Gee, I see this forum is not very crowded... :-\
I hope someone can help me with this.
I have an old computer with SSE2 CPU and a CUDA capable Nvidia card (512MB video RAM so it should be able to run seti apps). Right now it is running stock apps but I would like to run optimized apps instead.
I am unsure however which one to download. For example, for Astropulse I see there is an SSE2 application and a CUDA application. Problem is, both contain a app_info file which seems to direct ap works at them exclusively (that is if I use CUDA app_info file, only GPU is getting those works and not CPU, and vice versa).
Same applies with MB apps.
Which one do I choose to have both GPU and CPU work on tasks?
Thanks!
Hey,
What distribution and video driver are you running ? Are you OK with some manual gymnastics there?
Apart from the downloads I see here, a number of Linux users for Cuda multibeam so far build straight from my Berkeley commits lately, but certainly far from a good 'easy' solution. I'm running a Cuda 3.2 build for the Cuda multibeam, under Ubuntu x64, that I can try make available & would probably work quite well (assuming distro & driver are OK).
For CPU multibeam and Astropulse, I think there is something there from Lunatics/Urs-Echternacht in downloads, I haven't been watching those developments lately though, so can't describe their status/compatibility.
Whatever combinations of applications you choose, combining the app_info.xml files into one file, and make sure there is only one outer set of <app_info></app_info> tags surrounding the lot, followed by a boinc restart, or machine restart, should do the trick. Can look pretty complex with a lot of apps, so picking one first to start might be on option.
Quote from: Ruskie on November 30, 2013, 08:21:19 PM
I am unsure however which one to download. For example, for Astropulse I see there is an SSE2 application and a CUDA application. Problem is, both contain a app_info file which seems to direct ap works at them exclusively (that is if I use CUDA app_info file, only GPU is getting those works and not CPU, and vice versa).
Same applies with MB apps.
Which one do I choose to have both GPU and CPU work on tasks?
Each Linux download has an example_app_info_files folder with different app_info combinations, CPU MB, CPU AP, OpenCL ATI MB, OpenCL ATI AP etc, (But not Cuda),
So you can choose one of those to start with.
Claggy
Quote from: Jason G on November 30, 2013, 09:00:24 PM
Hey,
What distribution and video driver are you running ? Are you OK with some manual gymnastics there?
If I weren't, I wouldn't have installed Linux at all. ;)
I have a freshly installed Gnome Ubuntu 13.10 headless server which I operate via putty/vncserver, though I had to move to Kde 'cause gnome-session is broken in 13.10.
I used to run CentOS as I am an SAP employee and wanted to have an environment I could easily install SAP instances on, but as with most enterprise Linuxes, it has far too outdated repositories and it was too much a hassle to have BOINC run there.
Quote
Apart from the downloads I see here, a number of Linux users for Cuda multibeam so far build straight from my Berkeley commits lately, but certainly far from a good 'easy' solution. I'm running a Cuda 3.2 build for the Cuda multibeam, under Ubuntu x64, that I can try make available & would probably work quite well (assuming distro & driver are OK).
For CPU multibeam and Astropulse, I think there is something there from Lunatics/Urs-Echternacht in downloads, I haven't been watching those developments lately though, so can't describe their status/compatibility.
Whatever is optimized sounds good for me! Just explain how to make it work!
What I was wondering in the first hand, however, was just that. That I installed stock boinc, and then Lunatics app with an installer, and I see 1/2 CUDA threads working on seti@home WU, and 8 CPU threads, which of course alternate with other BOINC projects. From that I guess it should be possible to run more than one task type on the same machine but:
Quote
Whatever combinations of applications you choose, combining the app_info.xml files into one file, and make sure there is only one outer set of <app_info></app_info> tags surrounding the lot, followed by a boinc restart, or machine restart, should do the trick. Can look pretty complex with a lot of apps, so picking one first to start might be on option.
...but that's my "problem". This was precisely what I guessed. Now I admit I just looked at all the app_info quickly, therefore I might have missed something but it looked to me that all the app_info had "the same tags"... I mean there were one app for AP work units in CPU and one for AP work units on GPU that had the same "astropulse_v6" tags.
HOWEVER.
However I must be probably drunk because I cannot currently find them anymore - changed browsing platform but cannot see how this would change anything...
I now see only one AP optimized app suitable for my system, so I guess it is all right.
I should have found 1 AP app, 1 MB cpu app and 1 MB gpu app, I'll try to join everything in their app_info file (besides openin and closing tags) and see it that works!
Quote from: Claggy on November 30, 2013, 09:36:06 PM
Each Linux download has an example_app_info_files folder with different app_info combinations, CPU MB, CPU AP, OpenCL ATI MB, OpenCL ATI AP etc, (But not Cuda),
So you can choose one of those to start with.
Claggy
Found them. I will use those example to see how you can combine multiple app_info files into one. Thanks!
For example, this one: _app_info_apcpu+mbcpu.xml should be a good starting point, to which I should only add CUDA infos.
Quote from: Ruskie on November 30, 2013, 09:54:20 PM
For example, this one: _app_info_apcpu+mbcpu.xml should be a good starting point, to which I should only add CUDA infos.
Yep, AP and MB cpu should be a good starting point. post your proposed app_info.xml if you feel it warrants checking.
I'll try scare up a recent Cuda multibeam build as soon as I can get into that machine without interruptions from the Koalas getting into my sewer pipes.
Quote from: Jason G on November 30, 2013, 10:47:41 PM
Yep, AP and MB cpu should be a good starting point. post your proposed app_info.xml if you feel it warrants checking.
I'll try scare up a recent Cuda multibeam build as soon as I can get into that machine without interruptions from the Koalas getting into my sewer pipes.
Lol. I will. I am off to a concert photo shooting now, so no hurry, I won't have the time to hack into my linux box for a couple days at best.
Thank you!
Are you talking about the Q8300 with the GT210?
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7152899
That machine actually has SSSE3 on it.