Questions and Answers :
GPU applications :
More than one WU per GPU at the same time ?
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 · Next
Author | Message |
---|---|
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
1) If you "have followed this thread" you had to know that (post #4): Your GeForce GTX 480 (1503MB) can run 2 or 3 tasks at the same time (don't try this with GeForce GT 220 - it can't) 2) There are several places in app_info.xml where you need to change <count>1</count> to: <count>0.5</count> (but your NVIDIA GeForce GTS 240 (1024MB) is incapable to run efficiently more than 1 task, only Fermi NVIDIA GPUs can (4xx, 5xx)) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
I have done this but it only runs one task at a time anyway no matter how efficiently. it does a seti app in about 30 minutes while using MB of vram. why, even when I change the app_info.xml file does it not run 2 tasks at a time? It's like it just ignores it. is there something else I need to do like put the lunatics file in a specific place or run a specific program? |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
first off you have an nVidia 240 so there is no benefit to running more than 1 WU at a time. I assume you restarted the boinc after you changed your app_info. If so change your count back to 1 and leave it there. you'll need an nVidia 4XX or 5XX card to run 2 or more WU's or an ATI 5XXX or 6XXX card In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Sunny129 Send message Joined: 7 Nov 00 Posts: 190 Credit: 3,163,755 RAC: 0 |
did you also change <cmdline>-period_iterations_num 2 -instances_per_device 1</cmdline>to <cmdline>-period_iterations_num 2 -instances_per_device 2</cmdline> ? |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
No Where the hell is that? I cannot find it ------------------------ The reason I want to try this is for an experiment, i know that it does not run efficiently according to this thread Also I may be getting a gtx 570 soon so I want this worked out first. |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
did you also change <cmdline>-period_iterations_num 2 -instances_per_device 1</cmdline>to <cmdline>-period_iterations_num 2 -instances_per_device 2</cmdline> ? What does this do? |
Sunny129 Send message Joined: 7 Nov 00 Posts: 190 Credit: 3,163,755 RAC: 0 |
No hmm...that's interesting. here is my app_info.xml file for S@H (just the Multibeam section):
i could be wrong, but i was under the impression that changing only one or the other wasn't enough to get your GPU to run 2 tasks simultaneously, and that you had to edit both values ("count" & "instances per device"). if you don't have "instances per device" in the <cmdline> section of your app_info.xml, just add it. that should get 2 WU's running at the same time. the same thing goes for trying to run 3 at a time - "count" would have to be 0.33 and "instances per device" would have to be 3. |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
ok, that is weird your looks alot different to mine i have put the whole app_info file in <app_info> <app> <name>astropulse_v505</name> </app> <file_info> <name>ap_5.05r409_SSE.exe</name> <executable/> </file_info> <app_version> <app_name>astropulse_v505</app_name> <version_num>505</version_num> <platform>windows_intelx86</platform> <file_ref> <file_name>ap_5.05r409_SSE.exe</file_name> <main_program/> </file_ref> </app_version> <app_version> <app_name>astropulse_v505</app_name> <version_num>505</version_num> <platform>windows_x86_64</platform> <file_ref> <file_name>ap_5.05r409_SSE.exe</file_name> <main_program/> </file_ref> </app_version> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>AK_v8b_win_x64_SSE41.exe</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>603</version_num> <platform>windows_intelx86</platform> <file_ref> <file_name>AK_v8b_win_x64_SSE41.exe</file_name> <main_program/> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>603</version_num> <platform>windows_x86_64</platform> <file_ref> <file_name>AK_v8b_win_x64_SSE41.exe</file_name> <main_program/> </file_ref> </app_version> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>Lunatics_x38g_win32_cuda32.exe</name> <executable/> </file_info> <file_info> <name>cudart32_32_16.dll</name> <executable/> </file_info> <file_info> <name>cufft32_32_16.dll</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>610</version_num> <platform>windows_intelx86</platform> <plan_class>cuda_fermi</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>609</version_num> <platform>windows_intelx86</platform> <plan_class>cuda23</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>608</version_num> <platform>windows_intelx86</platform> <plan_class>cuda</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>610</version_num> <platform>windows_x86_64</platform> <plan_class>cuda_fermi</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>609</version_num> <platform>windows_x86_64</platform> <plan_class>cuda23</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>608</version_num> <platform>windows_x86_64</platform> <plan_class>cuda</plan_class> <avg_ncpus>0.040000</avg_ncpus> <max_ncpus>0.040000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>Lunatics_x38g_win32_cuda32.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_32_16.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_32_16.dll</file_name> </file_ref> </app_version> </app_info> As you can see i have cuda count as one. I change that to .5 all the time and save the file but it somehow changes back. i will try what you have suggested. Thanks |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
Also, if i were to dedicate my machine to gpu computing would i be better off with a Tesla c2075 than a GeForce product? |
Sunny129 Send message Joined: 7 Nov 00 Posts: 190 Credit: 3,163,755 RAC: 0 |
well it just came to my attention that you're crunching with a CUDA card, whereas i'm crunching on an ATI card. so aside from the fact that i'm only running 1 application version for Multibeam (v6.10) and you're running 6 versions (v6.08, v6.09, v6.10, v6.08x64, v6.09x64, and v6.10x64), there may be some additional differences in app_info.xml code simply due to us having GPUs based on completely different architectures. now that i'm aware that your running a CUDA card and that i have no experience with app_info.xml files for CUDA crunching, i'm not sure if the <cmdline>-period_iterations_num 2 -instances_per_device 2</cmdline> is necessary, or if it'll have any adverse affects. i'm hoping some folks with a bit more experience can chime in here and verify whether or not this would be a necessary step. |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
Ok, thank you anyway. |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
I don't think those parameters will do anything since they weren't built into the stock app. Now, Raistmer may have mentioned that those parameters are available in his readme of install instructions. I don't know since I don't follow much of the Cuda builds In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
1) -instances_per_device is used only for OpenCL ATI apps (and nVidia OpenCL AstroPulse) (made by Raistmer) CUDA NVIDIA apps (made by jason_gee) do not have/use it 2) you need to change <count>1</count> to: <count>0.5</count> in 6 places and then restart BOINC to make it read the file again. (but this is so obvious) 3) As you can see i have CUDA count as one. It can't "change back" by itself as BOINC only reads that file and never write to it. Possible: - If you reinstall Lunatics' Unified Installer (Lunatics_Win64_v0.38_setup.exe) it will create new app_info.xml (overwrite the existing) - maybe you save the new file to different place or different name (e.g. app_info.xml.txt) Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
I will try this when possible because I am currently running win 7 and I do not have admin rights (mum does)so i can only do this when mum is logged in and forgets to log off. Does anyone know how to find out an admin password when your a standard user because that would be very helpful in making my gpu do multiple wu's. |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
My mother is also very against scientific computing so that makes this all the more difficult because if she catches me well... |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Then leave the computer alone, as we already told you several times - this GPU is too weak for 2 tasks at once "your NVIDIA GeForce GTS 240 (1024MB) is incapable to run efficiently more than 1 task, only Fermi NVIDIA GPUs can (4xx, 5xx)" Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
skildude Send message Joined: 4 Oct 00 Posts: 9541 Credit: 50,759,529 RAC: 60 |
And you need to remember that your EULA states that you are forbidden from using the BOINC process on Computers where you do not have permission. Stealing admin passwords and computing in the manner you suggest tells us you should stop immediately and ask for permission In a rich man's house there is no place to spit but his face. Diogenes Of Sinope |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
Whatever I will figure this our on my own, thank you all for the help and advice. |
Aaron Send message Joined: 19 Jan 11 Posts: 17 Credit: 226,460 RAC: 0 |
my gpu (gts 240) is not so ineffiecient after all it completed 2 seti tasks at the same time in 7 minutes, which is weird because it takes 14 minutes to complete one task at a time. But anyway, i figured it out and now im happy :) so you guys gave me wrong advice :/ |
BilBg Send message Joined: 27 May 07 Posts: 3720 Credit: 9,385,827 RAC: 0 |
Your last 2 tasks are short tasks: WU true angle range is : 2.724356 When you get "normal" tasks with: WU true angle range about 0.42 ... then check the speed and speak again about the "wrong advice" The speed will be slower as reported over the years by numerous users of even GTX 280 or 295 Search and read the forums. If "it takes 14 minutes to complete one task at a time" it will take >30 minutes for 2 "normal" tasks at a time (even more). And maybe task with WU true angle range of 2.724356 will be computed in 3 minutes if only one at a time? With only "Number of tasks completed 9" using "SETI@home Enhanced (anonymous platform, nvidia GPU)" you don't have enough statistics to make conclusions: http://setiathome.berkeley.edu/host_app_versions.php?hostid=6234616 ;) As in old joke - you may think that "Scottish sheep are black" but the truth may be only that "In Scotland, there is at least one sheep, at least one side of which appears to be black from here" ;) http://en.wikipedia.org/wiki/Mathematical_joke#Stereotypes_of_mathematicians An astronomer, a physicist and a mathematician are on a train in Scotland. The astronomer looks out of the window, sees a black sheep standing in a field, and remarks, "How odd. Scottish sheep are black." "No, no, no!" says the physicist. "Only some Scottish sheep are black." The mathematician rolls his eyes at his companions' muddled thinking and says, "In Scotland, there is at least one sheep, at least one side of which appears to be black from here." P.S. And what exactly you needed to "figure it out"?? What was the problem? As I said: "It (app_info.xml) can't "change back" by itself as BOINC only reads that file and never write to it." Â - ALF - "Find out what you don't do well ..... then don't do it!" :) Â |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.