| Index | Recent Threads | Unanswered Threads | Who's Active | Guidelines | Search |
 |
World Community Grid Forums
Category: Support
Forum: GPU Support Forum
Thread: Dual GPU Setup (CPU Cores & GPU Task # Questions) |
| No member browsing this thread |
|
Thread Status: Active Total posts in this thread: 66
|
|
| Author |
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I am getting closer. Thanks underwater for getting me this close. below....4 core cpu with 2 x NVidia cards running 3 tasks per gpu....(580's) as a bonus has 1 x cpu task running concurrently a (2500k) cpu only task <app_config> <app> <name>hcc1</name> <max_concurrent>6</max_concurrent> <gpu_versions> <gpu_usage>.33</gpu_usage> <cpu_usage>.66</cpu_usage> </gpu_versions> </app> </app_config> Your app_config would have 4 tasks per gpu, at 2 tasks per cpu thread - understood ?? For Nvidia cards this is maxed out !!!! (big time) You should try 3 tasks per gpu (as above) but errors still may occur, in which case you need to drop back to 2 tasks per gpu.... do the math. My dual 580s work just fine at 3 tasks per card, but inferior cards ?? |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
underwater, out of curiosity, does that <cpu_usage>.66</cpu_usage> result in 3 GPU tasks being managed by 2 CPU cores? My math, sorry, says that 6 CPU task with 0.66 CPU allocation each, uses 3.96 cores. Suppose that the alternating [random asynch] processing frees up 1 non-hcc CPU thread. I'd be tempted to drop the setting to <cpu_usage>.5</cpu_usage> i.e. 2 cards, 6 GPU tasks with 2 GPU threads each assigned to a CPU core. Is the 2500 hyperthreaded?
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
I have found that majority of errors are caused during cpu stage with Nvidia cards, so I devoted as much cpu as possible to each task, the extra cpu task gained with this config was just a by-product which resulted unexpectedly.
----------------------------------------NON hyper threading, 4 core 2500K cpu with 2 x gpu cards running 3 tasks each should = .666 per cpu thread (4 cpu threads divided by 6 gpu tasks) and .333 per gpu task. (YES 3.996) and see the bonus 7th task running as cpu task. Here is the running app_config with bonus cpu task running  [Edit 1 times, last edit by Former Member at Mar 21, 2013 4:18:55 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Definitely interesting math... your system has no problem with this config, it appears. As suggested, you may want to try 0.5 on the CPU value. Could be your spare core starts to run jobs faster
----------------------------------------edit: Note sure, but this could already be reflected in the Elapsed / CPU time differential of the non HCC jobs. As far as scheduling is concerned GPU jobs get priority over CPU jobs in BOINC, but that's side info. [Edit 1 times, last edit by Former Member at Mar 21, 2013 4:19:00 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Sir Underwater - I just converted my i7-3930K to Win7.
I install the version 7.0.58 (x64). My app_config.xml <app_config> <app> <name>hcc1</name> <max_concurrent>32</max_concurrent> <gpu_versions> <gpu_usage>0.25</gpu_usage> <cpu_usage>1.00</cpu_usage> </gpu_versions> </app> The Event Log: 3/24/2013 4:29:38 PM | | Starting BOINC client version 7.0.58 for windows_x86_64 3/24/2013 4:29:38 PM | | log flags: file_xfer, sched_ops, task 3/24/2013 4:29:38 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6 3/24/2013 4:29:38 PM | | Data directory: C:\ProgramData\BOINC 3/24/2013 4:29:38 PM | | Running under account gandalf 3/24/2013 4:29:38 PM | | Processor: 12 GenuineIntel Intel(R) Core(TM) i7-3930K CPU @ 3.20GHz [Family 6 Model 45 Stepping 7] 3/24/2013 4:29:38 PM | | 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 popcnt aes syscall nx lm vmx tm2 dca pbe 3/24/2013 4:29:38 PM | | OS: Microsoft Windows 7: Professional x64 Edition, Service Pack 1, (06.01.7601.00) 3/24/2013 4:29:38 PM | | Memory: 15.94 GB physical, 31.89 GB virtual 3/24/2013 4:29:38 PM | | Disk: 119.14 GB total, 69.78 GB free 3/24/2013 4:29:38 PM | | Local time is UTC -7 hours 3/24/2013 4:29:38 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 285 (driver version 314.07, CUDA version 5.0, compute capability 1.3, 1024MB, 949MB available, 1063 GFLOPS peak) 3/24/2013 4:29:38 PM | | CUDA: NVIDIA GPU 1: GeForce GTX 285 (driver version 314.07, CUDA version 5.0, compute capability 1.3, 2048MB, 1939MB available, 1063 GFLOPS peak) 3/24/2013 4:29:38 PM | | CUDA: NVIDIA GPU 2: GeForce GTX 650 (driver version 314.07, CUDA version 5.0, compute capability 3.0, 1024MB, 961MB available, 813 GFLOPS peak) 3/24/2013 4:29:38 PM | | CUDA: NVIDIA GPU 3: GeForce GTX 650 (driver version 314.07, CUDA version 5.0, compute capability 3.0, 1024MB, 961MB available, 813 GFLOPS peak) 3/24/2013 4:29:38 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 285 (driver version 314.07, device version OpenCL 1.0 CUDA, 1024MB, 949MB available, 1063 GFLOPS peak) 3/24/2013 4:29:38 PM | | OpenCL: NVIDIA GPU 1: GeForce GTX 285 (driver version 314.07, device version OpenCL 1.0 CUDA, 2048MB, 1939MB available, 1063 GFLOPS peak) 3/24/2013 4:29:38 PM | | OpenCL: NVIDIA GPU 2: GeForce GTX 650 (driver version 314.07, device version OpenCL 1.1 CUDA, 1024MB, 961MB available, 813 GFLOPS peak) 3/24/2013 4:29:38 PM | | OpenCL: NVIDIA GPU 3: GeForce GTX 650 (driver version 314.07, device version OpenCL 1.1 CUDA, 1024MB, 961MB available, 813 GFLOPS peak) 3/24/2013 4:29:38 PM | World Community Grid | Found app_config.xml 3/24/2013 4:29:38 PM | | Config: use all coprocessors 3/24/2013 4:29:38 PM | | Config: GUI RPCs allowed from: 3/24/2013 4:29:38 PM | | 192.168.1.100 3/24/2013 4:29:38 PM | | 192.168.1.101 3/24/2013 4:29:38 PM | | 192.168.1.103 3/24/2013 4:29:38 PM | World Community Grid | URL http://www.worldcommunitygrid.org/; Computer ID 2326477; resource share 100 3/24/2013 4:29:38 PM | World Community Grid | General prefs: from World Community Grid (last modified 27-Nov-2012 21:24:41) 3/24/2013 4:29:38 PM | World Community Grid | Host location: none 3/24/2013 4:29:38 PM | World Community Grid | General prefs: using your defaults 3/24/2013 4:29:38 PM | | Reading preferences override file 3/24/2013 4:29:38 PM | | Preferences: 3/24/2013 4:29:38 PM | | max memory usage when active: 16327.50MB 3/24/2013 4:29:38 PM | | max memory usage when idle: 16327.50MB 3/24/2013 4:29:38 PM | | max disk usage: 10.00GB 3/24/2013 4:29:38 PM | | (to change preferences, visit a project web site or select Preferences in the Manager) 3/24/2013 4:29:38 PM | | Not using a proxy 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500971201205101502_0 using hcc1 version 705 in slot 10 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500975201205101502_0 using hcc1 version 705 in slot 11 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500979201205101502_0 using hcc1 version 705 in slot 0 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500987201205101502_0 using hcc1 version 705 in slot 2 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500851201205101504_0 using hcc1 version 705 in slot 3 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500977201205101502_1 using hcc1 version 705 in slot 1 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500981201205101502_0 using hcc1 version 705 in slot 4 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500985201205101502_0 using hcc1 version 705 in slot 6 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500883201205101504_0 using hcc1 version 705 (nvidia_hcc1) in slot 7 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500837201205101504_0 using hcc1 version 705 (nvidia_hcc1) in slot 9 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500933201205101503_0 using hcc1 version 705 (nvidia_hcc1) in slot 5 3/24/2013 4:29:38 PM | World Community Grid | Restarting task X0900133500843201205101504_0 using hcc1 version 705 (nvidia_hcc1) in slot 12 3/24/2013 4:34:40 PM | | Re-reading cc_config.xml 3/24/2013 4:34:40 PM | | Not using a proxy 3/24/2013 4:34:40 PM | | Config: use all coprocessors 3/24/2013 4:34:40 PM | | Config: GUI RPCs allowed from: 3/24/2013 4:34:40 PM | | 192.168.1.100 3/24/2013 4:34:40 PM | | 192.168.1.101 3/24/2013 4:34:40 PM | | 192.168.1.103 3/24/2013 4:34:40 PM | | log flags: file_xfer, sched_ops, task 3/24/2013 4:34:40 PM | World Community Grid | Found app_config.xml 3/24/2013 4:34:50 PM | World Community Grid | General prefs: from World Community Grid (last modified 27-Nov-2012 21:24:41) 3/24/2013 4:34:50 PM | World Community Grid | Host location: none 3/24/2013 4:34:50 PM | World Community Grid | General prefs: using your defaults 3/24/2013 4:34:50 PM | | Reading preferences override file 3/24/2013 4:34:50 PM | | Preferences: 3/24/2013 4:34:50 PM | | max memory usage when active: 16327.50MB 3/24/2013 4:34:50 PM | | max memory usage when idle: 16327.50MB 3/24/2013 4:34:50 PM | | max disk usage: 10.00GB 3/24/2013 4:34:50 PM | | (to change preferences, visit a project web site or select Preferences in the Manager) 3/24/2013 4:35:08 PM | World Community Grid | Computation for task X0900133500837201205101504_0 finished 3/24/2013 4:35:08 PM | World Community Grid | Starting task X0900133500863201205101505_0 using hcc1 version 705 (nvidia_hcc1) in slot 8 3/24/2013 4:35:11 PM | World Community Grid | Started upload of X0900133500837201205101504_0_0 3/24/2013 4:35:11 PM | World Community Grid | Started upload of X0900133500837201205101504_0_1 3/24/2013 4:35:12 PM | World Community Grid | Computation for task X0900133500883201205101504_0 finished 3/24/2013 4:35:12 PM | World Community Grid | Starting task X0900133500951201205101503_0 using hcc1 version 705 (nvidia_hcc1) in slot 7 3/24/2013 4:35:13 PM | World Community Grid | Finished upload of X0900133500837201205101504_0_1 The image of the BOINC Manager:  Please explain this data. Why am I only seeing four GPU tasks running? Doesn't the log indicate 13 GPU tasks should be running? Respectfully...Confused. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
1) Already by setting the <cpu_usage> to 1.0 you're overriding the <max_concurrent> of 32. The limit as you put it is 12 for HCC. Of course, if there's not enough GPU jobs, then the rest will be filled out by CPU jobs of different or same kind.
|
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Gandalf,
We have issues with the maths here to sort out. 1st - close Boinc completely, use Ctrl Alt Delete, everything in your power to ensure it is not running a service in the background. I suspect that it WAS running when you set up your app_config file (maybe not - just my guess) Your cpu is 12 threaded - nice !! GPU's - 4 in total - hope you got adequate and reliable psu. Really .25 tasks per gpu might be a bit high we'll try 3 first in a little while shall we ?? So you had 4 gpus doing 4 tasks each with 1 cpu thread allocated to each task. Already you have 16 cpu threads asked, but cpu only has 12. try this app_config.... for 12 gpu tasks (3 each card) and 4 cpu tasks Don't need max concurrent line anymore either, without that restriction you may get extra cpu thread ?? bonus if you do, no skin if you don't <app_config> <app> <name>hcc1</name> <gpu_versions> <gpu_usage>.33</gpu_usage> <cpu_usage>.66</cpu_usage> </gpu_versions> </app> </app_config> Good luck |
||
|
|
branjo
Master Cruncher Slovakia Joined: Jun 29, 2012 Post Count: 1892 Status: Offline Project Badges: 
|
Nope, your log indicates 4 concurrent GPU WU's + 8 CPU tasks. I am guessing your app_config file ends with </app_config>.
----------------------------------------AFAIK, HCC1 GPGPU needs OpenCL v 1.1 or higher, your 285's have 1.0, this might be a problem. This does not makes a sense: <gpu_usage>0.25</gpu_usage> <cpu_usage>1.00</cpu_usage> It is asking your BOINC to run 16 GPU tasks on 16 CPU threads while you have only 12 of them. Try this app_config.xml: <app_config> <app> <name>hcc1</name> <gpu_versions> <gpu_usage>.25</gpu_usage> <cpu_usage>.75</cpu_usage> </gpu_versions> </app> </app_config> If you can use also your 285's, it has to result in 16 concurrent GPU tasks using all 12 CPU threads. If 285's will be excluded, you will be running 8 concurrent GPU tasks (on 650's) using 6 CPU threads + 6 CPU tasks. Cheers and keep us informed about your progress   Crunching@Home since January 13 2000. Shrubbing@Home since January 5 2006  [Edit 1 times, last edit by branjo at Mar 24, 2013 5:58:54 PM] |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
 underwater - As always, as soon as I post something things change. I changed the app_config to 1 + 0.50, and then remembered three things. Reboot, reboot and reboot. I'll remove that line per your suggestion. Like the tortoise, I may be slower than you guys, but with your help, I'll get there just in time for WCG to drop the HCC1 GPUs. P.S. - FYI - As in Signature shows, I'm running two 285 and two 650 GPUs in the same computer and it's working. Yea... Also, my PSU is 1000 watts. Ain't it a kicker. |
||
|
|
Former Member
Cruncher Joined: May 22, 2018 Post Count: 0 Status: Offline |
Why reboot when from 7.0.54 the app_config.xml change is simply applied via the Advanced menu Read Config file function?
|
||
|
|
|