So your 4600 seconds vs. his 6600 seconds would make Catalyst about 30% faster.
No, you can't say that. Because I'm crunching 4 concurrently there is an advantage for the per task times. If things still run correctly, he should get at least some improvement on elapsed time per task by trying 2x. It would be interesting to see what happened.
Quote:
I think I would keep using an older distro until Catalyst supported something newer.
I'm using fglrx drivers that were current in mid 2014. Everything works well so I'm happy to leave things as they are for the moment. I have no experience of more recent versions so have no knowledge of any problems there might be.
I'm glad others are "happy" with catalyst, but I'm hear because it's a disaster for me. I'm a performance "nut" too; so you can be sure this is a last resort option for me and I have little choice. It's a choice between doing any BOINC work with my GPUs or none at all. I feel pretty good I got E@Home to work this way; I haven't got SETI@Home to work and I consider that my main project.
Question: The last time I ran 2x I was running anonymous platform w/ Lunatics apps. Do I need to run anon--make my own app_config.xml for all apps--in order to alter the predetermined resource requirements of each app, or can I do that in a override file so I can stay "official"? Anon is fun, I remember, but I want to get new apps as they get released without having to watch all my projects for updates and changes.
To run multiple GPU tasks at a time here at Einstein check out your Einstein@home prefs and "GPU utilization factor" settings. Set it to 0.5 to run x2, 0.33 to run x3 and so on.
The setting will take effect after you download new work for the GPU and will then effect all work in your cache. A simple update to the project server will not work in this case.
The alternative is to use an app_config.xml to have more control and be able to change the settings by doing a Options -> Read config files in Boinc manager.
More info on app_config.xml in the documentation here
For the Application_Name tag use einsteinbinary_BRP6 for "Binary Radio Pulsar Search (Parkes PMPS XT)" = "BRP6" work and einsteinbinary_BRP4G for "Binary Radio Pulsar Search (Arecibo, GPU)" = "BRP4G" work.
I'm glad others are "happy" with catalyst, but I'm hear because it's a disaster for me.
What exactly is the model of your GPU. I know it's a Pitcairn series but that covers a wide range of actual models. Mine are MSI HD7850 2GB.
Quote:
Question: The last time I ran 2x I was running anonymous platform w/ Lunatics apps. Do I need to run anon--make my own app_config.xml for all apps--in order to alter the predetermined resource requirements of each app, or can I do that in a override file so I can stay "official"?
You don't need anonymous platform (which used app_info.xml - not app_config.xml). Just go to your account page and click on the EAH project prefs link. There you will find a GPU utilization factor for BRP apps (with the usual CYA type warnings). If you set it to 0.5 (and then download new BRP6 work) your machine will automatically start doing them 2x (provided your driver doesn't have some issue with it - I'm being cautious) :-). This should be fine for running 2x. If you want to go back to running singly, just set it back to 1 (and download new work). Also, there's nothing to do if the app version changes.
You can achieve the same thing with more control (and the possibility of more knobs to tweak) by creating a local app_config.xml file. You then can use the 're-read config files' option in BOINC Manager to get immediate change. Try the GPU utilization factor method first. If you don't want to risk trashing a whole bunch of tasks, run your work cache down to a low value before making any changes. Best of luck. I'd love to know how you get on :-).
I've got two 7870 cards. I've got my eye on the R9 390, but I'm not sure that would fix any of my problems. That's one of the reasons I'm here, doing this research: to find out what my options are and determine if upgrading would help. PITCARN is a old, I know, but the future is really uncertain. If you know more about how the current development of AMDGPU/Vulcan/Kernel will play out with respect to BOINC, I would be grateful to learn about it. I read Phoronix's stuff, but it's not like OpenCL is his focus.
Thanks for pointing me to project preferences; I haven't poked around there in years. I made a change and did an "update" on this project, but still running 1x. It looks like other settings I changed got updated (like project shares), but app resource requirements didn't change; running GPU app still shows 0.5CPU+1GPU in task list.
I made a change and did an "update" on this project, but still running 1x. It looks like other settings I changed got updated (like project shares),
That is all normal. The "multiplicity" change only gets implmented on your host after it downloads work. That is unlike most other preference, which generally take effect when you do a project update.
OK, thanks. They should be able to produce more (and use more power doing so) than my 7850s. Even though old, I regard that series to be very suited to crunching.
Quote:
If you know more about how the current development of AMDGPU/Vulcan/Kernel will play out with respect to BOINC, I would be grateful to learn about it. I read Phoronix's stuff, but it's not like OpenCL is his focus.
I have no idea about this. I value stable and efficient crunching which means I don't really investigate/buy stuff until it's been reasonably well tested (and reported on) by others. Almost invariably I find that really high end stuff is overpriced for what it can produce. I'm talking exclusively about crunching.
Quote:
It looks like other settings I changed got updated (like project shares), but app resource requirements didn't change; running GPU app still shows 0.5CPU+1GPU in task list.
I did say, "If you set it to 0.5 (and then download new BRP6 work) ..." :-). Your tasks list shows a download of 11 new BRP6 tasks at 15 Feb 2016, 19:28:49 UTC. The previous download was a day earlier. If you want to try out 2x immediately and also protect the bulk of your work cache from being trashed, by suspending them for their protection, you could use the following procedure.
* Set up an app_config.xml file (see below) in the Einstein project directory.
* Suspend ALL EXCEPT 4 of your current BRP6 tasks - the two crunching and the next two that are ready to start.
* Use BOINC Manager (or boinccmd if CLI) to 're-read config files'.
* When you are satisfied that there are 4 tasks crunching correctly (the original two plus two new ones) you can un-suspend all the others.
* If things crash and burn, remove app_config.xml, remove the website pref settings and then restart BOINC.
* Resume crunching the protected tasks singly.
Below is a suitable app_config.xml. This will override any website preferences and can be edited and 're-read' whenever you want to make an immediate change. You should be able to copy and paste directly from your screen.
[pre]
einsteinbinary_BRP6
0.5
0.5
einsteinbinary_BRP4G
0.5
0.5
[/pre]
I noticed you have a couple of BRP4G tasks mixed in with the BRP6s. I've added an extra block above just for them. I don't do any BRP4G so I'm not sure if I've got the exactly correct for them. The documentation that Holmis linked to has instructions about how to get the correct short name if the above is incorrect.
Thanks, I had missed that
)
Thanks, I had missed that link.
So your 4600 seconds vs. his 6600 seconds would make Catalyst about 30% faster.
That's a bit much. I think I would keep using an older distro until Catalyst supported something newer.
RE: I thought it would
)
For the CUDA version yes / OpenCL sadly no.
I suspect you might find a logging option for boinc to log it locally.
It would be within the power of the E@H devs to log the device it was running on as boinc passes a "--device " in the run string.
Eg.
RE: I think I would keep
)
These are very wise words.
RE: So your 4600 seconds
)
No, you can't say that. Because I'm crunching 4 concurrently there is an advantage for the per task times. If things still run correctly, he should get at least some improvement on elapsed time per task by trying 2x. It would be interesting to see what happened.
I'm using fglrx drivers that were current in mid 2014. Everything works well so I'm happy to leave things as they are for the moment. I have no experience of more recent versions so have no knowledge of any problems there might be.
Cheers,
Gary.
Sorry for the delay... I'm
)
Sorry for the delay...
I'm glad others are "happy" with catalyst, but I'm hear because it's a disaster for me. I'm a performance "nut" too; so you can be sure this is a last resort option for me and I have little choice. It's a choice between doing any BOINC work with my GPUs or none at all. I feel pretty good I got E@Home to work this way; I haven't got SETI@Home to work and I consider that my main project.
Question: The last time I ran 2x I was running anonymous platform w/ Lunatics apps. Do I need to run anon--make my own app_config.xml for all apps--in order to alter the predetermined resource requirements of each app, or can I do that in a override file so I can stay "official"? Anon is fun, I remember, but I want to get new apps as they get released without having to watch all my projects for updates and changes.
To run multiple GPU tasks at
)
To run multiple GPU tasks at a time here at Einstein check out your Einstein@home prefs and "GPU utilization factor" settings. Set it to 0.5 to run x2, 0.33 to run x3 and so on.
The setting will take effect after you download new work for the GPU and will then effect all work in your cache. A simple update to the project server will not work in this case.
The alternative is to use an app_config.xml to have more control and be able to change the settings by doing a Options -> Read config files in Boinc manager.
More info on app_config.xml in the documentation here
For the Application_Name tag use einsteinbinary_BRP6 for "Binary Radio Pulsar Search (Parkes PMPS XT)" = "BRP6" work and einsteinbinary_BRP4G for "Binary Radio Pulsar Search (Arecibo, GPU)" = "BRP4G" work.
RE: I'm glad others are
)
What exactly is the model of your GPU. I know it's a Pitcairn series but that covers a wide range of actual models. Mine are MSI HD7850 2GB.
You don't need anonymous platform (which used app_info.xml - not app_config.xml). Just go to your account page and click on the EAH project prefs link. There you will find a GPU utilization factor for BRP apps (with the usual CYA type warnings). If you set it to 0.5 (and then download new BRP6 work) your machine will automatically start doing them 2x (provided your driver doesn't have some issue with it - I'm being cautious) :-). This should be fine for running 2x. If you want to go back to running singly, just set it back to 1 (and download new work). Also, there's nothing to do if the app version changes.
You can achieve the same thing with more control (and the possibility of more knobs to tweak) by creating a local app_config.xml file. You then can use the 're-read config files' option in BOINC Manager to get immediate change. Try the GPU utilization factor method first. If you don't want to risk trashing a whole bunch of tasks, run your work cache down to a low value before making any changes. Best of luck. I'd love to know how you get on :-).
Cheers,
Gary.
I've got two 7870 cards.
)
I've got two 7870 cards. I've got my eye on the R9 390, but I'm not sure that would fix any of my problems. That's one of the reasons I'm here, doing this research: to find out what my options are and determine if upgrading would help. PITCARN is a old, I know, but the future is really uncertain. If you know more about how the current development of AMDGPU/Vulcan/Kernel will play out with respect to BOINC, I would be grateful to learn about it. I read Phoronix's stuff, but it's not like OpenCL is his focus.
Thanks for pointing me to project preferences; I haven't poked around there in years. I made a change and did an "update" on this project, but still running 1x. It looks like other settings I changed got updated (like project shares), but app resource requirements didn't change; running GPU app still shows 0.5CPU+1GPU in task list.
RE: I made a change and
)
That is all normal. The "multiplicity" change only gets implmented on your host after it downloads work. That is unlike most other preference, which generally take effect when you do a project update.
RE: I've got two 7870
)
OK, thanks. They should be able to produce more (and use more power doing so) than my 7850s. Even though old, I regard that series to be very suited to crunching.
I have no idea about this. I value stable and efficient crunching which means I don't really investigate/buy stuff until it's been reasonably well tested (and reported on) by others. Almost invariably I find that really high end stuff is overpriced for what it can produce. I'm talking exclusively about crunching.
I did say, "If you set it to 0.5 (and then download new BRP6 work) ..." :-). Your tasks list shows a download of 11 new BRP6 tasks at 15 Feb 2016, 19:28:49 UTC. The previous download was a day earlier. If you want to try out 2x immediately and also protect the bulk of your work cache from being trashed, by suspending them for their protection, you could use the following procedure.
* Suspend ALL EXCEPT 4 of your current BRP6 tasks - the two crunching and the next two that are ready to start.
* Use BOINC Manager (or boinccmd if CLI) to 're-read config files'.
* When you are satisfied that there are 4 tasks crunching correctly (the original two plus two new ones) you can un-suspend all the others.
* If things crash and burn, remove app_config.xml, remove the website pref settings and then restart BOINC.
* Resume crunching the protected tasks singly.
Below is a suitable app_config.xml. This will override any website preferences and can be edited and 're-read' whenever you want to make an immediate change. You should be able to copy and paste directly from your screen.
[pre]
einsteinbinary_BRP6
0.5
0.5
einsteinbinary_BRP4G
0.5
0.5
[/pre]
I noticed you have a couple of BRP4G tasks mixed in with the BRP6s. I've added an extra block above just for them. I don't do any BRP4G so I'm not sure if I've got the exactly correct for them. The documentation that Holmis linked to has instructions about how to get the correct short name if the above is incorrect.
Good luck with getting it going.
Cheers,
Gary.