You could always override the project's default reservations with one of your own in a app_config.xml
Sure, and I do that already in a few cases. For example, the bulk of my GTX650 hosts run in a particular venue where the GPU utilisation factor is set to 0.5 so that tasks are processed 2x. I've found that a few can squeeze a bit more performance by running 3x. They are still in the same venue but I override the 0.5 factor with 0.33 in an app_config.xml.
But I can't see that strategy working for FGRP3. In the FGRP3 technical news thread, Bernd said, "so far only the FFT will run on the GPU, the rest of the computation (which is still a lot) is done on the CPU." I take that to mean that overall performance will suffer if I were to try reducing the fractional CPU allocation through an app_config.xml
Quote:
(and if you're running one of the latest Boinc dev versions, you can do it by app_versions too)
I wasn't aware of "app_versions" so I tried googling app_versions.xml. When I got nothing appropriate, I went and reread the app_config.xml documentation and saw the new .... stuff which I don't remember seeing the last time I looked - probably quite a while ago. I notice you need 7.3+ to use it. The latest I'm using (on a few hosts) is 7.2.33. Even if I were to go find a 7.3+ version, how would this help? Once again, I could change the fractional CPU requirement to the probable detriment of overall efficiency. So, I can't really see either of these options helping, or am I just missing something?
There isn't a FGRP3 GPU app listed on the 'applications' page at the moment so it might be a while before there is a further attempt to run one. I'll just wait and see how it performs when it is relaunched.
(and if you're running one of the latest Boinc dev versions, you can do it by app_versions too)
I wasn't aware of "app_versions" so I tried googling app_versions.xml. When I got nothing appropriate, I went and reread the app_config.xml documentation and saw the new .... stuff which I don't remember seeing the last time I looked - probably quite a while ago. I notice you need 7.3+ to use it. The latest I'm using (on a few hosts) is 7.2.33. Even if I were to go find a 7.3+ version, how would this help? Once again, I could change the fractional CPU requirement to the probable detriment of overall efficiency. So, I can't really see either of these options helping, or am I just missing something?
There isn't a FGRP3 GPU app listed on the 'applications' page at the moment so it might be a while before there is a further attempt to run one. I'll just wait and see how it performs when it is relaunched.
It's supposed in the following dev versions (in spite of what the wiki says): On Boinc dev versions 7.2.29, 7.2.30, 7.2.32, 7.2.34, 7.2.35, 7.2.36, 7.2.37, 7.2.38 and 7.2.39
It's supposed in the following dev versions (in spite of what the wiki says): On Boinc dev versions 7.2.29, 7.2.30, 7.2.32, 7.2.34, 7.2.35, 7.2.36, 7.2.37, 7.2.38 and 7.2.39
OK, thanks! I'll certainly try the option once a GPU app reappears. Strange that the feature is in everything BUT 7.2.33. I did a bit of browsing and saw that at one point 7.2.28 was released to the public but then withdrawn/replaced by 7.2.31 and eventually 7.2.33. I guess that implies that 7.2.33 is a rebadged 7.2.28/7.2.31 with fixes but not the new feature.
I don't have time to follow BOINC development closely and I certainly don't have time to experiment on a range of versions over a large number of machines. If there is any particular version in the above list that stands out as better/more stable than the others, I'd appreciate the advice.
It's supposed in the following dev versions (in spite of what the wiki says): On Boinc dev versions 7.2.29, 7.2.30, 7.2.32, 7.2.34, 7.2.35, 7.2.36, 7.2.37, 7.2.38 and 7.2.39
OK, thanks! I'll certainly try the option once a GPU app reappears. Strange that the feature is in everything BUT 7.2.33. I did a bit of browsing and saw that at one point 7.2.28 was released to the public but then withdrawn/replaced by 7.2.31 and eventually 7.2.33. I guess that implies that 7.2.33 is a rebadged 7.2.28/7.2.31 with fixes but not the new feature.
Exactly so. They released v7.2.28 as 'work completed for this release', and moved on to testing the next round of new features. But they had to call it back for snagging...
Quote:
I don't have time to follow BOINC development closely and I certainly don't have time to experiment on a range of versions over a large number of machines. If there is any particular version in the above list that stands out as better/more stable than the others, I'd appreciate the advice.
I'd forget anything except v7.2.33 for regular work, and v7.2.39 and successors for testing. Basically, follow the numbers that appear on http://boinc.berkeley.edu/download_all.php.
I'd forget anything except v7.2.33 for regular work, and v7.2.39 and successors for testing. Basically, follow the numbers that appear on http://boinc.berkeley.edu/download_all.php.
Well I got some more of these work units. ATM I am running all 4 cpu cores and three tasks at a time on the gpu. When ever one runs it takes a full cpu core and 1/3 of the gpu per task, so instead of running 7 tasks I am only running 6. Is this the way it is supposed to be? It seems like a lot of resources being used for 1 task.
Well I got some more of these work units. ATM I am running all 4 cpu cores and three tasks at a TIME ON THE GPU. When ever one runs it takes a full cpu core and 1/3 of the gpu per task, so instead of running 7 tasks I am only running 6. Is this the way it is supposed to be? It seems like a lot of resources being used for 1 task.
Yes, until the project people have had a chance to develop the app further this is the way it's going to be. I fully expect this app to take the same route as the BRP-GPU app once did and get refined for every development cycle, as to a time line on that I haven't seen anything posted yet.
Now I'm running 3 of these devils and only have 4 tasks in progress. My poor RAC.
If you wish to run FGRP3 CPU tasks on all the available cores, you will need to preclude FGRP3 GPU tasks which have the high CPU requirement. You should read this information for some ideas you might like to implement.
RE: You could always
)
Sure, and I do that already in a few cases. For example, the bulk of my GTX650 hosts run in a particular venue where the GPU utilisation factor is set to 0.5 so that tasks are processed 2x. I've found that a few can squeeze a bit more performance by running 3x. They are still in the same venue but I override the 0.5 factor with 0.33 in an app_config.xml.
But I can't see that strategy working for FGRP3. In the FGRP3 technical news thread, Bernd said, "so far only the FFT will run on the GPU, the rest of the computation (which is still a lot) is done on the CPU." I take that to mean that overall performance will suffer if I were to try reducing the fractional CPU allocation through an app_config.xml
I wasn't aware of "app_versions" so I tried googling app_versions.xml. When I got nothing appropriate, I went and reread the app_config.xml documentation and saw the new .... stuff which I don't remember seeing the last time I looked - probably quite a while ago. I notice you need 7.3+ to use it. The latest I'm using (on a few hosts) is 7.2.33. Even if I were to go find a 7.3+ version, how would this help? Once again, I could change the fractional CPU requirement to the probable detriment of overall efficiency. So, I can't really see either of these options helping, or am I just missing something?
There isn't a FGRP3 GPU app listed on the 'applications' page at the moment so it might be a while before there is a further attempt to run one. I'll just wait and see how it performs when it is relaunched.
Cheers,
Gary.
RE: RE: (and if you're
)
It's supposed in the following dev versions (in spite of what the wiki says): On Boinc dev versions 7.2.29, 7.2.30, 7.2.32, 7.2.34, 7.2.35, 7.2.36, 7.2.37, 7.2.38 and 7.2.39
Claggy
RE: It's supposed in the
)
OK, thanks! I'll certainly try the option once a GPU app reappears. Strange that the feature is in everything BUT 7.2.33. I did a bit of browsing and saw that at one point 7.2.28 was released to the public but then withdrawn/replaced by 7.2.31 and eventually 7.2.33. I guess that implies that 7.2.33 is a rebadged 7.2.28/7.2.31 with fixes but not the new feature.
I don't have time to follow BOINC development closely and I certainly don't have time to experiment on a range of versions over a large number of machines. If there is any particular version in the above list that stands out as better/more stable than the others, I'd appreciate the advice.
Cheers,
Gary.
RE: RE: It's supposed in
)
Exactly so. They released v7.2.28 as 'work completed for this release', and moved on to testing the next round of new features. But they had to call it back for snagging...
I'd forget anything except v7.2.33 for regular work, and v7.2.39 and successors for testing. Basically, follow the numbers that appear on http://boinc.berkeley.edu/download_all.php.
RE: I'd forget anything
)
OK, thanks very much!
Cheers,
Gary.
Well I got some more of these
)
Well I got some more of these work units. ATM I am running all 4 cpu cores and three tasks at a time on the gpu. When ever one runs it takes a full cpu core and 1/3 of the gpu per task, so instead of running 7 tasks I am only running 6. Is this the way it is supposed to be? It seems like a lot of resources being used for 1 task.
RE: Well I got some more of
)
Yes, until the project people have had a chance to develop the app further this is the way it's going to be. I fully expect this app to take the same route as the BRP-GPU app once did and get refined for every development cycle, as to a time line on that I haven't seen anything posted yet.
Well now I am running 2 of
)
Well now I am running 2 of these and in total have only 5 tasks in progress.
Now I'm running 3 of these
)
Now I'm running 3 of these devils and only have 4 tasks in progress. My poor RAC.
RE: Now I'm running 3 of
)
If you wish to run FGRP3 CPU tasks on all the available cores, you will need to preclude FGRP3 GPU tasks which have the high CPU requirement. You should read this information for some ideas you might like to implement.
Cheers,
Gary.