Message boards :
Number crunching :
Lunatics - Should I be using a different MB GPU app?
Message board moderation
Author | Message |
---|---|
Price Family - Brighton Send message Joined: 4 Nov 00 Posts: 42 Credit: 48,039,168 RAC: 233 |
BOINC 6.10.60. GPU:AMD HD 6570 Is there a better MB GPU app for my setup? I've finally downloaded Lunatics and I'm not getting very good performance on the MB v7 GPU tasks. They're only about twice as fast as the MB CPU tasks. Virtually all the run time is in the GPU. MB usage: GPU Temperature c66.0, GPU Load 96+%. I've now started getting AP GPU tasks and the performance is very good, 9-10 times faster than an AP CPU task (similar to the stock AP GPU tasks that used to run). They spend c25% of the runtime in the CPU. AP usage: GPU Temperature c65.0, GPU Load 65-75% with short peaks of 100%. I compared the Task details from a current and stock GPU AP pending task and the GPU and CPU capability descriptions are pretty much the same. I then compared them with the Task details of an MB GPU pending task and the GPU and CPU capability descriptions are totally different. I think that the MB GPU apps I'm running are for a better GPU than the one I have. FYI the actual application running names are: setiathome_v7 7.03 (opencl_ati5_sah) astropulse_v6 6.04 (ati_opencl_100) The old stock task name: astropulse_v6 6.06 (ati_opencl_100) Hope all this makes sense. Rick |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
One thing you can do for MB tasks is go to your projects folder. Open "mb_cmdline_win_x86_SSE_OpenCL_ATi_HD5.txt" file with Notepad and insert -sbs 256. Save the file. On next WU start these parameters are in use. It increases single buffer size and should speed it a little bit up. You dont need to stop Boinc. With each crime and every kindness we birth our future. |
MarkJ Send message Joined: 17 Feb 08 Posts: 1139 Credit: 80,854,192 RAC: 5 |
One other thing you might find is leaving a CPU core free to feed the GPU. OpenCL apps seem to benefit greatly from this. Just set the "use % processors" to some value less than 100 (eg 87.5% for your i7-2600 machine). Also check your driver version. It says unknown which might just be BOINC doesn't know, but maybe the driver isn't the best one for your card. BOINC blog |
Price Family - Brighton Send message Joined: 4 Nov 00 Posts: 42 Credit: 48,039,168 RAC: 233 |
Thanks. I've added that to the text file (it's the only thing in it) so we'll see how it goes. |
Price Family - Brighton Send message Joined: 4 Nov 00 Posts: 42 Credit: 48,039,168 RAC: 233 |
Sorry, I meant to mention these points. I have got a free CPU and the latest driver 13.4 (as of a few weeks ago). |
Mike Send message Joined: 17 Feb 01 Posts: 34255 Credit: 79,922,639 RAC: 80 |
Sorry, I meant to mention these points. I have got a free CPU and the latest driver 13.4 (as of a few weeks ago). I`ve noticed that already. Lets see how the parameters are working. I´ll check it later. With each crime and every kindness we birth our future. |
Price Family - Brighton Send message Joined: 4 Nov 00 Posts: 42 Credit: 48,039,168 RAC: 233 |
I believe I've made a mistake in the following statement. I think that the MB GPU apps I'm running are for a better GPU than the one I have. There seem to be two naming conventions For ATI (and other?) GPUs. They are either opencl_[GPU type] or [GPU type]_opencl as in my two apps below. setiathome_v7 7.03 (opencl_ati5_sah) astropulse_v6 6.04 (ati_opencl_100) I assumed the first convention was better but I've just noticed the thread 'Stuck AstroPulse v6 v6.06 (opencl_ati_100) task' which says that the GPU that this task runs on is fairly minimal. So I now think the MB GPU app is for a more basic GPU. Hopefully there is a setiathome_v7 app available that uses the second naming convention and will perform well with my GPU. Rick |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
There seem to be two naming conventions For ATI (and other?) GPUs. They are either opencl_[GPU type] or [GPU type]_opencl as in my two apps below. The naming convention is only there because of a work around that Eric did so Boinc 6 clients can receive OpenCL work, Boinc 6 clients don't do OpenCL detection, so Eric used the assumption that clients/GPUs with a certain driver version would have OpenCL support, So basically the planclass of opencl_ goes to a Boinc 7 client, and anything with ati_ or cuda_ go to a Boinc 6 client, the apps supplied in each case are the same. For the v7 AMD apps, Eric used the opencl_ati_ and opencl_ati5_ planclasses (further devided into pre and post Cat 13.1 planclasses), for Stock the opencl_ati_ is the HD5 app, while opencl_ati5_ is the non-HD5 app, while for Annonymous platform use, we at Lunatics put the planclasses the other way round, so when running Anonymous Platform, the HD5 app will show up as using the opencl_ati5_ planclasses, and the non-HD5 app as opencl_ati_ Claggy |
Price Family - Brighton Send message Joined: 4 Nov 00 Posts: 42 Credit: 48,039,168 RAC: 233 |
I see. So what I'm running is what there is. I can live with that. I must stop assuming that what I'm assuming is correct. Is there any configuration I can set to give AP GPU tasks preference when requesting new tasks? Obviously I'd rather run these when available. I've had so few that the estimated run time is still over 100 hours. Rick |
Wiggo Send message Joined: 24 Jan 00 Posts: 34744 Credit: 261,360,520 RAC: 489 |
Under your SETI Computing Preferences make these changes. SETI@home Enhanced: no SETI@home v7: no AstroPulse v6: yes .., accept work from other applications?: yes Cheers. |
©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.