Message boards :
Number crunching :
Run Multiple WU's on Your GPU
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 . . . 11 · Next
Author | Message |
---|---|
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
I have been trying to run the app_config file shown above and whenever I do the Boinc event log shows an error message saying "App file not found". Has anyone else had this problem? Something is funky then, I would suggest going thru the steps again from creating the file in Notepad, to saving the file in the right place, to then stopping and restarting Boinc. |
Send message Joined: 4 Mar 10 Posts: 6 Credit: 1,057,111 RAC: 0 |
thanks this is working nicely Crunching 8 WUs at once on my GTX690 card |
Send message Joined: 22 Mar 10 Posts: 5 Credit: 32,810,611 RAC: 0 |
I have been trying to run the app_config file shown above and whenever I do the Boinc event log shows an error message saying "App file not found". Has anyone else had this problem? Yes, and haven't found the answer yet. Did try the ideas mentioned above. edit: Mine wanted <use_all_gpus>1</use_all_gpus> in the cc_config, apparently. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
I have been trying to run the app_config file shown above and whenever I do the Boinc event log shows an error message saying "App file not found". Has anyone else had this problem? There are other things that some require that others don't too, some require a dummy plug for the 2nd gpu: http://www.overclock.net/t/384733/the-30-second-dummy-plug While others require the use of the <use_all_gpus> line, some require BOTH! Unfortunately it is NOT a one size fits all type thing. Even running two units at once doesn't ALWAYS work, I can easily make it run on my two Nvidia cards but doing the exact same thing on my AMD cards means it works for one unit then stops, and it will only run a single unit from them on! |
Send message Joined: 15 Oct 07 Posts: 14 Credit: 150,634,885 RAC: 0 |
I am using the following app_INFO.xml file, after much fiddling and searching, with no real luck, one of my team mates gave me a copy of his which is now humming along quite nicely in two machines ( though a little slow on a 5850..about to receive the 6950 which came out of the main rig)... I have 6 Milkyway WU's running on the 7950 in one of my i7's (my main rig , which also has an nVidia 660Ti, running 6 POEM WU's concurrently), with the MW Wu's taking about 4 minutes or so. <app> <name>milkyway</name> </app> <file_info> <name>milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe</name> <executable/> </file_info> <app_version> <app_name>milkyway</app_name> <version_num>102</version_num> <flops>1.0e11</flops> <avg_ncpus>0.05</avg_ncpus> <max_ncpus>1</max_ncpus> <plan_class>ati14ati</plan_class> <coproc> <type>ATI</type> <count>0.166</count> </coproc> <cmdline></cmdline> <file_ref> <file_name>milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe</file_name> <main_program/> </file_ref> </app_version> </app_info> what I WOULD like to know, if someone has been testing etc, is this: Does the processing time improve at all with increasing the clock speed or memory speed on the GPU ? I am sure I remember something a while back which ( although different projects ARE benefited) said that GPU's could even be under-clocked and voltages reduced a bit (to reduce heat) for MW on a GPU... Thanks in advance, Veebee |
Send message Joined: 18 Jul 09 Posts: 300 Credit: 303,581,398 RAC: 1,022 |
GPU clock speed has a direct bearing on completion times. But MW uses little memory so having excess memory speed only produces heat. I keep my GPU clock at 900MHz and my memory clock at 300MHz. Collatz, for example, requires memory speed. If you run it, turn the memory all the way up and ventilate the room. |
Send message Joined: 18 Jul 09 Posts: 300 Credit: 303,581,398 RAC: 1,022 |
I have been trying to run the app_config file shown above and whenever I do the Boinc event log shows an error message saying "App file not found". Has anyone else had this problem? I'm trying once again to run with the app_config file vs. an app_info file. I am using the app_config file as written earlier in this thread: <app_config> <app> <name>milkyway</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.05</cpu_usage> </gpu_versions> </app> </app_config> Here is the app_config file as it appears in my projects folder: <?xml version="1.0"?> -<app_config> -<app> <name>milkyway</name> -<gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.05</cpu_usage> </gpu_versions> </app> </app_config> Here is my Boinc event log since I started up: 4/26/2013 3:22:29 PM | | No config file found - using defaults 4/26/2013 3:22:29 PM | | Starting BOINC client version 7.0.64 for windows_x86_64 4/26/2013 3:22:29 PM | | log flags: file_xfer, sched_ops, task 4/26/2013 3:22:29 PM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6 4/26/2013 3:22:29 PM | | Data directory: C:\ProgramData\BOINC 4/26/2013 3:22:29 PM | | Processor: 3 AuthenticAMD AMD Phenom(tm) II X3 720 Processor [Family 16 Model 4 Stepping 2] 4/26/2013 3:22:29 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 htt pni cx16 popcnt syscall nx lm svm sse4a osvw ibs skinit wdt page1gb rdtscp 3dnowext 3dnow 4/26/2013 3:22:29 PM | | OS: Microsoft Windows 7: Home Premium x64 Edition, Service Pack 1, (06.01.7601.00) 4/26/2013 3:22:29 PM | | Memory: 4.00 GB physical, 8.00 GB virtual 4/26/2013 3:22:29 PM | | Disk: 465.66 GB total, 419.39 GB free 4/26/2013 3:22:29 PM | | Local time is UTC -5 hours 4/26/2013 3:22:29 PM | | CAL: ATI GPU 0: ATI Radeon HD 5800 series (Cypress) (CAL version 1.4.1741, 1024MB, 991MB available, 5760 GFLOPS peak) 4/26/2013 3:22:29 PM | | OpenCL: AMD/ATI GPU 0: ATI Radeon HD 5800 series (Cypress) (driver version CAL 1.4.1741 (VM), device version OpenCL 1.2 AMD-APP (938.2), 1024MB, 991MB available, 5760 GFLOPS peak) 4/26/2013 3:22:29 PM | Milkyway@Home | [error] State file error: missing application file milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe 4/26/2013 3:22:29 PM | Milkyway@Home | Found app_config.xml 4/26/2013 3:22:29 PM | Milkyway@Home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 252041; resource share 900 4/26/2013 3:22:29 PM | | General prefs: from http://setiathome.berkeley.edu/ (last modified 18-Mar-2012 07:20:13) 4/26/2013 3:22:29 PM | | Computer location: home 4/26/2013 3:22:29 PM | | General prefs: no separate prefs for home; using your defaults 4/26/2013 3:22:29 PM | | Reading preferences override file 4/26/2013 3:22:29 PM | | Preferences: 4/26/2013 3:22:29 PM | | max memory usage when active: 3071.48MB 4/26/2013 3:22:29 PM | | max memory usage when idle: 4054.35MB 4/26/2013 3:22:29 PM | | max disk usage: 100.00GB 4/26/2013 3:22:29 PM | | (to change preferences, visit a project web site or select Preferences in the Manager) 4/26/2013 3:22:29 PM | | Not using a proxy 4/26/2013 3:22:33 PM | Milkyway@Home | work fetch resumed by user 4/26/2013 3:22:36 PM | Milkyway@Home | update requested by user 4/26/2013 3:22:40 PM | Milkyway@Home | Sending scheduler request: Requested by user. 4/26/2013 3:22:40 PM | Milkyway@Home | Requesting new tasks for ATI 4/26/2013 3:22:42 PM | Milkyway@Home | Scheduler request completed: got 16 new tasks 4/26/2013 3:22:44 PM | Milkyway@Home | File milkyway_separation_1.02_windows_x86_64__opencl_amd_ati.exe exists already, skipping download 4/26/2013 3:22:44 PM | Milkyway@Home | Starting task de_p_82_1s_dr8_4_2_1366886102_957162_0 using milkyway version 102 (opencl_amd_ati) in slot 2 4/26/2013 3:23:35 PM | Milkyway@Home | Computation for task de_p_82_1s_dr8_4_2_1366886102_957162_0 finished 4/26/2013 3:23:35 PM | Milkyway@Home | Starting task de_p_82_1s_dr8_4_2_1366886102_957167_0 using milkyway version 102 (opencl_amd_ati) in slot 2 4/26/2013 3:23:47 PM | Milkyway@Home | Sending scheduler request: To fetch work. 4/26/2013 3:23:47 PM | Milkyway@Home | Reporting 1 completed tasks 4/26/2013 3:23:47 PM | Milkyway@Home | Requesting new tasks for ATI 4/26/2013 3:23:49 PM | Milkyway@Home | Scheduler request completed: got 16 new tasks 4/26/2013 3:23:51 PM | Milkyway@Home | Starting task de_p_82_1s_dr8_4_2_1366886102_957169_0 using milkyway version 102 (opencl_amd_ati) in slot 4 My system is happily crunching one WU at a time, not two. Is it because of the extra line about the xml version in the config file? I also noticed the error message in the Boinc event log. Anyone have any ideas how I can get this to work properly? Edit: Hmmm...I go and check my computer and now it's running two WUs! I don't know why it didn't from the beginning but I'm not going to argue. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
The first " no config file found" is because you don't have that file, such as one saying use all gpus or something in the Boinc directory. Did you stop and restart Boinc after you edited the app_config file? Also what do you mean when you say "here is my app_config file" and then "here it is in my projects folder"? They should be IDENTICAL not different as you show. Are you using NOTEPAD in Windows and saving the file as a text type file with the extension .xml? This is the first one: <app_config> <app> <name>milkyway</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.05</cpu_usage> </gpu_versions> </app> </app_config> and this is the one in your projects directory: <?xml version="1.0"?> -<app_config> -<app> <name>milkyway</name> -<gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.05</cpu_usage> </gpu_versions> </app> </app_config> I see dashes, an ?xml line, things that should not be there. It looks like you are using a word processing program and not NOTEPAD. Word processing programs add hidden stuff to the file for possible editing tweaks, Boinc does not recognize those and says what the heck, and then won't process the file. As for the error line it could be that it didn't find the file and then downloaded it on its own, that is how it SHOULD work. As you said though since it is now working as you want it to I might just leave it alone and let it does its thing. |
Send message Joined: 18 Jul 09 Posts: 300 Credit: 303,581,398 RAC: 1,022 |
I am using Notepad. When I edit the file, it opens in notepad (W7-64) and I get what should appear, no extra spaces, characters, etc. When I simply open it, it opens in FF and that is where the extra stuff comes from. Am I the only one to see these differences? I'm hoping they are normal. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
I don't know as I never open mine in FF, I have the default set to Notepad. I guess as long as you don't edit or save it after viewing it in FF it should be okay, after all you are just looking at it. |
Send message Joined: 4 Nov 12 Posts: 96 Credit: 251,528,484 RAC: 0 |
<app_config> Thanks for this. I'm now running 4 GPU tasks simultaneously where I was only doing one before on my SLIed cards two days ago. I had to update BOINC today to get it to work though. Yesterday I got 2 task to run, one on each GPU. Not sure how I'm going to get 8 GPU tasks to run effectively tomorrow... ;) This is putting my 680 cards to good use. They're at about 85% to 90% average GPU load per card now (they're 4GB cards, so memory isn't really an issue), clocked at 1175MHz each, 65°C, which is a low enough temp for them, but I'm not wanting to push them much higher for nearly 24/7 crunching. Anyway, thanks again guys. Getting a lot more crunching use out of my computer now, and hopefully back on track to reach my 10 million credits date goal. |
Send message Joined: 4 Mar 12 Posts: 45 Credit: 460,275,437 RAC: 2,158 |
I am running a vishera 8 core amd fx processor with a radeon xfx 6950 graphics card. If I use all cores for cpu tasks in boic with mw@home it takes forever to complete 1 gpu mw@home task. It also barely utilizes the gpu when looked at in gpu-z. If I free up 1 cpu core, by setting the cpu's used to 99% I can tear through mw@home gpu work units like a maniac. Usually taking anywhere from 36-47 sec per work unit. When I check gpu usage with gpu-z I'm at 100%. I acutally had to increase my fan speed on the gpu because these work units were heating it up something nice :) lol I can't see why someone would want to run 8 simultaneous gpu based wu's at once... I know my card is no where near a super duper new card, but it's nothing too crappy either. Is everyone leaving one cpu core free for their gpu work to utilize the cpu when needed? Just curious. I'd love to update to a dual crossfire config with 2 7900 series cards, but I think I'm doing ok at the moment. :) - Penguin |
Send message Joined: 3 Mar 13 Posts: 84 Credit: 779,527,712 RAC: 0 |
Since i got serious about GPU crunching i always keep one core free to run the cup part of boinc apps, seti, einstein, milky, (colatz is `just for fun`) and they all work better for it. |
Send message Joined: 24 Feb 09 Posts: 620 Credit: 100,587,625 RAC: 0 |
Just seen this - and noticed no reply to the specific .... so ... The dash(s) are a graphics representation of additional characters that word processors place into files for their own use - they are varying control characters. When such a file containing those characters is viewed most of them cannot be seen by the human eye - they are hidden characters. However, that doesn't mean they are not there, and usually the "-" (no quotes) is the indicator that you are using a "dirty" editor that uses hidden control characters (most editors do, don't misplace the use of the word "dirty"). When BOINC reads the file, it also reads those hidden characters, and basically says "wtf is that" and gives up. To avoid that always - without fail - use Notepad to make config (etc) files. Notepad is a "clean" tool that does not put in hidden control characters, and therefore when BOINC reads it, its a happy bunny. If the file is falling over, make sure its formed inside Notepad first, no other Tool before ripping apart anything else on a fault-finding odyssey :) Its use will save 2 to 3 secs on the load/unload of WUs between gpu/cpu if configured for multiple WUs per GPU (not all of them, but most of them). If your normal one-per-gpu crunch time is (say) 50 secs+ per WU, then for sure its a difference, but not earth shattering. If you have a fast GPU pumping them at around 20 to 30secs a WU, then its vital that an app_info file is used if your aim is maximising credits; as the 2-3 secs load unload time represents 15%+ or so of total crunch time; not the 2% or so if your base line time was 50secs or so actual crunch time. So .... if your unaided crunching time was 50-60 secs, then for sure, you will gain a little, but hardly anything to worry about. Run a 6XXX or 7XXX high end variant - then its essential one is used to cut down "dead time" if you are to maximise credits. There is always a downside to things (!). This one is that by using the file as discussed, all the GPU(s) capacity is utilised to the full 100% capacity, there is no "wasted" capacity compared to only running at circa 90%+ GPU utilisation. Therefore, if you try to do something else as well ..... the PC is not going to be a happy bunny, and response on the extra task will be horrific as all GPU capacity is pointed at BOINC via the app_info. App_Info's are good for "fire and forget" crunchers dedicated to crunching, but for machines that also are used for browsing et al at the same time, its trial and error; because its pointless, as the Browsing sessions suck up any spare capacity (and more). So its pointless frankly doing this if you multi-task doing others things as well on the same machine. Pays your money takes your choice as they say...... On a fast machine, this works well on aka high end variants of 6XX or 7XX. High end 5XX is trial and error experiment to see how yours goes. On a slow machine (low end 5XX, low end 6XX and low end 7XX that is used to multi-task on browsing non-BOINC - or using a low end variant of any GPU Class or type on BOINC dedicated machines - don't bother, it will slow it down horribly, a real pain. |
Send message Joined: 8 May 09 Posts: 3339 Credit: 524,010,781 RAC: 0 |
I am running a vishera 8 core amd fx processor with a radeon xfx 6950 graphics card. It depends on the project and the brand of card, on Collatz you can use all the cpu's for crunching and the gpu will not be affected, but at other projects leaving a cpu core open is very beneficial. Also at some projects you can use all your cpu cores if you have an AMD gpu but if you are using an Nvidia gpu it is best ot leave one open. What I do is go into the Boinc Manager and look at the numbers in the status column. It will tell how much cpu it is using, then I adjust the cpu cores on that machines accordingly. At Collatz, for example, the status column shows a cpu usage of 0.02% whether a cpu core is left free or not. But at DistRTgen it shows 0.866% of a cpu core being used by the gpu, and I leave one open. |
Send message Joined: 4 Mar 12 Posts: 45 Credit: 460,275,437 RAC: 2,158 |
Since i got serious about GPU crunching i always keep one core free to run the cup part of boinc apps, With your RAC at the level I can tell you know what you're doing! ;) Happy crunching. |
Send message Joined: 12 Jul 11 Posts: 2 Credit: 4,832,979 RAC: 0 |
Thanks for this :) , Was using a app file , but it stopped working , dropped this config file in wola :D Running 2 HD 7950 , 2 WU per gpu , Its taking at the moment around 1 minuet for every 4 WU to complete , GPUs 100% @ 75 C I run a Dual core ivy Pentium , but only do GPU crunching so the CPU is never been used by anything other than the GPU WU when needed . |
Send message Joined: 23 Nov 09 Posts: 29 Credit: 17,119,258 RAC: 0 |
You should clock that GPU to 1200MHz core frequency and you'll be completing a work unit every 9 seconds (4 every 36 seconds). I just earned 350,000+ yesterday with just one 7950. |
Send message Joined: 12 Jul 11 Posts: 2 Credit: 4,832,979 RAC: 0 |
Im running at 1100 Mhz Currently , takes 1min 20 to do 6 WU ( 3 on each card ) |
Send message Joined: 22 Mar 10 Posts: 5 Credit: 32,810,611 RAC: 0 |
Im running at 1100 Mhz Currently , takes 1min 20 to do 6 WU ( 3 on each card ) My results are similar to this, running a 7870xt (7950 rebranded by the marketing department). I can run one in 42 seconds, or two in 83 seconds, a speed increase of less than 2%. Adding a third gives no increase at all. At first, I thought this might be a method that only works on Nvidias, since this has been how it's worked for me on both the old and new clients, with a couple of different mobos and a few different AMD CPUs. (Current CPU is an X6 1045t.) I do leave spare cores. Does anyone have any suggestions as to why it seems to work amazingly for some people, but barely, or not at all, for others with similar hardware? |
©2024 Astroinformatics Group