Message boards :
Number crunching :
Lunatics what to chose?
Message board moderation
Author | Message |
---|---|
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Hey there. Im looking to optimise my littel PC for Seti, its a I7-4790K with a GTX-1080 and 32GB. When installing Lunatics what CPU type should be the best, SSE4.1 or AVX? And what would be the best to use with a 1080? Im told that Ive need to make a app_config.xml and have boinc run two tasks at the same time on the GPU. And a mb*SoG.txt with some settings (a very long line). So my guess would be that the 1080 should run SoG, but Im not sure about the CPU. And if Ive chose wrong, who do one change from AVX to SSE4.1 or vise-versa? Uninstall and reinstall or? |
Wiggo Send message Joined: 24 Jan 00 Posts: 34871 Credit: 261,360,520 RAC: 489 |
AVX for the CPU and SoG for the 1080. ;-) Cheers. |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Thanks - so now I've just have to find out how to change from SSE4.1 to AVX ! |
Wiggo Send message Joined: 24 Jan 00 Posts: 34871 Credit: 261,360,520 RAC: 489 |
Just rerun the installer again to correct that. ;-) Someone else will have to help you with app_config.xml and cmdline settings for the SoG app though. Cheers. |
Brent Norman Send message Joined: 1 Dec 99 Posts: 2786 Credit: 685,657,289 RAC: 835 |
If I remember right, there is no SSE4 app, it just installs SSSE3. So I'd use AVX. EDIT: IN the seti@home project directory ... For app_config.xml (a plain TXT file - use Notepad) This should work well: <app_config> <app> <name>setiathome_v8</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>1.0</cpu_usage> </gpu_versions> </app> <app> <name>astropulse_v7</name> <gpu_versions> <gpu_usage>0.5</gpu_usage> <cpu_usage>0.5</cpu_usage> </gpu_versions> </app> </app_config> You don't have to restart BOINC for this, just read the config files. |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Done the app_config.xml, with more or less the same numbers. and done mb*SoG.txt with this: -sbs 256 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp -cpu_lock -high_perf Not impressed, ~430 gflops on a 1080 - my 780 get more that that...- so have to figure out what all that means and tune it to my GTX... Michael. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Done the app_config.xml, with more or less the same numbers. Change to -sbs 1024 -period_iterations_num 10 -spike_fft_thresh 4096 -tune 1 64 1 4 -oclfft_tune_gr 256 -oclfft_tune_lr 16 -oclfft_tune_wg 256 -oclfft_tune_ls 512 -oclfft_tune_bn 64 -oclfft_tune_cw 64 -hp -cpu_lock -high_perf With each crime and every kindness we birth our future. |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
One have told me to drop the cpu_lock, but will wait and see where it ends up. Day 5 on that PC on my own account and its still climping in RAC. What would you think is a good RAC on a PC like this? EVGA GTX 1080 Hydrocopper FTW. I7-4790K @ 4.6gHz no HT Running 4 tasks on the GTX, and 50/100% the CPU (half the cores at full load). Michael. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
It wasn't me that told you to drop it but I've never had any good experience with using it. Yes it improves the time on 1 work unit while sacrificing the crunching time of the others on the same GPU |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
It wasn't me that told you to drop it but I've never had any good experience with using it. Yes it improves the time on 1 work unit while sacrificing the crunching time of the others on the same GPU That`s correct Zalster. With each crime and every kindness we birth our future. |
Keith J. LaGue Send message Joined: 17 May 99 Posts: 59 Credit: 40,441,387 RAC: 0 |
It wasn't me that told you to drop it but I've never had any good experience with using it. Yes it improves the time on 1 work unit while sacrificing the crunching time of the others on the same GPU Oops...I was wondering why the discrepancy in my run times. Thanks for posting that tip. I wonder if using -cpu_lock_fixed_cpu N would make a noticeable difference? |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
It wasn't me that told you to drop it but I've never had any good experience with using it. Yes it improves the time on 1 work unit while sacrificing the crunching time of the others on the same GPU Not really. Once it was an option because -cpu_lock didn`t work as intended. Now it does, so no difference. Only if you run just one instance on GPU you can bind it to a specific cpu core. With each crime and every kindness we birth our future. |
Filipe Send message Joined: 12 Aug 00 Posts: 218 Credit: 21,281,677 RAC: 20 |
One have told me to drop the cpu_lock, but will wait and see where it ends up. With this PC you will get a RAC close to 40.000 running 24/7 |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
So drop GPU_lock?? |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13746 Credit: 208,696,464 RAC: 304 |
So drop GPU_lock?? Only if you run just one instance on GPU you can bind it to a specific cpu core. Since you're running multiple WUs on the GPU, it may be worthwhile. Grant Darwin NT |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
So drop GPU_lock?? It depends on how much CPU cores are available/free. In most cases -cpu_lock helps. With each crime and every kindness we birth our future. |
Zalster Send message Joined: 27 May 99 Posts: 5517 Credit: 528,817,460 RAC: 242 |
So drop GPU_lock?? Try it with and without. Monitor the times as they crunch and see what the times do. Make sure the work units are similiar...ie the names of the work units, especially the non BLC ones. I've found some 08 work units are faster than others, so try to get a sample of different dates and write them down, then compare the times of work units from the same tape without the cpu lock and see what the times look like. If there is a significant amount of difference in the time to complete then don't use it. Most of this is trial and error, you have to test it out on your system and see what works best |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
So first thing is to wait, and let the RAC get stable :) Next thing is to fiddle with CPU_Lock, and find out what the avg WU-time is then - or do my RAC go up, or down ? But dam this thing is a long time to get the RAC stabile on... Michael. |
Grant (SSSF) Send message Joined: 19 Aug 99 Posts: 13746 Credit: 208,696,464 RAC: 304 |
So first thing is to wait, and let the RAC get stable :) Nope. It will take (depending on outages etc) a couple of months for RAC to get to it's usual levels. As Zalster said, check the run times of similar WUs against each other with the different settings to see what does (or doesn't) help. Grant Darwin NT |
Stauning Send message Joined: 10 Nov 00 Posts: 25 Credit: 67,819,196 RAC: 749 |
Well with 1-2 days of seti-server downtime every week, its not a easy task to messure it by RAC your right. Oki Im on it now, Ill hope. Thanks for all the help. Michael. |
©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.