Could you run a later version of the 6.10 client? I suspect a CUDA version mismatch or something like that at this point. The latest clients reports the CUDA version at the time it detects the GPU.
Your GPU may not be able to support CUDA 2.3, when you launch your application without the libcudart.so included with the app it finds the one installed on the system which might be older than 2.3. ----- Rom From: Francisco Sanz García [mailto:[email protected]] Sent: Tuesday, October 20, 2009 4:35 AM To: Rom Walton Cc: Bernd Machenschalk; [email protected] Subject: Re: [boinc_dev] Simple cuda aplication. I don't unsderstand it very well: s...@localhost:~/BOINC/samples$ ls projects/test.test/ boinc_finish_called hello_cuda hello_cuda2 init_data.xml input.txt output.txt (I don't need libcudart.so.2.3 here. why?) s...@localhost:~/BOINC/samples$ cat projects/test.test/hello_cuda #!/bin/sh exec ../../projects/test.test/hello_cuda2 "$@" (It works without exec)(hello_cuda2 is the binary) s...@localhost:~/BOINC/samples$ cat client_state_save.xml <client_state> <host_info> <p_fpops>1e9</p_fpops> <p_calculated>2e9</p_calculated> <m_nbytes>2e9</m_nbytes> <d_total>1e11</d_total> <d_free>1e11</d_free> </host_info> <project> <master_url>http://test.test</master_url> <project_name>test_project</project_name> </project> <app> <name>hello_cuda</name> <user_friendly_name>hello_cuda</user_friendly_name> </app> <file_info> <name>hello_cuda</name> <executable/> <status>1</status> </file_info> <file_info> <name>input.txt</name> <status>1</status> </file_info> <file_info> <name>libcudart.so.2.3</name> <executable/> </file_info> <file_info> <name>output.txt</name> <generated_locally/> <upload_when_present/> <max_nbytes>1e6</max_nbytes> </file_info> <app_version> <app_name>hello_cuda</app_name> <api_version>1.1</api_version> <file_ref> <file_name>hello_cuda</file_name> <main_program/> </file_ref> <file_ref> <file_name>libcudart.so.2.3</file_name> </file_ref> <coproc> <type>CUDA</type> </coproc> </app_version> <workunit> <name>hello_cuda_wu</name> <app_name>hello_cuda</app_name> <command_line> --device 0 </command_line> <file_ref> <file_name>input.txt</file_name> <open_name>in</open_name> </file_ref> </workunit> <result> <name>hello_cuda_result</name> <wu_name>hello_cuda_wu</wu_name> <state>2</state> <report_deadline>2e9</report_deadline> <file_ref> <file_name>output.txt</file_name> <open_name>out</open_name> </file_ref> </result> <user_network_request>3</user_network_request> </client_state> and then: (works fine) s...@localhost:~/BOINC/samples$ cp client_state_save.xml client_state.xml ; ./boinc --skip_cpu_benchmarks;20-Oct-2009 10:36:05 [---] Unrecognized tag in cc_config.xml: <skip_cpu_benchmarks/> Unrecognized XML in CONFIG::parse_options: skip_cpu_benchmarks/ 20-Oct-2009 10:36:05 [---] Unrecognized tag in cc_config.xml: <unsigned_apps_ok/> Unrecognized XML in CONFIG::parse_options: unsigned_apps_ok/ 20-Oct-2009 10:36:05 [---] Starting BOINC client version 6.6.37 for i686-pc-linux-gnu 20-Oct-2009 10:36:05 [---] log flags: task, file_xfer, sched_ops 20-Oct-2009 10:36:05 [---] Libraries: libcurl/7.18.0 OpenSSL/0.9.8g zlib/1.2.3 c-ares/1.5.1 20-Oct-2009 10:36:05 [---] Data directory: /home/spac/0910/bifi/cliente_boinc/BOINC_6_6_40/BOINC/samples 20-Oct-2009 10:36:05 [---] Processor: 2 GenuineIntel Intel(R) Core(TM)2 Duo CPU T7250 @ 2.00GHz [Family 6 Model 15 Stepping 13] 20-Oct-2009 10:36:05 [---] Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm pbe nx lm constant_tsc arch_perfmon pebs bts pni dtes64 monitor ds_cpl vmx est tm2 ssse3 cx16 xtpr pdcm lahf_lm ida tpr_shadow vnm 20-Oct-2009 10:36:05 [---] OS: Linux: 2.6.29-2-686 20-Oct-2009 10:36:05 [---] Memory: 1.98 GB physical, 4.66 GB virtual 20-Oct-2009 10:36:05 [---] Disk: 110.00 GB total, 24.74 GB free 20-Oct-2009 10:36:05 [---] Local time is UTC +2 hours 20-Oct-2009 10:36:05 [---] NVIDIA GPU 0: GeForce 8400M G (driver version 0, compute capability 1.1, 127MB, est. 2GFLOPS) 20-Oct-2009 10:36:05 [---] Not using a proxy 20-Oct-2009 10:36:05 [---] Version change (0.0.0 -> 6.6.37) 20-Oct-2009 10:36:05 [test_project] URL: http://test.test/; Computer ID: not assigned yet; location: (none); project prefs: default 20-Oct-2009 10:36:05 [---] No general preferences found - using BOINC defaults 20-Oct-2009 10:36:05 [---] Preferences limit memory usage when active to 953.67MB 20-Oct-2009 10:36:05 [---] Preferences limit memory usage when idle to 1716.61MB 20-Oct-2009 10:36:05 [---] Preferences limit disk usage to 10.00GB 20-Oct-2009 10:36:05 [---] Preferences limit # CPUs to 1 20-Oct-2009 10:36:05 [---] Suspending network activity - user request 20-Oct-2009 10:36:05 [test_project] Starting hello_cuda_result 20-Oct-2009 10:36:05 [test_project] Starting task hello_cuda_result using hello_cuda version 0 20-Oct-2009 10:36:08 [test_project] Computation for task hello_cuda_result finished 2009/10/19 Rom Walton <[email protected]> Could you check if the first argument in hello_cuda2 contains the path to the executable in both cases? This is starting to ring some bells for me, but I had thought we had already fixed this in BOINC. ----- Rom -----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Francisco Sanz García Sent: Monday, October 19, 2009 9:34 AM To: Bernd Machenschalk Cc: [email protected] Subject: Re: [boinc_dev] Simple cuda aplication. Yes, works with exec. 2009/10/19 Bernd Machenschalk <[email protected]> > Francisco Sanz García wrote, On 19.10.2009 14:03 Uhr: > >> mmm, I see it's not necessary to do: >> >> LD_LIBRARY_PATH=../../projects/test.test/, simply wrapping the >> application: >> >> #!/bin/sh >> ../../projects/test.test/hello_cuda2 "$@" >> >> works. >> >> > > Can you confirm that > > #!/bin/sh > exec ../../projects/test.test/hello_cuda2 "$@" > > works, too? > > Bernd > > El 19 de octubre de 2009 13:54, Bernd Machenschalk < >> [email protected]> escribió: >> >> >> >>> Francisco Sanz García wrote, On 19.10.2009 12:58 Uhr: >>> >>> >>> >>>> Ok, then, if the script is: >>>> >>>> #!/bin/sh >>>> echo $PWD > /tmp/libr.log >>>> echo $LD_LIBRARY_PATH >> /tmp/libr.log >>>> LD_LIBRARY_PATH=../../projects/test.test/ >>>> ../../projects/test.test/hello_cuda2 "$@" >>>> echo $PWD >> /tmp/libr.log >>>> echo $LD_LIBRARY_PATH >> /tmp/libr.log >>>> >>>> then: cat libr.log >>>> >>>> ~/BOINC/samples/slots/0 >>>> >>>> >>>> >>>> >>> Hm. >>> >>> Anybody has any clue why >>> >>> >>> >>>> ../../projects/test.test/ >>>> >>>> >>>> >>>> >>> works better than >>> >>> ../../projects/test.test:.:../.. >>> >>> >>>> >>>> >>> as LD_LIBRARY_PATH? >>> >>> Is setting the LD_LIBRARY_PATH actually necessary to make the script >>> work? >>> >>> Or what else is affected by the script? >>> >>> From a different user I know that it works bot if you call the App as a >>> sub-process and exec the App, so the pid doesn't seem to make a >>> difference. >>> >>> >>> Still confused... >>> >>> Bernd >>> >>> >>> >>> >> _______________________________________________ >> boinc_dev mailing list >> [email protected] >> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev >> To unsubscribe, visit the above URL and >> (near bottom of page) enter your email address. >> >> > > _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
