Questions and Answers :
GPU applications :
cuda & akv8 app_info.xml
Message board moderation
Author | Message |
---|---|
Noman Send message Joined: 15 Aug 09 Posts: 6 Credit: 528,646 RAC: 0 |
I have searched the forums but was unable to find any definitive answer. I have a box that I want to run the optimized AK v8 ssse3 binary but it also has a CUDA capable nvidia card. It sees the cuda card but never gets any tasks. here is my current app_info.xml <app_info> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>AK_V8_linux64_ssse3</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>528</version_num> <file_ref> <file_name>AK_V8_linux64_ssse3</file_name> <main_program/> </file_ref> </app_version> <app> <name>astropulse_v505</name> </app> <file_info> <name>ap_5.06r168_sse3_x86_64-pc-linux-gnu</name> <executable/> </file_info> <app_version> <app_name>astropulse_v505</app_name> <version_num>506</version_num> <file_ref> <file_name>ap_5.06r168_sse3_x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> </app_version> </app_info> <app_info> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</name> <executable/> </file_info> <file_info> <name>libcudart.so.2</name> <executable/> </file_info> <file_info> <name>libcufft.so.2</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>528</version_num> <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>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>603</version_num> <plan_class>cuda</plan_class> <avg_ncpus>1.0000</avg_ncpus> <max_ncpus>1.0000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>605</version_num> <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>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>606</version_num> <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>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>607</version_num> <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>setiathome-CUDA-6.08.x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> </app_info> |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
if it helps any here is mine. i am not running ap and i cut out all the extraneous versions to simplify it. too many version entries not only get confusing but can also be that many more lines for syntax errors. <app_info> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>AK_V8_linux64_ssse3</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>603</version_num> <file_ref> <file_name>AK_V8_linux64_ssse3</file_name> <main_program/> </file_ref> </app_version> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>setiathome-6.08.CUDA_2.2_x86_64-pc-linux-gnu</name> <executable/> </file_info> <file_info> <name>libcudart.so.2</name> <executable/> </file_info> <file_info> <name>libcufft.so.2</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>608</version_num> <plan_class>cuda</plan_class> <avg_ncpus>0.350000</avg_ncpus> <max_ncpus>0.350000</max_ncpus> <coproc> <type>CUDA</type> <count>1</count> </coproc> <file_ref> <file_name>setiathome-6.08.CUDA_2.2_x86_64-pc-linux-gnu</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2</file_name> </file_ref> <file_ref> <file_name>libcufft.so.2</file_name> </file_ref> </app_version> </app_info> also you should have the cuda libraries (from the appropriate cuda 2.2 or 2.3 toolkit to match your driver) located in the same seti data directory as your application and this xml file as well as a copy of the application executable in the binary path of the machine, typically /usr/bin.. be VERY sure you have the seti data path included in /etc/ld.so.conf and that you have run ldconfig to update the library search paths. you can verify all is well by changing into the seti directory and doing ldd <seti-app-filename> and you should see an unbroken list of libraries. if so then it should be good to go. also in the data dir place a txt file with no extension called number_of_gpus and place a 1 in the first character of the first line if you run 1 cuda card or a 2 if you run 2 cuda cards. then in the boinc directory make sure you have a cc_config.xml containing <cc_config> <log_flags> <coproc_debug>1</coproc_debug> </log_flags> <options> <use_all_gpus>1</use_all_gpus> </options> </cc_config> and it should 'just work'. :) also make sure you are running a recent version of boinc. early versions had various breakages concerning cuda. for 1 gpu, 6.6.36 works well , for 2 gpus, try 6.9.x or 6.10.x. the 6.6.36 has problems with 2 gpus (at least in linux it does) |
Noman Send message Joined: 15 Aug 09 Posts: 6 Credit: 528,646 RAC: 0 |
I have updated my app_info.xml and added the stuff to cc_config.xml but have yet to see any cuda tasks or any helpful coprocessor debug messages. I will wait a day or two and update. Thanks |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
I have updated my app_info.xml and added the stuff to cc_config.xml but have yet to see any cuda tasks or any helpful coprocessor debug messages. I will wait a day or two and update. ok here's hoping.... when mine did that i discovered i didn't have the number_of_gpus file present so it just turned getting workunits off. |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
I have updated my app_info.xml and added the stuff to cc_config.xml but have yet to see any cuda tasks or any helpful coprocessor debug messages. I will wait a day or two and update. boinc doesn't say something like it is not going to use the card it finds does it? what card is it, and more importantly how much vidram? i have never been able to get cuda to work on a 256mb card while using it for my desktop at the same time, simply not enough memory. |
Noman Send message Joined: 15 Aug 09 Posts: 6 Credit: 528,646 RAC: 0 |
ok I am running Boinc 6.6.36 I am not running anything on this computer but boinc/seti From messages: "Configured to use all coprocessors" "CUDA device: GeForce 8400 GS (driver version 0, compute capability 1.1, 255MB, est. 5GFLOPS)" I did not have a number_of_gpus file so I created it so it only has a 1 the scheduler is not even asking for GPU tasks and there is not cuda device in my client_state.xml |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
ok if you are using that card for your desktop as well chances are it thinks there is not enough memory to run cuda so it is ignoring it. |
Noman Send message Joined: 15 Aug 09 Posts: 6 Credit: 528,646 RAC: 0 |
that could be but the only graphics is a linux commandline, maybe I pop in another card to do the desktop. |
Noman Send message Joined: 15 Aug 09 Posts: 6 Credit: 528,646 RAC: 0 |
its working now! so I will mark this answered and maybe open another because it is REAL slow. |
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
its working now! so I will mark this answered and maybe open another because it is REAL slow. 5 gflops will be slow. |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
|
Chuck Gorish Send message Joined: 19 Jun 00 Posts: 156 Credit: 29,589,106 RAC: 0 |
Hi, a recent C2D or C2Q processor, has average of 2,0 - 3,7 GFLOPS per core. true, but i believe he is looking at average cuda wu times he sees around and when comparing them 5gflops will be a bit slower. it still will process a wu faster than a cpu will but it will be hard pressed to do more than 1 wu in 1 to 1.5 hrs when a cpu might take 1.5 to 2.5 hrs for the same kind of wu. boinc reports my gtx285 card at 127 gflops and my tesla at 74 gflops and i only get 2 to 5 wu /hr each out of them (going by the elapsed time in boinc). when i had my 8600gt running as my 2nd card i was lucky to get 1 wu every 1 to 1.5 hrs out of it. that was reported by boinc as 14 gflops. |
©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.