If I want to get GW GPU (Beta) tasks only (no CPU ones), which settings should I set ?
If I have:
Quote:
Run CPU versions of applications for which GPU versions are available NO
Run beta/test application versions? YES
Gravitational Wave S6 Directed Search (CasA) NO
I don't seem to be getting any tasks.
However if I change to:
Quote:
Gravitational Wave S6 Directed Search (CasA) YES
then I'm getting CPU tasks as well (even though "Run CPU versions of applications for which GPU versions are available = NO")
Use CPU (Enforced by version 6.10+): no
Use ATI GPU (Enforced by version 6.10+): yes
Use NVIDIA GPU (Enforced by version 6.10+): no
Use INTEL GPU (Enforced by version 7.0.27+): no
...perhaps you still have "use CPU" set to "yes" still?
Yes. In current system configuration Beta GPU versions don't count as availabe in the sense of "Run CPU versions of applications for which GPU versions are available".
I have also noticed one oddity here.
When I first enable to get new tasks on a machine (GW GPU enabled), the request takes too long and after a while, I get a HTTP server error.
Then the next request seems to work, though the WUs come as "lost". This happened on 2 different machines now, so it's not just a coincidence.
Quote:
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI Scheduler request failed: HTTP internal server error
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI
Scheduler request completed: got 12 new tasks Resent lost task h1_0999.35_S6Directed__S6CasAf40a_999.75Hz_289_3
Another oddity is that, while on an AMD system, the GPU usage seems to be quite steady, on NV it jumps constantly between 70-99% (with a few sec period).
I have also noticed one oddity here.
When I first enable to get new tasks on a machine (GW GPU enabled), the request takes too long and after a while, I get a HTTP server error.
Then the next request seems to work, though the WUs come as "lost". This happened on 2 different machines now, so it's not just a coincidence.
Quote:
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI Scheduler request failed: HTTP internal server error
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI
Scheduler request completed: got 12 new tasks Resent lost task h1_0999.35_S6Directed__S6CasAf40a_999.75Hz_289_3
Another oddity is that, while on an AMD system, the GPU usage seems to be quite steady, on NV it jumps constantly between 70-99% (with a few sec period).
Whet tool are you measuring the "GPU usage" with, i.e. what exactly is it counting as "usage"?
I'd guess that the relevant difference is not really the manufacturer of the card, but the memory bandwidth. Could it be that your NVidia card has less GPU memory bandwidth than your ATI?
BTW, any estimation when GW GPU will move from Beta to standard app ?
Honestly I doubt that we'll do this in the next 50d, i.e. before the end of the run.
This has to do with both the code itself as well as the way Beta application versions are implemented in the server code Einstein@Home is currently using.
Both (application and server code) will not be used beyond the current S6CasA run, which is why I doubt that we'll spend any more work on either issue.
I have also noticed one oddity here.
When I first enable to get new tasks on a machine (GW GPU enabled), the request takes too long and after a while, I get a HTTP server error.
Then the next request seems to work, though the WUs come as "lost". This happened on 2 different machines now, so it's not just a coincidence.
Quote:
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI Scheduler request failed: HTTP internal server error
Sending scheduler request: To fetch work.
Requesting new tasks for CPU and ATI
Scheduler request completed: got 12 new tasks Resent lost task h1_0999.35_S6Directed__S6CasAf40a_999.75Hz_289_3
Another oddity is that, while on an AMD system, the GPU usage seems to be quite steady, on NV it jumps constantly between 70-99% (with a few sec period).
Whet tool are you measuring the "GPU usage" with, i.e. what exactly is it counting as "usage"?
I'd guess that the relevant difference is not really the manufacturer of the card, but the memory bandwidth. Could it be that your NVidia card has less GPU memory bandwidth than your ATI?
BM
I'm using HWiNFO of course ;) It reports this via NVAPI, so it should be reliable. Though nobody exactly knows what the term "GPU usage" means. Btw, Afterburner reports the same.
The GPU is a 660 Ti.
Could that be because of how the app is written? It seems that the GW OpenCL ATI app does some CPU pre- and post-processing for each WU, while the NV app does this during the entire run ?
If I want to get GW GPU
)
If I want to get GW GPU (Beta) tasks only (no CPU ones), which settings should I set ?
If I have:
I don't seem to be getting any tasks.
However if I change to:
then I'm getting CPU tasks as well (even though "Run CPU versions of applications for which GPU versions are available = NO")
Any clues?
-----
these are my CPU/GPU
)
these are my CPU/GPU settings:
...perhaps you still have "use CPU" set to "yes" still?
RE: Any clues? Yes. In
)
Yes. In current system configuration Beta GPU versions don't count as availabe in the sense of "Run CPU versions of applications for which GPU versions are available".
BM
BM
Oh indeed, I had "Use
)
Oh indeed, I had "Use CPU"=yes. I'm getting old... Thanks!
BTW, any estimation when GW GPU will move from Beta to standard app ?
-----
I have also noticed one
)
I have also noticed one oddity here.
When I first enable to get new tasks on a machine (GW GPU enabled), the request takes too long and after a while, I get a HTTP server error.
Then the next request seems to work, though the WUs come as "lost". This happened on 2 different machines now, so it's not just a coincidence.
Another oddity is that, while on an AMD system, the GPU usage seems to be quite steady, on NV it jumps constantly between 70-99% (with a few sec period).
-----
RE: I have also noticed one
)
I do have the same problem on my main rig.
RE: Another oddity is that,
)
Whet tool are you measuring the "GPU usage" with, i.e. what exactly is it counting as "usage"?
I'd guess that the relevant difference is not really the manufacturer of the card, but the memory bandwidth. Could it be that your NVidia card has less GPU memory bandwidth than your ATI?
BM
BM
RE: BTW, any estimation
)
Honestly I doubt that we'll do this in the next 50d, i.e. before the end of the run.
This has to do with both the code itself as well as the way Beta application versions are implemented in the server code Einstein@Home is currently using.
Both (application and server code) will not be used beyond the current S6CasA run, which is why I doubt that we'll spend any more work on either issue.
BM
BM
RE: RE: I have also
)
I'll take a look at that on Monday (tomorrow).
BM
BM
RE: RE: Another oddity is
)
I'm using HWiNFO of course ;) It reports this via NVAPI, so it should be reliable. Though nobody exactly knows what the term "GPU usage" means. Btw, Afterburner reports the same.
The GPU is a 660 Ti.
Could that be because of how the app is written? It seems that the GW OpenCL ATI app does some CPU pre- and post-processing for each WU, while the NV app does this during the entire run ?
-----