Message boards :
Number crunching :
app_info tuning
Message board moderation
Author | Message |
---|---|
doug Send message Joined: 10 Jul 09 Posts: 202 Credit: 10,828,067 RAC: 0 |
I'm using the Lunatics apps and I started messing with the app_info.xml file after reading a post on this board and monitoring GPU usage using GPU-Z. At first I only modified the MB task to allow it to run 2 tasks at once since it seemed I was only using ~30% of the GPU most of the time. That worked great. Now I crunch 2 MB ATI tasks at a time and use ~70% of the GPU. I just modified my app_info.xml to allow 2 AP WU's to run at a time after monitoring with GPU-Z. I'm still playing with it to see how that goes. My question is, what other things can I do to tune tasks with? I searched the Lunatics site and didn't really find anything. I also haven't really found anything on this board. I see that there are command line arguments that you can give to the task. I think my ATI card is maxed out at 2 concurrent tasks, so now I'm looking for sage advice on what else I can do, in essence tuning guidelines. Here's a snippet of my app_info file and you can see we also have things like unroll, ffa_block and ffa_block_fetch that can be modified. Suggestions? Doug <app_version> <app_name>astropulse_v505</app_name> <version_num>506</version_num> <avg_ncpus>0.04</avg_ncpus> <max_ncpus>0.2</max_ncpus> <platform>windows_intelx86</platform> <plan_class>ati13ati</plan_class> <cmdline>-instances_per_device 2 -unroll 6 -ffa_block 4096 -ffa_block_fetch 2048</cmdline> <coproc> <type>ATI</type> <count>0.5</count> </coproc> <file_ref> <file_name>ap_5.06_win_x86_SSE2_OpenCL_ATI_r521.exe</file_name> <main_program/> </file_ref> <file_ref> <file_name>AstroPulse_Kernels_r521.cl</file_name> <copy_file/> </file_ref> </app_version> |
kittyman Send message Joined: 9 Jul 00 Posts: 51469 Credit: 1,018,363,574 RAC: 1,004 |
I'm not running any ATI cards, but you could do a forum search for 'unroll' and check some other posts for some possible hints. "Freedom is just Chaos, with better lighting." Alan Dean Foster |
doug Send message Joined: 10 Jul 09 Posts: 202 Credit: 10,828,067 RAC: 0 |
I will do that. I'm also curious about setting the <count>x</count> to .75 since AP, true to its name, seems to pulse the GPU. I guess I'm limiting those pulses by using x=.5. Make sense? Doug |
Fred J. Verster Send message Joined: 21 Apr 04 Posts: 3252 Credit: 31,903,643 RAC: 0 |
[ADDED] These 2 things, have nothing todo, with each other and is the way the app. uses the GPU(s), except the loading and unloading by CPU, the GPU load is variable. [color=green] By Fine-Tuning these UNROLL(2 -18)? and ffa_block & fetch: 4096/10240 & 2048/5120, as I tried, also.[color] By the way, I can't use GPUz anymore, cause it tells rubbish, 2 GPUs, using (CROSS-FIRE) when displaying something, but act as 2x 1GPU when 'crunching'. But CPUz (0.53/54/55) all give 100% load for the 1st GPU (Load or Not!) and the 2nd gives a more realistic reading and shows the actual load, atleast I hope! (Maybe someone noticed this behavior on DUAL-GPU-CARDs)? With <count>1</count>, you instuct to use the whole GPU, <count>0.5</count>, you 'instruct' the GPU the use half, so 2 can be run.?! Do I say this right? With UNROLL, ffa_block & ffa_block_fetch, you can overule defaults and try what's best for your GPU(s). Take a look at results and the one from your wingmen/womem, or other hosts, running ATI and OpenCL. |
Claggy Send message Joined: 5 Jul 99 Posts: 4654 Credit: 47,537,079 RAC: 4 |
First of all, try putting the unrolled value to 10, that was what I was running on my GTX460 and HD5770 (before the HD failed). Claggy |
doug Send message Joined: 10 Jul 09 Posts: 202 Credit: 10,828,067 RAC: 0 |
I'll try setting unroll to 10 on my next test. What does unroll mean? I can't find an explanation of it anywhere. Doug |
HAL9000 Send message Joined: 11 Sep 99 Posts: 6534 Credit: 196,805,888 RAC: 57 |
I'll try setting unroll to 10 on my next test. What does unroll mean? I can't find an explanation of it anywhere. You might want to have a read in the Let's talk about ATI AP -unroll, -ffa_block, and -ffa_block_fetch thread. It explains what the values mean & has the data from several test attempts from various people. SETI@home classic workunits: 93,865 CPU time: 863,447 hours Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[ |
doug Send message Joined: 10 Jul 09 Posts: 202 Credit: 10,828,067 RAC: 0 |
I completely missed that message thread. I guess others have been down this road before. Thanks. Doug |
©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.