Message boards :
Number crunching :
NVIDIA GeForce GTX Titan
Message board moderation
Previous · 1 · 2 · 3 · 4
Author | Message |
---|---|
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
Has it thrown any more errors since you got that .dll? Cheers. |
Dave Send message Joined: 26 Jun 10 Posts: 18 Credit: 9,976,949 RAC: 0 |
Looks good now, I don't know if it's running optimally but I haven't seen any more issues in the log file. I've been doing a bunch of configuring and reboots, so I'm about 12 hours into it's first continuous stretch. I'd be interested in any pointers for optimizing performance. |
ElricM Send message Joined: 4 Oct 03 Posts: 4 Credit: 607,981,200 RAC: 228 |
Hi there, i have my Titan running for serveral days ... the RAC is dissapointing Today i got 38K WU's crunched :( 10-Mar-2013 23:20:22 [---] No config file found - using defaults 10-Mar-2013 23:20:22 [---] Starting BOINC client version 7.0.28 for windows_x86_64 10-Mar-2013 23:20:22 [---] log flags: file_xfer, sched_ops, task 10-Mar-2013 23:20:22 [---] Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6 10-Mar-2013 23:20:22 [---] Data directory: C:\ProgramData\BOINC 10-Mar-2013 23:20:22 [---] Running under account Volker 10-Mar-2013 23:20:22 [---] Processor: 8 GenuineIntel Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz [Family 6 Model 58 Stepping 9] 10-Mar-2013 23:20:22 [---] Processor: 256.00 KB cache 10-Mar-2013 23:20:22 [---] 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 htt tm pni ssse3 cx16 sse4_1 sse4_2 syscall nx lm vmx smx tm2 popcnt aes pbe 10-Mar-2013 23:20:22 [---] OS: Microsoft Windows 7: Ultimate x64 Edition, Service Pack 1, (06.01.7601.00) 10-Mar-2013 23:20:22 [---] Memory: 3.95 GB physical, 7.90 GB virtual 10-Mar-2013 23:20:22 [---] Disk: 465.66 GB total, 410.97 GB free 10-Mar-2013 23:20:22 [---] Local time is UTC +1 hours 10-Mar-2013 23:20:22 [---] NVIDIA GPU 0: GeForce GTX TITAN (driver version 314.14, CUDA version 5.0, compute capability 3.5, 4096MB, 8384512MB available, 4707 GFLOPS peak) 10-Mar-2013 23:20:22 [---] OpenCL: NVIDIA GPU 0: GeForce GTX TITAN (driver version 314.14, device version OpenCL 1.1 CUDA, 6144MB, 8384512MB available) 10-Mar-2013 23:20:22 [SETI@home] Found app_info.xml; using anonymous platform 10-Mar-2013 23:20:22 [SETI@home] URL http://setiathome.berkeley.edu/; Computer ID 6775938; resource share 100 10-Mar-2013 23:20:22 [SETI@home] General prefs: from SETI@home (last modified 19-Jan-2013 22:22:34) 10-Mar-2013 23:20:22 [SETI@home] Host location: none 10-Mar-2013 23:20:22 [SETI@home] General prefs: using your defaults My app_info.xml <app_info> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>AK_v8b2_win_x64_SSE3.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_v8b2_win_x64_SSE3.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_v8b2_win_x64_SSE3.exe</file_name> <main_program/> </file_ref> </app_version> <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>Lunatics_x41zc_win32_cuda50.exe</name> <executable/> </file_info> <file_info> <name>cudart32_50_35.dll</name> <executable/> </file_info> <file_info> <name>cufft32_50_35.dll</name> <executable/> </file_info> <file_info> <name>mbcuda.cfg</name> </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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</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>0.080000</avg_ncpus> <max_ncpus>0.080000</max_ncpus> <coproc> <type>CUDA</type> <count>0.25</count> </coproc> <file_ref> <file_name>Lunatics_x41zc_win32_cuda50.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>cudart32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>cufft32_50_35.dll</file_name> </file_ref> <file_ref> <file_name>mbcuda.cfg</file_name> </file_ref> </app_version> </app_info> Any ideas ? Greeting from ElricM Nr. 8 in Germany :) |
Wingless Wonder Send message Joined: 14 May 99 Posts: 14 Credit: 12,157,146 RAC: 0 |
<plan_class>cuda_fermi</plan_class>I don't think it will make a difference, timewise, if you change this to 'cuda_kepler'? |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
Hi there, Have patience, RAC is a rolling Average, it'll take many weeks for it to stabilse, several days is too short a time for it to stabilse. Claggy |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Any ideas ? Yes, a small thing if you're hoping to wind things out to the max. I'd add mbcuda.cfg settings as per the supplied example, with the following content: [mbcuda] processpriority = abovenormal pfblockspersm = 15 # maybe even 16 for this one pfperiodsperlaunch = 200 won't be a huge impact, but if OK usability-wise while running 4 at a time like that the subtle CPU usage reduction & GPU usage/latency reduction impact does add up over time. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
Oddbjornik Send message Joined: 15 May 99 Posts: 220 Credit: 349,610,548 RAC: 1,728 |
I understand the process priority, but could you give a hint as to what the other two parameters mean? And perhaps suggest values for a 4GB GTX 680? |
ExchangeMan Send message Joined: 9 Jan 00 Posts: 115 Credit: 157,719,104 RAC: 0 |
Hi there, Actually, based on some rough calculations I've performed on my Titan that's right in the ball park. My first calculations indicated in the 35k-40k range. How many are you running at 1 time? Going from 1 to 2 increases work unit crunching quite a bit. 2 to 3 maybe a couple of % more. |
jason_gee Send message Joined: 24 Nov 06 Posts: 7489 Credit: 91,093,184 RAC: 0 |
Those are pulsefinding settings, controlling how much work gets pushed into the multiprocessors at one time. increased 'blockspersm' takes advantage of superscalar execution capabilities in later model GPUs during pulsefinding. pfperiodsperlaunch controls the length of execution. 15 & 200 are the settings that worked well on my 680 (determined under bench conditions), so reasonable starting points for TITAN given the similarities. Defaults, by necessity of course, are usually conservative to favour machine usability while crunching on more mid-low end GPUs. I would have thought given the new GPU geometry, versus the way new builds fill the GPU, 2 or 3 app instances at one time would be optimal (depending on the machine), though that can vary from machine to machine. With the TITAN GPU itself there are a number of refinements squarely directed at limitations encountered with Kepler 1's, namely unlocked voltage control & better 'turbo boost'. Like 'our' application defaults, I'd imagine nVidia's defaults are conservative, by necessity, to work under wide conditions, so squeezing the most possible out of this card will likely involve leveraging the new controls offered by Turbo Boost 2. In particular the resulting core clock & memory speed would likely have significant effects, especially running 3 or 4 tasks at a time. "Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions. |
ElricM Send message Joined: 4 Oct 03 Posts: 4 Credit: 607,981,200 RAC: 228 |
Currently i have four WU's running on my Titan. Greeting, ElricM |
ElricM Send message Joined: 4 Oct 03 Posts: 4 Credit: 607,981,200 RAC: 228 |
Ok, i will test these parameters :) |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
<plan_class>cuda_fermi</plan_class>I don't think it will make a difference, timewise, if you change this to 'cuda_kepler'? plan_class is only a lable to differentiate between app versions. Makes no difference whatsoever to the actual crunching by the app. Work arriving is assigned a plan_class - that in turn is used by boinc to choose the correct app. Only makes a difference in stock, when there are different apps to choose from. For us, having all the lables in place is neccesaay to pick up all work that might possibly be present on the rig that is running an installer from a stock state. A person who won't read has no advantage over one who can't read. (Mark Twain) |
Richard Haselgrove Send message Joined: 4 Jul 99 Posts: 14650 Credit: 200,643,578 RAC: 874 |
<plan_class>cuda_fermi</plan_class>I don't think it will make a difference, timewise, if you change this to 'cuda_kepler'? It's probably a good idea not to move away from the project's pre-defined version numbering and plan_class allocations. As William says, it'll make no difference at all while you're running a single application: but if you ever change to a different application in the future, you'll lose all your cached work, unless you remember exactly what changes you've made, and apply them to the app_info stub for the new application before installing it |
William Send message Joined: 14 Feb 13 Posts: 2037 Credit: 17,689,662 RAC: 0 |
<plan_class>cuda_fermi</plan_class>I don't think it will make a difference, timewise, if you change this to 'cuda_kepler'? So, in principle you can pick any version number and plan class you like. If you wanted you could place an entry starting <app_version> <app_name>setiathome_enhanced</app_name> <version_num>999</version_num> <platform>windows_intelx86</platform> <plan_class>foobar</plan_class> at the top of you app_info.xml and all new work would be assigned a version number of 9.99 and a plan class of foobar [and it would show up like that in the manager]. However your old work would be marked <version_num>610</version_num> <platform>windows_intelx86</platform> <plan_class>cuda_fermi</plan_class> so 6.10/cuda_fermi and if you don't have an entry in your app_info.xml that says exactly that, all the tasks currently on the host would be discarded by BOINC - and that would never be reported back to the server! They would time out after a fashion. Of course with resend_lost_results in place, on the next server contact the server notices the missing tasks and will resend them in batches of 20. To provide for work that might possibly be on the host is the sole reason for that horribly bloated app_info.xml with multiple app_version entries. If you wanted you could trim it down to one entry per app. So as Richard has mentioned, if you were to change those entries and then to rerun the installer at a later time [or use the next one out], you would lose the altered tags and work on the host would no longer be recognised. We had a lot of fun when a Lunatics dev decided it was good practise to increase the version number with a fresh optimised app release... NB Don't try this at home ;) A person who won't read has no advantage over one who can't read. (Mark Twain) |
©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.