this just a feature?
I know I have low-end NVidia cards (8400 GS and 9500 GS), but the processing times I'm getting using my GPU's is worse than if I don't use them at all (e.g. a q6600 with no CUDA beat my Q8400 with a 8400 GS almost by a factor of 2 -- I doubt the q6600 is overclocked that much). I've disabled use of my GPU's by Einstein (if I had better cards, I'd use them).
I've just tried some Cudas on my machine, and it was a total waste of time. My machine crunches ABPS without GPU in ~8,000-9,000 seconds, the use of my GPU elongated this runtime to 13,000 seconds, an increase of more than 50%.
How is it possible to have extremely longer crunch times when the crunching power was much uncreased?
My GPU isn't the fastest as well, but if it's not faster with it than without, why bother sending WUs at all? Why does an increase if power make the crunching so much wasteful?
RE: Does configuring BOINC
)
As long as you keep
in your Einstein@Home preferences, I see no way to exclude S5 tasks.
Gruß,
Gundolf
[edit]You could try to find hints in the server logs by clicking the "Last contact" link in the Your computers list. I can't, because your computers are hidden.[/edit]
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: RE: Does configuring
)
When I attempt to access the "Last Contact" link - All I get is:
Oops! This link appears to be broken.
There are only 10 kind of people in the world, those that understand binary and those that don't!
RE: When I attempt to
)
You are right, I also get
at mine and the Index of /sched_logs is empty.
I seem to remember some other user complain about that a while back.
Sorry,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
RE: . . . Also, understand
)
I was wondering about The CPU needs for the CUDA application.
Collatz shows a small bit of CPU with the GPU allowing for a
quad to do 4 tasks, plus the Collatz GPU task.
It seems that Einstein needs 100% of a CPU with the GPU, So I can only
run 3 other tasks while doing an Einstein CPU/GPU task.
Also, it seems that with Einstein CUDA is running, Collatz will not download
any new tasks. Yet, when Collatz CUDA is running Einstein will download their
CUDA tasks. That is how my system switched from running Collatz GPU to
Einstein's CUDA. I did not remember that Einstein had any CUDA applications,
or if it did, I would have to manually install the GPU app, like the others
I have looked into.
Now, tell me, how long would one of these task run, normally?
On my AMD Phenom 9650 quad running 2.3GHz and a GT220 card, the
manager estimated 4.5 - 4.75 hours to run. Is that good, bad,
so-so?
Never ran their CUDA apps before, so it would be nice to get an
idea on how my system is doing.
RE: I was wondering about
)
Correct. The next version of the app will put more load on the GPU and less on the CPU.
That must be a coincidence. The E@H app has no direct say in the decision what tasks are downloaded for later execution. This is managed by the BOINC core client. Maybe if your pipeline length is small, BOINC will decide not to download additional CUDA tasks if it already has a E@H task.
What BOINC estimates and what it finally takes in reality to complete a task are not necessarily the same. I'd expect that on your system, ABP2 tasks should take quite a bit less than 3 hours. For example, on my AMD X2 4850e with a rather mid-range 9800 GT Eco card, it takes about 3hr 15 minutes to complete a CUDA task (6 hours when run on the CPU only). The difference between CPU and CUDA runtime is less drastic on Intel CPUs, btw.
CU
HB
REmember GPU tasks are always
)
REmember GPU tasks are always done FIFO ... so, if BOINC thinks it has enough tasks it will never download more GPU tasks ...
I have reported this bug, posted the logs with the flags on and everything where BOINC will run the buffer dry, download some GPU tasks, run dry, download some tasks, run dry ... ad inf ... no interest by UCB ...
There are other issues with Strict FIFO, but you are getting bit by one of those that UCB is not interested in looking at ...
Bikeman wrote: "Correct.
)
Bikeman wrote:
"Correct. The next version of the app will put more load on the GPU and less on the CPU."
When we can use the next version?
RE: When we can use the
)
I'll ask the devs to give an update. Stay tuned.
HB
Just an observation from last
)
Just an observation from last night. Since the web update I can see run time vs. cpu time.
Up to this morning the run time equals cpu time (6,791.43 vs 6,717.98). Now the run time is smaller, than the cpu time (5,644.96 vs 6,716.27).
Is there something going on or is this just a feature?
RE: I know I have low-end
)
I've just tried some Cudas on my machine, and it was a total waste of time. My machine crunches ABPS without GPU in ~8,000-9,000 seconds, the use of my GPU elongated this runtime to 13,000 seconds, an increase of more than 50%.
How is it possible to have extremely longer crunch times when the crunching power was much uncreased?
My GPU isn't the fastest as well, but if it's not faster with it than without, why bother sending WUs at all? Why does an increase if power make the crunching so much wasteful?
Grüße vom Sänger
