app_info for AP500, AP503, MB603 and MB608

Message boards : Number crunching : app_info for AP500, AP503, MB603 and MB608
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 . . . 12 · Next

AuthorMessage
dyeman
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 23
Credit: 8,493,140
RAC: 0
Australia
Message 878122 - Posted: 22 Mar 2009, 0:33:41 UTC - in response to Message 876943.  

Not true. With 6.6.14 and above, you can receive (I have received) 603 work for AK_v8 as part of a standard work fetch from the project, as well as 608 CUDA work. You can also (and I have done this as a test, but I don't recommend it) ask for and receive AK_v8 work branded as 608, and CUDA work branded as 608, all in the same configuration.

The point is, the <plan_class>cuda</plan_class> line (in the 608 section in MarkJ's example) now serves to differentiate that section from the 603 section: MultiBeam and MultiBeam CUDA, although being identical tasks and datafiles, can now be tracked separately through the system and requested individually.

(And Fred - it's also why I'm looking forward with such anticipation to the 608 --> 603 rebranding script for VLAR. One tester, at your service whenever called upon!)


Is it possible to have a setup where MB WUs are only fetched for CUDA processing, but AK V8 is still defined? The reason for the question is that I'd prefer to do MBs only on CUDA and leave the CPU free for AP work. However Fred's 603 rebranding script will require the 603 AK V8 entry in app-info to process the vlars, in which case CPU-based 603 MB work might be fetched.
ID: 878122 · Report as offensive
Profile RottenMutt
Avatar

Send message
Joined: 15 Mar 01
Posts: 1011
Credit: 230,314,058
RAC: 0
United States
Message 878470 - Posted: 23 Mar 2009, 3:04:58 UTC
Last modified: 23 Mar 2009, 3:06:07 UTC

<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>

what is coproc count for?
ID: 878470 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 878517 - Posted: 23 Mar 2009, 6:47:40 UTC - in response to Message 878470.  

<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>

what is coproc count for?


As far as I understand its the number of coproc's it needs to process a single work unit. So in this case 1 wu needs 1 co-processor (or gpu) to run. If you have more co-processors then you can run more wu at the same time.
BOINC blog
ID: 878517 · Report as offensive
Profile Voyager
Volunteer tester
Avatar

Send message
Joined: 2 Nov 99
Posts: 602
Credit: 3,264,813
RAC: 0
United States
Message 879209 - Posted: 25 Mar 2009, 21:36:11 UTC

Can't find the boinc log with G flops on it. also don't have cuda 608exe. Will cuda mod exe. work?
ID: 879209 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 879214 - Posted: 25 Mar 2009, 21:45:04 UTC - in response to Message 879209.  
Last modified: 25 Mar 2009, 21:49:50 UTC

Can't find the boinc log with G flops on it. also don't have cuda 608exe. Will cuda mod exe. work?


If you want to use Raistmer's apps, use the 'MB_6.08_mod_CUDA_V10.exe' from his V11 download, instead of the stock app,
just add the file name to the app_info, and drop the file into your seti folder.

Claggy
ID: 879214 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 880358 - Posted: 29 Mar 2009, 7:16:15 UTC - in response to Message 879209.  

Can't find the boinc log with G flops on it. also don't have cuda 608exe. Will cuda mod exe. work?


In the boinc log (just click on the messages tab in BOINCmgr) it will be near the top, along with the device driver version and graphics card details. Don't forget all the extra zeroes seeing as its in gigaflops.
BOINC blog
ID: 880358 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 880392 - Posted: 29 Mar 2009, 10:36:03 UTC - in response to Message 879209.  

Can't find the boinc log with G flops on it. also don't have cuda 608exe. Will cuda mod exe. work?

Yes, it will.
YOu need either rename it to match stock app name, or do correct app_info file.
ID: 880392 · Report as offensive
Profile RottenMutt
Avatar

Send message
Joined: 15 Mar 01
Posts: 1011
Credit: 230,314,058
RAC: 0
United States
Message 880478 - Posted: 29 Mar 2009, 15:36:13 UTC - in response to Message 880392.  
Last modified: 29 Mar 2009, 16:13:45 UTC

what is wrong with my app_info.xml file, I replaced AK_v8_win_SSSE3x.exe with AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe. when ever bonic tries to execute a regular "setiathome_enhanced 6.03" (non cuda) it the application AK_v8b... crashes with a windows error and the returns the result as an error of coarse. cpu is a Q9450 and is ssse3 and sse4.1 compatible.

is the 6.03 wu not compatilble with AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe or is the problem with how the wu is passed off to the application?

or could it be because i used the 64 bit app???

<app_info>

<app>
<name>astropulse</name>
</app>
<file_info>
<name>ap_5.00r103_SSE3.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse</app_name>
<version_num>500</version_num>
<flops>8122702500</flops>
<file_ref>
<file_name>ap_5.00r103_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app>
<name>astropulse_v5</name>
</app>
<file_info>
<name>ap_5.03r112_SSE3.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse_v5</app_name>
<version_num>503</version_num>
<flops>18276080625</flops>
<file_ref>
<file_name>ap_5.03r112_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe</name>
<executable/>
</file_info>
<file_info>
<name>MB_6.08_mod_CUDA_V10.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft.dll</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3-1-1a_upx.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>603</version_num>
<platform>windows_intelx86</platform>
<flops>41121181406</flops>
<file_ref>
<file_name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe</file_name>
<main_program/>
</file_ref>
</app_version>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.127970</avg_ncpus>
<max_ncpus>0.127970</max_ncpus>
<flops>30000000000</flops>
<plan_class>cuda</plan_class>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V10.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
</app_version>
</app_info>

ID: 880478 · Report as offensive
frigens
Volunteer tester

Send message
Joined: 23 Feb 09
Posts: 3
Credit: 949,887
RAC: 0
United States
Message 880865 - Posted: 30 Mar 2009, 21:22:06 UTC

I've one GTX295 (counted as 2 CUDAs) @106GFLOPS each and 8600GTS @20GLOPS so I have total 3 CPU, should I use GTX295's GLOPS value, 8600's value or average (106+106+20)/3=77.33GLOPS in the calculation?

Appreciate any advise. Many thanks in advance
ID: 880865 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 880869 - Posted: 30 Mar 2009, 21:55:00 UTC - in response to Message 880865.  

I've one GTX295 (counted as 2 CUDAs) @106GFLOPS each and 8600GTS @20GLOPS so I have total 3 CPU, should I use GTX295's GLOPS value, 8600's value or average (106+106+20)/3=77.33GLOPS in the calculation?

Appreciate any advise. Many thanks in advance

Depends what you want to achieve, and how you want BOINC to work for you.

The idea of the FLOPS calculations in this thread was to compensate for the different efficiencies and estimates of the four different applications in the thread title. If you get the ratio right, there will be no systematic variation in the DCF (Duration Correction Factor) when a task finishes - because you have already applied your own correction factor.

In your situation, you have five correction factors to contend with (the final app being split into MB/608/GTX295 and MB/608/8600GTS), and only four FLOPS figures to juggle with. With a speed diffence of more than 5::1 between your CUDA cards, the sawteeth on your DCF graph will be sharp enough to cut down a rainforest.

The BOINC approach would be safety-first: plug in the value of your slowest card. Then, according to BOINC, you should never download too much work and risk missing deadlines. But that doesn't bear inspection: in between the (relatively rare) task completions by your 8600GTS card, DCF will be driven rapidly downwards by the (frequent) completions on the GTX295. Because, in current versions of BOINC, you only have a single DCF, and it controls the CPU apps as well, that could lead to you downloading an excess of AP tasks that you subsequently have difficulty completing on time.

So, thinking about it (and I'm making this up as I go along), I suggest you put the speed of the fastest card into Mark's formula. Every time the 8600GTS finishes a task, work fetch will be inhibited for a while (and, depending on the size of your cache, you may see a burst of EDF ('high priority') running on the CPU), but you should never be at risk of missing a deadline.
ID: 880869 · Report as offensive
Profile Voyager
Volunteer tester
Avatar

Send message
Joined: 2 Nov 99
Posts: 602
Credit: 3,264,813
RAC: 0
United States
Message 880880 - Posted: 30 Mar 2009, 22:06:20 UTC

I had my flops wrong for 503 (bad math)my DCF was at 2hrs. and I got mass 503s , changed it and it's back about right now.Worked out because this was just before the shut down.Now have lots of work but no 608s poor cuda.
ID: 880880 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 880973 - Posted: 31 Mar 2009, 11:57:28 UTC - in response to Message 880478.  

what is wrong with my app_info.xml file, I replaced AK_v8_win_SSSE3x.exe with AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe. when ever bonic tries to execute a regular "setiathome_enhanced 6.03" (non cuda) it the application AK_v8b... crashes with a windows error and the returns the result as an error of coarse. cpu is a Q9450 and is ssse3 and sse4.1 compatible.

is the 6.03 wu not compatilble with AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe or is the problem with how the wu is passed off to the application?

or could it be because i used the 64 bit app???

<app_info>

<app>
<name>astropulse</name>
</app>
<file_info>
<name>ap_5.00r103_SSE3.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse</app_name>
<version_num>500</version_num>
<flops>8122702500</flops>
<file_ref>
<file_name>ap_5.00r103_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app>
<name>astropulse_v5</name>
</app>
<file_info>
<name>ap_5.03r112_SSE3.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse_v5</app_name>
<version_num>503</version_num>
<flops>18276080625</flops>
<file_ref>
<file_name>ap_5.03r112_SSE3.exe</file_name>
<main_program/>
</file_ref>
</app_version>

<app>
<name>setiathome_enhanced</name>
</app>
<file_info>
<name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe</name>
<executable/>
</file_info>
<file_info>
<name>MB_6.08_mod_CUDA_V10.exe</name>
<executable/>
</file_info>
<file_info>
<name>cudart.dll</name>
<executable/>
</file_info>
<file_info>
<name>cufft.dll</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3-1-1a_upx.dll</name>
<executable/>
</file_info>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>603</version_num>
<platform>windows_intelx86</platform>
<flops>41121181406</flops>
<file_ref>
<file_name>AK_v8b_win_x64_SSSE3x_GPU_CPU_team_V10.exe</file_name>
<main_program/>
</file_ref>
</app_version>
<app_version>
<app_name>setiathome_enhanced</app_name>
<version_num>608</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.127970</avg_ncpus>
<max_ncpus>0.127970</max_ncpus>
<flops>30000000000</flops>
<plan_class>cuda</plan_class>
<file_ref>
<file_name>MB_6.08_mod_CUDA_V10.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>cudart.dll</file_name>
</file_ref>
<file_ref>
<file_name>cufft.dll</file_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3-1-1a_upx.dll</file_name>
</file_ref>
<coproc>
<type>CUDA</type>
<count>1</count>
</coproc>
</app_version>
</app_info>


You have the 64 bit app and the platform is saying you are a 32 bit one. Further up in the thread there is discussion about the platform to use.
BOINC blog
ID: 880973 · Report as offensive
Vager

Send message
Joined: 1 May 04
Posts: 10
Credit: 1,066,439
RAC: 0
Canada
Message 881513 - Posted: 2 Apr 2009, 2:59:42 UTC - in response to Message 880973.  

Does the new 6.6.17 client "Elapsed Time" counter now work properly with CUDA WU's. I've managed to get it working and I noticed that the counter is contstantly counting. My GPU thermals have gone up so to me it looks like its running ok. Anyone else notice this?

Vager
ID: 881513 · Report as offensive
Profile Voyager
Volunteer tester
Avatar

Send message
Joined: 2 Nov 99
Posts: 602
Credit: 3,264,813
RAC: 0
United States
Message 881549 - Posted: 2 Apr 2009, 4:23:33 UTC

Running 6.6.17 with MJ apfo just got some 608s and running fine 4 503s on cpu 1 608 on gpu.! cool no baby sitting. was waiting to get some mbs to see it work, and it does. Even though I screwed up on my 503 flops and got too many 503s ,all running high priority.did some math and aborted about 50 of them ,but even though my cpu has too many tasks 6.6.17 still dled cuda tasks so I'm a happy cruncher. I thought my rac at 8550 was pretty good but now with the gpu working..well who knows????????????????????
ID: 881549 · Report as offensive
Galadriel

Send message
Joined: 24 Jan 09
Posts: 42
Credit: 8,422,996
RAC: 0
Romania
Message 881810 - Posted: 2 Apr 2009, 22:42:45 UTC
Last modified: 2 Apr 2009, 22:43:37 UTC

hey guys.

one question/request

who can provide me a working app_info (using optimized apps) file just for astropulse and mb on a machine wich has no cuda capable card.
intented is the use of boin ver 6.6.18 or above.

thx in advance
ID: 881810 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 881816 - Posted: 2 Apr 2009, 23:14:24 UTC - in response to Message 881810.  

hey guys.

one question/request

who can provide me a working app_info (using optimized apps) file just for astropulse and mb on a machine wich has no cuda capable card.
intented is the use of boin ver 6.6.18 or above.

thx in advance

It's supplied with AP/MB combos, nothing new for 6.6.18 specifically needed for non-CUDA host IMHO.
ID: 881816 · Report as offensive
Galadriel

Send message
Joined: 24 Jan 09
Posts: 42
Credit: 8,422,996
RAC: 0
Romania
Message 881848 - Posted: 3 Apr 2009, 0:57:51 UTC - in response to Message 881816.  

ok thx for the reply raistmer. apreciated
ID: 881848 · Report as offensive
gmnewell

Send message
Joined: 19 Feb 09
Posts: 26
Credit: 13,751,185
RAC: 0
United States
Message 882010 - Posted: 3 Apr 2009, 17:06:06 UTC

This procedure worked great. Two comments and one observation:

Comment 1: In step #6, I think you mean p_fpops instead of p_flops.?.

Comment 2: It would be helpful if the instructions included multiple CUDA cards. I assumed that the number of CUDA needed to be 2 (instead of 1). But, I was not sure how to fill out the multibeam 608 GFlops rating. I averaged the two cards GFlop ratings.

Observation: BOINC version 6.6.20 did not take full advantage of both CUDA cards. In version 6.4.7, a single CPU core would keep both CUDA cards busy. Now, the second card is listed as "waiting to run".
ID: 882010 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 882041 - Posted: 3 Apr 2009, 19:23:31 UTC - in response to Message 882010.  

This procedure worked great. Two comments and one observation:

Comment 1: In step #6, I think you mean p_fpops instead of p_flops.?.

Comment 2: It would be helpful if the instructions included multiple CUDA cards. I assumed that the number of CUDA needed to be 2 (instead of 1). But, I was not sure how to fill out the multibeam 608 GFlops rating. I averaged the two cards GFlop ratings.

Observation: BOINC version 6.6.20 did not take full advantage of both CUDA cards. In version 6.4.7, a single CPU core would keep both CUDA cards busy. Now, the second card is listed as "waiting to run".

Richard Haselgrove answered the a question about Multiple Cuda cards flops values earlier in this thread,

Try changing the CUDA number back to 1, MarkJ seems to think it's the number of coproc's it needs to process a single work unit, see earlier in this thread.

Have you removed any <ncpus> entries from your cc_config?, Boinc 6.6.x doesn't need it anymore, it should be able to tell how many GPU's you have without using any work arounds.

Claggy
ID: 882041 · Report as offensive
MarkJ Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 08
Posts: 1139
Credit: 80,854,192
RAC: 5
Australia
Message 882169 - Posted: 4 Apr 2009, 3:54:53 UTC - in response to Message 882010.  

This procedure worked great. Two comments and one observation:

Comment 1: In step #6, I think you mean p_fpops instead of p_flops.?.

Comment 2: It would be helpful if the instructions included multiple CUDA cards. I assumed that the number of CUDA needed to be 2 (instead of 1). But, I was not sure how to fill out the multibeam 608 GFlops rating. I averaged the two cards GFlop ratings.

Observation: BOINC version 6.6.20 did not take full advantage of both CUDA cards. In version 6.4.7, a single CPU core would keep both CUDA cards busy. Now, the second card is listed as "waiting to run".


Agree with comment 1, it is now <p_fpops>. Not sure if it was a typo or the older version (6.6.15 when I wrote it).

BOINC should see all the cuda-capable cards and they should work without any additional changes. It does have a problem when you mix different cards with different f_pops as you can only give BOINC one value.

As Claggy mentioned if you have an <ncpus> statement in your cc_config file you should remove it as its no longer required under the 6.6 versions of BOINC.

Also as Claggy mentioned the coproc <count> is used to tell BOINC how many cuda devices are needed to run a single wu, so this should remain at 1.
BOINC blog
ID: 882169 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 . . . 12 · Next

Message boards : Number crunching : app_info for AP500, AP503, MB603 and MB608


 
©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.