Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Rob.B on 29 Apr 2009, 10:00:14 pm
-
During the latest Seti work famine I ran down my cashe and moved to 6.6.20. I was using the V10 team mod app's, but I believe these should not be used with 6.6.15 upwards.
So what should I be using and where do I find it.
There are lots of versions flying around and I am utterly confused as to what I should download and install.
Rob.B
-
Take latest CUDA MB build + use app_info template file from SETI main forum.
As alternative you can use already downloaded V10 CUDA MB but remove teamed CPU apps,replace app_info and remove ncpus field from cc_config.xml
-
Thanks for that pointer, I'll give it a go. Currently I have only AP running OP app with an APP_Info I sorted my selt, but cant get non CUDA MB. CUDA is running stock app's.
A bit of a mess really, as I am not an AP fan.
Rob.
-
Take latest CUDA MB build + use app_info template file from SETI main forum.
Sorry for being stupid but I can't find this template, or the "Seti main forum". I have even looked on the seti@home forums to no advantage.
Rob.
-
Take latest CUDA MB build + use app_info template file from SETI main forum.
Sorry for being stupid but I can't find this template, or the "Seti main forum". I have even looked on the seti@home forums to no advantage.
Rob.
http://setiathome.berkeley.edu/forum_thread.php?id=52589&nowrap=true#876502
"app_info for AP500, AP503, MB603 and MB608"
-
Thaks you for the link Richard.
Rob.
-
Followed the instructions to the letter and blew away my Boinc Manager! Won't start properly, can't unintall, cant repair, sant shut the client. Well and truly screwed!
Tried butting back the pre op app structure, no go. Ended up having to physically delete both boinc directory structures, then uninstall, then reinstall. Now back at 6.6.20 stock apps but seeing only one GPU. I think I'll go back to 6.4.7 again. (This was my second try at movong to 6.6.20).
Thanks anyway.
Rob.
-
Third time lucky, with a clean install I have followed the template instructions and now the AP's are running OP-APP and MB is running stock CUDA, but only one CUDA is being used (I have two).
I have tried amending this entry in the app_info file from one to two, but when set to two no CUDA work starts.
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
I have even reinstated cc_config with an entry: <use_all_gpus>1</use_all_gpus>
No luck.
This seems far more complex than it need to be. Any ideas on this one?
-
Don't touch <count> field. It's for another purpose.
What log of your BOINC client at startup?
-
OK, reset that field.
The log shows 1 GPU, I downgraded back to 6.4.7 and it see's two GUP's again. Now back at 6.6.20 and only one GPU is recognised. It's all very strange.
I'm going to be working away for a week or two, so my wife will be keeping an eye on my server, so I'm going back to 6.4.7 until I get back.
Rob.
-
OK one last try.
reboot, update drivers for GUP's, last update about five or six weeks ago. check 6.4.7 can see and use both cards. OK save directories, remove app_info.xml and nogupu file;upgrade to 6.6.20! As you can see from the log it now sees both GPUS but has trashed all MB work, none in progress as I ran them to completeion first and reported before trying this. (Not that i don't trust this move to 6.6.20 or anything). Four MB WU's lost.
Now what? It's running on stock app's but I'm doing nothing until I can get MB running on both CUDA and CPU.
03/05/2009 09:34:06 Starting BOINC client version 6.6.20 for windows_intelx86
03/05/2009 09:34:06 log flags: task, file_xfer, sched_ops
03/05/2009 09:34:06 Libraries: libcurl/7.19.4 OpenSSL/0.9.8j zlib/1.2.3
03/05/2009 09:34:06 Data directory: C:\ProgramData\BOINC
03/05/2009 09:34:06 Running under account Robert
03/05/2009 09:34:07 Processor: 4 AuthenticAMD AMD Phenom(tm) 9850 Quad-Core Processor [x86 Family 16 Model 2 Stepping 3]
03/05/2009 09:34:07 Processor features: fpu tsc pae nx sse sse2 pni 3dnow mmx
03/05/2009 09:34:07 OS: Microsoft Windows Vista: Ultimate x86 Editon, Service Pack 1, (06.00.6001.00)
03/05/2009 09:34:07 Memory: 3.00 GB physical, 6.20 GB virtual
03/05/2009 09:34:07 Disk: 465.76 GB total, 364.36 GB free
03/05/2009 09:34:07 Local time is UTC +1 hours
03/05/2009 09:34:07 CUDA devices: GeForce 9800 GTX/9800 GTX+ (driver version 18250, CUDA version 1.1, 512MB, est. 86GFLOPS), GeForce 9400 GT (driver version 18250, CUDA version 1.1, 512MB, est. 8GFLOPS)
03/05/2009 09:34:07 SETI@home [error] No app version for result: windows_intelx86 608 cuda
03/05/2009 09:34:07 SETI@home [error] No app version for result: windows_intelx86 608 cuda
03/05/2009 09:34:07 SETI@home [error] No app version for result: windows_intelx86 608 cuda
03/05/2009 09:34:07 SETI@home [error] State file error: result 07fe09ac.32724.72.8.8.255_0 not found
03/05/2009 09:34:07 SETI@home [error] State file error: result 07fe09ac.32724.481.8.8.36_1 not found
03/05/2009 09:34:07 Not using a proxy
03/05/2009 09:34:07 Version change (6.4.7 -> 6.6.20)
03/05/2009 09:34:07 SETI@home URL: http://setiathome.berkeley.edu/; Computer ID: 4793552; location: home; project prefs: home
03/05/2009 09:34:07 GPUGRID URL: http://www.gpugrid.net/; Computer ID: 34718; location: (none); project prefs: default
03/05/2009 09:34:07 SETI@home General prefs: from SETI@home (last modified 14-Mar-2009 05:57:17)
03/05/2009 09:34:07 SETI@home Computer location: home
03/05/2009 09:34:07 General prefs: using separate prefs for home
03/05/2009 09:34:07 Reading preferences override file
03/05/2009 09:34:07 Preferences limit memory usage when active to 1534.79MB
03/05/2009 09:34:07 Preferences limit memory usage when idle to 2762.62MB
03/05/2009 09:34:07 Preferences limit disk usage to 10.00GB
03/05/2009 09:34:07 Preferences limit # CPUs to 4
03/05/2009 09:34:08 Running CPU benchmarks
03/05/2009 09:34:08 Suspending computation - running CPU benchmarks
03/05/2009 09:34:08 SETI@home Sending scheduler request: Requested by user.
03/05/2009 09:34:08 SETI@home Not reporting or requesting tasks
03/05/2009 09:34:33 SETI@home Scheduler request completed: got 0 new tasks
03/05/2009 09:34:39 Benchmark results:
03/05/2009 09:34:39 Number of CPUs: 4
03/05/2009 09:34:39 2530 floating point MIPS (Whetstone) per CPU
03/05/2009 09:34:39 5214 integer MIPS (Dhrystone) per CPU
03/05/2009 09:34:40 SETI@home Restarting task ap_21mr09ab_B2_P0_00098_20090501_22613.wu_1 using astropulse_v5 version 503
03/05/2009 09:34:40 SETI@home Restarting task ap_20mr09ae_B3_P0_00213_20090501_29643.wu_2 using astropulse_v5 version 503
03/05/2009 09:34:40 SETI@home Restarting task ap_21mr09ab_B2_P0_00100_20090501_22613.wu_0 using astropulse_v5 version 503
03/05/2009 09:34:40 SETI@home Restarting task ap_21mr09ab_B2_P0_00096_20090501_22613.wu_0 using astropulse_v5 version 503
-
Well, both GPU available so you should be able to use them both now.
Try to compose correct app_info then update to opt apps.
-
Used the same app_info as last time, working fine now. AP's and MB on CPU using opp app's. When I swap in the optimised CUDA app', should I swap the dll's as well?
Rob.
-
Not nessesary.
-
Raistmer,
Is that I have no need to move to OP App for CUDa, or I do not need to swap the dll's? Sorry to ask but this move has fried my brain!
If I should move to OP app for cuda, which version should I use?
Thanks again.
Rob.
-
OK just rebooted (part of a non-crunch app upgrade), and I'm back to one GPU! Vista still see's two GPU's. Is this an issue with 6.6.xx or what?
I'm fast getting fed up of the grief this migration has caused, no wonder Seti is loosing CPU's to other projects. I wonder how many have given up and moved to GPUGrid, and/or Einstein?.
Rob.
-
This is silly!
I can swap from one to two GPU's as follows:
1. Disable/susspend all GPU based projects; reboot. Result 2 GPU's
2. Disable/suspend SETI leave GPUGRID (only one task) active; reboot. Result 1 GPU.
3. Disable/suspend GPUGRID, leave SETI enabled (many CUDA tasks). Result 2 GPU's.
4. Download extra GPUGRID WU's to provide more than on task at reboot. Result 2 GPU's.
This is a Boinc client thing!!!!!
No wonder the transition has been hellish, if it depends on what is available at reboot.
Note: When adding op apps for Milkyway (from ZSLIP - non GPU) 6.6.20 deleted all CUDA (both SETI and GPUGRID) tasks and downloaded a new set's, the deleted tasks still show as active in my task list on the web site. (P.S the Milkyway tasks did not vanish and the op app is working fine.)
Not happy with this version al all!
-
report on BOINC forum, there are no BOINC devs AFAIK...