New to AP, where to start?

Message boards : Number crunching : New to AP, where to start?
Message board moderation

To post messages, you must log in.

AuthorMessage
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1411355 - Posted: 4 Sep 2013, 13:16:29 UTC

Going on 14 years of Classic and MB crunching I would like to wet my feet in the AP realm but I'm not sure how to proceed. Yes, I know I check the box in the Lunatics .41 installer (GREAT job BTW guys) but Where/how is it easiest to control how many to run on GPU only? app_info or app_config or what combo of both?

I want to try on a dedicated cruncher that I just put together from spare parts I had laying around. Then, after I've got the hang of how it behaves, get it going on my other machines.

Specs: Athlon2 x2 215, GTS450 1GB, 4GB RAM

I currently run 2 MB at a time, I don't know if the 450 has the ummph to do 2 APs at a time. Maybe I should try 1 AP (when available) and 1 MB but still do 2 MB when there are no APs. Where to start?

Thanks in advance.
</Tazz>
ID: 1411355 · Report as offensive
Profile Fred E.
Volunteer tester

Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,140,697
RAC: 0
United States
Message 1411362 - Posted: 4 Sep 2013, 13:40:14 UTC

Suggest you use app_config.xml because it is easier to modify and re-read without restarting BOINC. You can also set a max concurrent of 1 for AP in app_config if you want to limit the number of AP tasks running. With this design, the max concurrent also impacts the cpu tasks - I wish the design was different.
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.
ID: 1411362 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1411387 - Posted: 4 Sep 2013, 14:38:49 UTC - in response to Message 1411362.  

With this design, the max concurrent also impacts the cpu tasks - I wish the design was different.


Here's what I've got:
Lunatics app_info for MB only. <count>.5</count> is the only change made.

This app_config:
<app_config>
<app>
<name>astropulse_v6</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>.51</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v7</name>
<gpu_versions>
<gpu_usage>.49</gpu_usage>
<cpu_usage>.06</cpu_usage>
</gpu_versions>
</app>
</app_config>

After I check AP in the .41 installer do I have to change anything in the app_info? Just the <count> back to .5 for MB?

BTW I already have 'Use at most %50.00 processors' leaving 1 free core. Does this look like it might work, I don't want to trash a bunch of work (again) not knowing what I'm doing.
</Tazz>
ID: 1411387 · Report as offensive
Profile Fred E.
Volunteer tester

Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,140,697
RAC: 0
United States
Message 1411393 - Posted: 4 Sep 2013, 15:01:46 UTC

The app_config.xml file looks okay. You'll get a message when it is read via the menu and at startup - could put it in now to see if there's an error message.

I usually keep the counts in app_info.xml consistent although others have said that's not necessary. Unless you have made other edits (flops, etc) you should not have to do anything else. I loke to preview the app_info.xml before startup.

Look at the readme file in the docs directory - may not be there now if you didn't elect AP / Nvidia during install, but it will be there after a new install. File name is ReadMe_AstroPulse_OpenCL_NV.txt. It has info on command line parameters to improve performance. They cut my times in half. Perhaps someone can suggest starting values for your 450.

Also, you may find that AP doesn't load your system as much as you might expect. I run 3 of these OpenCL tasks at a time on my 670 without much strain, but can only run 2 CUDA 5.0 Multibeam at a time without getting into problems. Astropulse does put more load on my cpu and I have to watch temps (which is why I use the max_concurrent switch).
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.
ID: 1411393 · Report as offensive
Profile ivan
Volunteer tester
Avatar

Send message
Joined: 5 Mar 01
Posts: 783
Credit: 348,560,338
RAC: 223
United Kingdom
Message 1411955 - Posted: 5 Sep 2013, 22:14:14 UTC - in response to Message 1411393.  

Also, you may find that AP doesn't load your system as much as you might expect. I run 3 of these OpenCL tasks at a time on my 670 without much strain, but can only run 2 CUDA 5.0 Multibeam at a time without getting into problems.

Interesting, Fred. I put a GTX 660 Ti into a Linux box using a self-compiled CUDA 5.5 app a week ago and found that it runs 100-110 credit MB WUs in 40 mins or less with one, two, three, or four concurrent processes (it had a GTS 8800 512 in it before that). Since the 660 Ti is the same as a 670 but with less memory bandwidth, I was interested to see that your 100-credit jobs are running in 1,000 seconds rather than my 2,400. But, you're running Windows so perhaps the difference is there (i.e. your maxing out at 2 per 1000 secs might suggest that I max out at 5 per 2500 secs...). I'm currently waiting a few weeks for the RAC to level off before I try changing the number of instances, probably first to five, then back to three if that proves worse overall. Unfortunately I'm not aware of a GPU-Z for Linux -- nvidia-smi has a place to report GPU load but it's not available on the 660 Ti.
ID: 1411955 · Report as offensive
Profile Fred E.
Volunteer tester

Send message
Joined: 22 Jul 99
Posts: 768
Credit: 24,140,697
RAC: 0
United States
Message 1412148 - Posted: 6 Sep 2013, 14:17:06 UTC - in response to Message 1411955.  

Also, you may find that AP doesn't load your system as much as you might expect. I run 3 of these OpenCL tasks at a time on my 670 without much strain, but can only run 2 CUDA 5.0 Multibeam at a time without getting into problems.

Interesting, Fred. I put a GTX 660 Ti into a Linux box using a self-compiled CUDA 5.5 app a week ago and found that it runs 100-110 credit MB WUs in 40 mins or less with one, two, three, or four concurrent processes (it had a GTS 8800 512 in it before that). Since the 660 Ti is the same as a 670 but with less memory bandwidth, I was interested to see that your 100-credit jobs are running in 1,000 seconds rather than my 2,400. But, you're running Windows so perhaps the difference is there (i.e. your maxing out at 2 per 1000 secs might suggest that I max out at 5 per 2500 secs...). I'm currently waiting a few weeks for the RAC to level off before I try changing the number of instances, probably first to five, then back to three if that proves worse overall. Unfortunately I'm not aware of a GPU-Z for Linux -- nvidia-smi has a place to report GPU load but it's not available on the 660 Ti.

I'm still experimenting with values in mbcuda.cfg and count, and have not found the optimal settings. I look at the APR instead of RAC to evaluate setting changes, but you have to wait a few days after a change to get to a point where most validations are from new completions rather than from pendings at the time of change.

Guess we're waiting for some Astropulse tasks to see if Tazz has more questions.
Another Fred
Support SETI@home when you search the Web with GoodSearch or shop online with GoodShop.
ID: 1412148 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1412210 - Posted: 6 Sep 2013, 16:35:55 UTC

I've got everything selected to do AP, just waiting for some. I do not have any command line parameters set because I have no idea what they mean.
</Tazz>
ID: 1412210 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1412456 - Posted: 7 Sep 2013, 2:16:46 UTC - in response to Message 1412210.  

OOhhhh, Ohhhh, I got one. Let's see what happens!
</Tazz>
ID: 1412456 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1413094 - Posted: 9 Sep 2013, 0:13:52 UTC - in response to Message 1412456.  

OOhhhh, Ohhhh, I got one. Let's see what happens!


That one went well. It stopped 1 core to use and ran along with the MB task. 8269.98 seconds validated for 815.85 credits.

I added -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 to the ap_cmdline_win_x86_SSE2_OpenCL_NV.txt file. That will work, right?

</Tazz>
ID: 1413094 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1413184 - Posted: 9 Sep 2013, 7:14:04 UTC - in response to Message 1413094.  

OOhhhh, Ohhhh, I got one. Let's see what happens!


That one went well. It stopped 1 core to use and ran along with the MB task. 8269.98 seconds validated for 815.85 credits.

I added -unroll 10 -ffa_block 6144 -ffa_block_fetch 1536 to the ap_cmdline_win_x86_SSE2_OpenCL_NV.txt file. That will work, right?


Yes, it should work.



With each crime and every kindness we birth our future.
ID: 1413184 · Report as offensive
Profile Tazz
Volunteer tester
Avatar

Send message
Joined: 5 Oct 99
Posts: 137
Credit: 34,342,390
RAC: 0
Canada
Message 1414717 - Posted: 12 Sep 2013, 18:00:39 UTC

OK, I thought that I was beginning to understand how this worked but something don't seem right. This is the app_config that I started out with:

<app_config>
<app>
<name>astropulse_v6</name>
<max_concurrent>1</max_concurrent>
<gpu_versions>
<gpu_usage>.51</gpu_usage>
<cpu_usage>1.0</cpu_usage>
</gpu_versions>
</app>
<app>
<name>setiathome_v7</name>
<gpu_versions>
<gpu_usage>.49</gpu_usage>
<cpu_usage>.06</cpu_usage>
</gpu_versions>
</app>
</app_config>

From watching it run on two of my hosts I concluded that <max_concurrent>1</max_concurrent> will only let 1 AP run at a time so I didn't really need to divide up the GPU with the .49 and .51 since there's only one GPU in each of those two machines.

I changed the GPU usage to .50 for AP and .50 for MB (and yes I did save it), with BOINC running, and then did the "Read config file" command in BOINC. It's still showing that they're using 0.06CPUs + .49GPUs. I restarted the computer and it's still showing the 'old' settings.

In another host I have 2 video cards so I changed to <max_concurrent>2</max_concurrent> and left the .51 and .49
I'm thinking that this will not let 2 APs run on 1 GPU, but I'll have 1 AP and 1 MB on each.
</Tazz>
ID: 1414717 · Report as offensive

Message boards : Number crunching : New to AP, where to start?


 
©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.