I can now see that you have quite a few new tasks so I guess that the website pref settings are finally working and you have 4 tasks running instead of 2. If this is so, you can ignore the app_config.xml file if you don't want to bother with that.
There are no error tasks showing up so hopefully all is going smoothly. If the crunch time for tasks done 2x turns out to be less than 12,000 secs, you are on a win. With 4 GPU tasks crunching you will have 2 less 'available' CPU cores. If you previously had CPU tasks on all 8 cores, you now should have 6 only with 2 'waiting to run'. Incidently, if all 8 cores were running tasks, this might explain why your GPU tasks were taking a lot longer than mine on a per task basis. AMD GPUs really do benefit from having the extra CPU support, particularly if it's an AMD CPU (from experience with the ones I've tried).
E@Home is working as hoped; I now have 2x on both cards.
I'm aware that GPU tasks take some CPU and that it varies by app and that GPU tasks greatly benefit from having uncontested access to CPU resources. Since my 8-core really has only 4 ALUs, I had BOINC set to use only 50% of available CPUs. This kept BOINC from over subscribing the system to CPU apps. Given other default settings, this will leave 1 ALU free (5 free cores, 3 cores running CPU tasks). With 2x, however, I figure this setting would unnecessarily restrictive. So, I increased the available CPU usage limit to 75%. I now have 4 CPU tasks in addition to the 4 GPU tasks. I wish there was a better way to optimize or estimate the best combination. That's what the default resource requirements are for, I assume, but as you and many others have noticed, the defaults are not optimal for all systems, nor can we expect that.
I'm anxious to see tasks times. They'll be submitted in a few hours, but elapse times reported in the task list look completely normal.
Congratulations! Well done. I took a look and there is a new returned task showing 9627 secs. Partly done at 1x and partly at 2x. One more to go before you get into those done entirely at 2x.
Quote:
... I increased the available CPU usage limit to 75%.
That sounds like a good value if you wish to continue controlling this using % of CPUs method. The disadvantage of this is that BOINC is restricted even if you have no GPU tasks running. I tend to use app_config.xml. If you wanted 4 CPU tasks in addition to 4 GPU tasks you could set to 0.5 and to 1.0 and have BOINC set to use 100% of the CPU cores.
Quote:
I wish there was a better way to optimize or estimate the best combination.
There is!! It's called 'suck it and see' :-). I'm continually amazed that the defaults get it right most of the time. Particularly with AMD CPUs driving AMD discrete GPUs, you do need to experiment to find what works best. I have a Phenom II X4 with a HD7770 GPU. I'm running 3 GPU tasks and 2 CPU tasks. To achieve this at 100% CPUs for BOINC, I use of 0.33 and of 0.67. If you can understand that, you're home and hosed! You can't achieve this 3+2 combination with just the GPU utilization factor without also fiddling with the % CPUs BOINC setting. And, with app_config.xml, it's so quick and easy to change things when you want to test a different combination.
You now have several completed results showing crunch times around 11,500 secs. I thought you might get a bit better than that but at least it's a gain. My 7850s using fglrx (mid 2014) run 4x and do all 4 in about 18,000 - 18,500 secs.
Why can't you run an older fglrx driver if the latest versions give you trouble?
I'll consider app_config, but the documentation still doesn't make clear if, when using it, you get new app versions when they become available without adding them to the file. I don't mind all tasks getting aborted when I make an config file error, but getting the new app version details is a pain and I don't want to have to do that for all my projects, anyway.
Quote:
Why can't you run an older fglrx driver if the latest versions give you trouble?
I bet you can guess...the older one gives me trouble, too. ;-) The last driver I had working was actually Crimson (5.12) w/ kernel 4.2 & Xorg 1.17. But, I want/need to run kernel 4.3 (or newer) and when I try it, X11 hangs or starts and fails repeatedly preventing me from safely rebooting. I get the same symptoms when I try 5.9 Catalyst, too. I even tried the community patches. I'm working that angle, too, in parallel with this effort to get CLOVER working. But, after crashing my machine six times playing with Catalyst, I had to try it this way. I asked elsewhere and people with 4.4 say it's working and people with other hardware, too, but I haven't got confirmation from anyone with a similar system to mine saying it works. Perhaps I should try again, but I'm going to need a lot more encouragement and hard evidence. Plus, I'm sure Xorg 1.18 isn't supported, so I'd have to downgrade that just to try, but if it doesn't work, upgrade Xorg again just to restore what I have now. I've run ATi/AMD proprietary drivers on Linux since 3DFX died and this is the first time I've ever given up.
Aside from DPMS not working, radeonsi is better in other ways, too, though. Games that never worked with Catalyst now run fine. I don't play much these days, but as a test, it says a lot that the tables have turned.
App_config is very much simpler and more lightweight than app_info.
Yes, you get new versions and even new applications as released, and they will run as normal - version upgrades will probably use the same configuration values, completely new applications will run as if the app_config wasn't present until you get round to updating it. Both app_config and app_info operate project-by-project: there's no need to change any other project if you find one of them useful here.
App_info is the dangerous one which could lead to tasks getting aborted if you get is wrong: I can't even think of a way to kill jobs with app_config.
Well, I'm back. After happily working for 6 months, everything broke two days ago. The APP is complete missing and BOINC deleted 660 files. Did you experience this too? Or, is it just me?
I never did follow up with Paul IE, as I promised. I'm sorry about that. This is disappointing, though, because the academic year just started, and I can't see how I can look into this problem, now.
Which app? I saw your post in BRP4G thread and now in BRP6 thread all while the logs mention GW. BRP6 has ended, only timed out/error'd WUs will be resent. FGRPB1 work generator is currently enabled
I'm not thinking of a specific APP. I'm interested in why all of the GPU apps for E@H are not running now. It's not clear why. I posted here because I thought something related to ATi had broken. I see the Tech News post about BRP6 ending. Okay, that explains BRP6, at least.
See this log line:
Sun 02 Oct 2016 03:48:23 PM | Einstein@Home | Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
I assume that refers to all BRP apps, but maybe not. It's also not clear what this message means in this situation. Does this mean there are no WUs for the APPs that are available on my platform, or does it mean there are no BRP APPs for my platform. The latter is implied by the message, but that would be change since two days ago. But, according to the server status page, the BRP4G and BRP6 queues are empty. Am I reading that correctly? If so, there isn't any way to know if it's broken or not at the moment. Right?
But, isn't BRP4 a non-GPU BRP app? So, I still don't understand that log entry.
I can now see that you have
)
I can now see that you have quite a few new tasks so I guess that the website pref settings are finally working and you have 4 tasks running instead of 2. If this is so, you can ignore the app_config.xml file if you don't want to bother with that.
There are no error tasks showing up so hopefully all is going smoothly. If the crunch time for tasks done 2x turns out to be less than 12,000 secs, you are on a win. With 4 GPU tasks crunching you will have 2 less 'available' CPU cores. If you previously had CPU tasks on all 8 cores, you now should have 6 only with 2 'waiting to run'. Incidently, if all 8 cores were running tasks, this might explain why your GPU tasks were taking a lot longer than mine on a per task basis. AMD GPUs really do benefit from having the extra CPU support, particularly if it's an AMD CPU (from experience with the ones I've tried).
Cheers,
Gary.
Awesome! Thanks so much, yet
)
Awesome! Thanks so much, yet again, everyone.
E@Home is working as hoped; I now have 2x on both cards.
I'm aware that GPU tasks take some CPU and that it varies by app and that GPU tasks greatly benefit from having uncontested access to CPU resources. Since my 8-core really has only 4 ALUs, I had BOINC set to use only 50% of available CPUs. This kept BOINC from over subscribing the system to CPU apps. Given other default settings, this will leave 1 ALU free (5 free cores, 3 cores running CPU tasks). With 2x, however, I figure this setting would unnecessarily restrictive. So, I increased the available CPU usage limit to 75%. I now have 4 CPU tasks in addition to the 4 GPU tasks. I wish there was a better way to optimize or estimate the best combination. That's what the default resource requirements are for, I assume, but as you and many others have noticed, the defaults are not optimal for all systems, nor can we expect that.
I'm anxious to see tasks times. They'll be submitted in a few hours, but elapse times reported in the task list look completely normal.
RE: I now have 2x on both
)
Congratulations! Well done. I took a look and there is a new returned task showing 9627 secs. Partly done at 1x and partly at 2x. One more to go before you get into those done entirely at 2x.
That sounds like a good value if you wish to continue controlling this using % of CPUs method. The disadvantage of this is that BOINC is restricted even if you have no GPU tasks running. I tend to use app_config.xml. If you wanted 4 CPU tasks in addition to 4 GPU tasks you could set to 0.5 and to 1.0 and have BOINC set to use 100% of the CPU cores.
There is!! It's called 'suck it and see' :-). I'm continually amazed that the defaults get it right most of the time. Particularly with AMD CPUs driving AMD discrete GPUs, you do need to experiment to find what works best. I have a Phenom II X4 with a HD7770 GPU. I'm running 3 GPU tasks and 2 CPU tasks. To achieve this at 100% CPUs for BOINC, I use of 0.33 and of 0.67. If you can understand that, you're home and hosed! You can't achieve this 3+2 combination with just the GPU utilization factor without also fiddling with the % CPUs BOINC setting. And, with app_config.xml, it's so quick and easy to change things when you want to test a different combination.
Cheers,
Gary.
You now have several
)
You now have several completed results showing crunch times around 11,500 secs. I thought you might get a bit better than that but at least it's a gain. My 7850s using fglrx (mid 2014) run 4x and do all 4 in about 18,000 - 18,500 secs.
Why can't you run an older fglrx driver if the latest versions give you trouble?
Cheers,
Gary.
I'll consider app_config, but
)
I'll consider app_config, but the documentation still doesn't make clear if, when using it, you get new app versions when they become available without adding them to the file. I don't mind all tasks getting aborted when I make an config file error, but getting the new app version details is a pain and I don't want to have to do that for all my projects, anyway.
I bet you can guess...the older one gives me trouble, too. ;-) The last driver I had working was actually Crimson (5.12) w/ kernel 4.2 & Xorg 1.17. But, I want/need to run kernel 4.3 (or newer) and when I try it, X11 hangs or starts and fails repeatedly preventing me from safely rebooting. I get the same symptoms when I try 5.9 Catalyst, too. I even tried the community patches. I'm working that angle, too, in parallel with this effort to get CLOVER working. But, after crashing my machine six times playing with Catalyst, I had to try it this way. I asked elsewhere and people with 4.4 say it's working and people with other hardware, too, but I haven't got confirmation from anyone with a similar system to mine saying it works. Perhaps I should try again, but I'm going to need a lot more encouragement and hard evidence. Plus, I'm sure Xorg 1.18 isn't supported, so I'd have to downgrade that just to try, but if it doesn't work, upgrade Xorg again just to restore what I have now. I've run ATi/AMD proprietary drivers on Linux since 3DFX died and this is the first time I've ever given up.
Aside from DPMS not working, radeonsi is better in other ways, too, though. Games that never worked with Catalyst now run fine. I don't play much these days, but as a test, it says a lot that the tables have turned.
App_config is very much
)
App_config is very much simpler and more lightweight than app_info.
Yes, you get new versions and even new applications as released, and they will run as normal - version upgrades will probably use the same configuration values, completely new applications will run as if the app_config wasn't present until you get round to updating it. Both app_config and app_info operate project-by-project: there's no need to change any other project if you find one of them useful here.
App_info is the dangerous one which could lead to tasks getting aborted if you get is wrong: I can't even think of a way to kill jobs with app_config.
Well, I'm back. After
)
Well, I'm back. After happily working for 6 months, everything broke two days ago. The APP is complete missing and BOINC deleted 660 files. Did you experience this too? Or, is it just me?
I never did follow up with Paul IE, as I promised. I'm sorry about that. This is disappointing, though, because the academic year just started, and I can't see how I can look into this problem, now.
Someone more knowledgeable
)
Someone more knowledgeable said these files are Gravity Wave search files from a long time ago. So, probably that is not a real problem.
The problem is that there are not matching APPs for my system. Did that happen to anyone else but me? I'm starting to get the feeling it's just me.
Which app? I saw your post in
)
Which app? I saw your post in BRP4G thread and now in BRP6 thread all while the logs mention GW. BRP6 has ended, only timed out/error'd WUs will be resent. FGRPB1 work generator is currently enabled
I'm not thinking of a
)
I'm not thinking of a specific APP. I'm interested in why all of the GPU apps for E@H are not running now. It's not clear why. I posted here because I thought something related to ATi had broken. I see the Tech News post about BRP6 ending. Okay, that explains BRP6, at least.
See this log line:
Sun 02 Oct 2016 03:48:23 PM | Einstein@Home | Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
I assume that refers to all BRP apps, but maybe not. It's also not clear what this message means in this situation. Does this mean there are no WUs for the APPs that are available on my platform, or does it mean there are no BRP APPs for my platform. The latter is implied by the message, but that would be change since two days ago. But, according to the server status page, the BRP4G and BRP6 queues are empty. Am I reading that correctly? If so, there isn't any way to know if it's broken or not at the moment. Right?
But, isn't BRP4 a non-GPU BRP app? So, I still don't understand that log entry.
I also have a message that points me to this URI: https://einsteinathome.org/host/12191882/log . But, I don't know what it's telling me.