Astropulse CPU vs GPU?

Message boards : Number crunching : Astropulse CPU vs GPU?
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · Next

AuthorMessage
Profile Cliff Harding
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1432
Credit: 110,967,840
RAC: 67
United States
Message 1334970 - Posted: 5 Feb 2013, 21:41:15 UTC
Last modified: 5 Feb 2013, 21:41:43 UTC

Ever since I started working Opencl WUs, I noticed that I get them in drips & drabs. Out of the 44 AP tasks that are now processing only 4 are tagged for the GPU. How does the scheduler differentiate between which get tagged for CPU work vs GPU work? And, is there a way other than stopping all CPU WUs of having the scheduler send out more GPU work?


I don't buy computers, I build them!!
ID: 1334970 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34250
Credit: 79,922,639
RAC: 80
Germany
Message 1334971 - Posted: 5 Feb 2013, 21:45:06 UTC
Last modified: 5 Feb 2013, 21:45:28 UTC

The only way to do it is to disable CPU work fetch for a while.
With actual server conditions not that easy.
Otherwise its simple luck.


With each crime and every kindness we birth our future.
ID: 1334971 · Report as offensive
Profile skildude
Avatar

Send message
Joined: 4 Oct 00
Posts: 9541
Credit: 50,759,529
RAC: 60
Yemen
Message 1334972 - Posted: 5 Feb 2013, 21:50:44 UTC

you could use the rescheduler to move the WU's to the GPU or if you don't want AP work on your CPU, don't include the CPU AP portion in your app_info


In a rich man's house there is no place to spit but his face.
Diogenes Of Sinope
ID: 1334972 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1334976 - Posted: 5 Feb 2013, 21:56:35 UTC - in response to Message 1334970.  

The easiest way is too untick 'Use CPU' and untick Run only the selected applications 'SETI@home Enhanced', then Astropulse Wu's can only go to the GPU.

Claggy
ID: 1334976 · Report as offensive
spitfire_mk_2
Avatar

Send message
Joined: 14 Apr 00
Posts: 563
Credit: 27,306,885
RAC: 0
United States
Message 1335540 - Posted: 7 Feb 2013, 18:55:28 UTC

Since I run Lunatics and have app_info.xml file, I followed the above advice and edited the app_info file. Took out following chunk:
<app>
<name>astropulse_v6</name>
</app>
<file_info>
<name>ap_6.01r557_SSE2_331_AVX.exe</name>
<executable/>
</file_info>
<app_version>
<app_name>astropulse_v6</app_name>
<version_num>601</version_num>
<file_ref>
<file_name>ap_6.01r557_SSE2_331_AVX.exe</file_name>
<main_program/>
</file_ref>
</app_version>
<app>
<name>astropulse_v505</name>
</app>
<app_version>
<app_name>astropulse_v505</app_name>
<version_num>505</version_num>
<cmdline>-v505</cmdline>
<file_ref>
<file_name>ap_6.01r557_SSE2_331_AVX.exe</file_name>
<main_program/>
</file_ref>
</app_version>

Everything seems to be running right.
ID: 1335540 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336318 - Posted: 9 Feb 2013, 20:18:58 UTC

Hi, just wondering...

How many AP tasks can a GF 560Ti 1.3Gb or 660GTX 2Gb do at a time? How many CPU cores would that need?

Just want to know if it is worth waiting for Linux version of AP cuda/OpenCL.



p.s.
I'm happy with my i7-3930K CPU doing 6 AP units simultaneously producing 6 completed AP results every 13500 seconds while it is also crunching 6 GPU MB processes on above mentioned cards both doing 3 at a time. The 3930K uses Lunatics linux AVX build. CPU temperature rises to 74-79C, depending on core, when doing 6 AP simultaneously. GPU's stay at 63C.


To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1336318 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22149
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1336368 - Posted: 9 Feb 2013, 21:51:44 UTC

Using my GTX460 as a guide - two or three at a time, each taking about one hour. You i7 takes about 4 hours to do one per core active.

Remember, your actual mileage may vary....
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1336368 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336384 - Posted: 9 Feb 2013, 22:19:24 UTC - in response to Message 1336368.  

Using my GTX460 as a guide - two or three at a time, each taking about one hour. You i7 takes about 4 hours to do one per core active.

Remember, your actual mileage may vary....


Thanks for the info. So about 1400-2100 points per hour for a GTX460.

I'd be happy to know ..
How many cores do you have to reserve for a task, 1 per GPU or 1 per task? Or how many seconds CPU time does it take for a task that finishes in an hour?


To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1336384 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22149
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1336387 - Posted: 9 Feb 2013, 22:25:16 UTC

Something like that - I don't look at the points per hour very often.
I don't have any cores reserved on either of my GPU rigs. I did try it once on the one with a 460, and it might have sped up the GPU by a couple of percent, but I lost more by not having the credit from all six CPU cores. The CPU is a 6 core Phenom, and as I don't have any Intel based crunchers I can't comment on any advantages or otherwise by having a core reserved on say an i7.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1336387 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34250
Credit: 79,922,639
RAC: 80
Germany
Message 1336396 - Posted: 9 Feb 2013, 22:33:17 UTC

Rob you are very slow compared to other 460´s.
Do you check you results.
13000 seconds instead of 3500.
You would benefit a lot reserving a core.



With each crime and every kindness we birth our future.
ID: 1336396 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336398 - Posted: 9 Feb 2013, 22:35:53 UTC - in response to Message 1336387.  

Thank You Rob.
ID: 1336398 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336401 - Posted: 9 Feb 2013, 22:39:30 UTC - in response to Message 1336396.  

Rob you are very slow compared to other 460´s.
Do you check you results.
13000 seconds instead of 3500.
You would benefit a lot reserving a core.


Hi Mike,

The 13500 is from my CPU (6 units at that time). Rob is doing about 3 tasks in an hour (3600s).

Do you know how many seconds CPU time it takes for a task that finishes in an hour?

ID: 1336401 · Report as offensive
Profile Cliff Harding
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1432
Credit: 110,967,840
RAC: 67
United States
Message 1336403 - Posted: 9 Feb 2013, 22:41:56 UTC
Last modified: 9 Feb 2013, 22:45:02 UTC

Check out my i7/950 -- it is running 1 x GTX660SC/2Gb & 1 x GTX460SE/1Gb. Both are running Lunatics _x41zc cuda 5.0 @ 3 each & Opencl _r1761 @ 1 each. CPU MB tasks run an avg of 11 hrs, GPU MB avg 20-25 min, & AP Opencl avg 45 min. Normally I run at 80% of processors ( 6 cores). Opencl will explicitly reserve 1 core for each task running, and it will release it when the task is finished. Not the fastest in the Seti universe, but I am happy with the results so far. The machine is a Win7/64-bit running BOINC 7.0.48 and nVidia 313.96.


I don't buy computers, I build them!!
ID: 1336403 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22149
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1336406 - Posted: 9 Feb 2013, 22:47:22 UTC - in response to Message 1336401.  

Rob you are very slow compared to other 460´s.
Do you check you results.
13000 seconds instead of 3500.
You would benefit a lot reserving a core.


Hi Mike,

The 13500 is from my CPU (6 units at that time). Rob is doing about 3 tasks in an hour (3600s).

Do you know how many seconds CPU time it takes for a task that finishes in an hour?



Not quite what I said, my figure was based on an extrapolation of what my 460 delivers and knowing how much faster a 660 is than a 460 (I should have made that a bit clearer)
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1336406 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34250
Credit: 79,922,639
RAC: 80
Germany
Message 1336412 - Posted: 9 Feb 2013, 22:54:22 UTC - in response to Message 1336401.  
Last modified: 9 Feb 2013, 23:01:50 UTC

Rob you are very slow compared to other 460´s.
Do you check you results.
13000 seconds instead of 3500.
You would benefit a lot reserving a core.


Hi Mike,

The 13500 is from my CPU (6 units at that time). Rob is doing about 3 tasks in an hour (3600s).

Do you know how many seconds CPU time it takes for a task that finishes in an hour?



No i checked Robs host.
Just today 3 WUs longer than 13000 seconds.
Thats because driver hangs because of lack of CPU cycles.
Wouldn`t happen with a free CPU core.
He is wasting GPU cycles.


With each crime and every kindness we birth our future.
ID: 1336412 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336416 - Posted: 9 Feb 2013, 23:01:51 UTC - in response to Message 1336403.  
Last modified: 9 Feb 2013, 23:02:36 UTC

Check out my i7/950 -- it is running 1 x GTX660SC/2Gb & 1 x GTX460SE/1Gb. Both are running Lunatics _x41zc cuda 5.0 @ 3 each & Opencl _r1761 @ 1 each. CPU MB tasks run an avg of 11 hrs, GPU MB avg 20-25 min, & AP Opencl avg 45 min. Normally I run at 80% of processors ( 6 cores). Opencl will explicitly reserve 1 core for each task running, and it will release it when the task is finished. Not the fastest in the Seti universe, but I am happy with the results so far. The machine is a Win7/64-bit running BOINC 7.0.48 and nVidia 313.96.


Thank You Cliff,

I'll take a look at your computer(s) (the i7/950) on sunday.

Did I understand correctly: You are doing 3 GPU MB (cuda) and one GPU AP (opencl) at a time on both cards and an additional 4-6 CPU tasks. (4 when runnig AP opencl on both cards)

Now it is time for me to go to sleep. It is 1.00 AM in Finland.

Happy Crunching!
ID: 1336416 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22149
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1336418 - Posted: 9 Feb 2013, 23:02:59 UTC

I did try running with one core feeding the GPU about 6 months ago (with the current hardware and software set up) and the overall throughput (total tasks per hour)was less with one core set aside for feeding the GPU than having the default settings where there is a "free for all". I have no doubt that for other CPU+GPU configurations the results would be different.

(I have run GPU-X on a few occasions, and the load time (time when the GPU is essential idle, is only a few seconds - between 5 and 10, not a lot when the run time is way over a thousand).
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1336418 · Report as offensive
Profile petri33
Volunteer tester

Send message
Joined: 6 Jun 02
Posts: 1668
Credit: 623,086,772
RAC: 156
Finland
Message 1336420 - Posted: 9 Feb 2013, 23:07:30 UTC - in response to Message 1336412.  

Rob you are very slow compared to other 460´s.
Do you check you results.
13000 seconds instead of 3500.
You would benefit a lot reserving a core.


Hi Mike,

The 13500 is from my CPU (6 units at that time). Rob is doing about 3 tasks in an hour (3600s).

Do you know how many seconds CPU time it takes for a task that finishes in an hour?



No i checked Robs host.
Just today 3 WUs longer than 13000 seconds.
Thats because driver hangs because of lack of CPU cycles.
Wouldn`t happen with a free CPU core.
He is wasting GPU cycles.


Oh, Yes. (My bad, i did not check.) :D


To overcome Heisenbergs:
"You can't always get what you want / but if you try sometimes you just might find / you get what you need." -- Rolling Stones
ID: 1336420 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34250
Credit: 79,922,639
RAC: 80
Germany
Message 1336421 - Posted: 9 Feb 2013, 23:07:45 UTC - in response to Message 1336418.  

I did try running with one core feeding the GPU about 6 months ago (with the current hardware and software set up) and the overall throughput (total tasks per hour)was less with one core set aside for feeding the GPU than having the default settings where there is a "free for all". I have no doubt that for other CPU+GPU configurations the results would be different.

(I have run GPU-X on a few occasions, and the load time (time when the GPU is essential idle, is only a few seconds - between 5 and 10, not a lot when the run time is way over a thousand).


Thats just because 1 core is not enough on AMD machines.

I finnish 2 APs in less an hour.
Thats ~3000 seconds.
You finnish in about 13000 - 15000 seconds.
Do you see the difference ?
I`m running APs over 2 years now and am fully aware of each conditions.

Claggy has also a 460 and does faster than my ATI.



With each crime and every kindness we birth our future.
ID: 1336421 · Report as offensive
Profile Cliff Harding
Volunteer tester
Avatar

Send message
Joined: 18 Aug 99
Posts: 1432
Credit: 110,967,840
RAC: 67
United States
Message 1336424 - Posted: 9 Feb 2013, 23:10:29 UTC - in response to Message 1336416.  

Check out my i7/950 -- it is running 1 x GTX660SC/2Gb & 1 x GTX460SE/1Gb. Both are running Lunatics _x41zc cuda 5.0 @ 3 each & Opencl _r1761 @ 1 each. CPU MB tasks run an avg of 11 hrs, GPU MB avg 20-25 min, & AP Opencl avg 45 min. Normally I run at 80% of processors ( 6 cores). Opencl will explicitly reserve 1 core for each task running, and it will release it when the task is finished. Not the fastest in the Seti universe, but I am happy with the results so far. The machine is a Win7/64-bit running BOINC 7.0.48 and nVidia 313.96.


Thank You Cliff,

I'll take a look at your computer(s) (the i7/950) on sunday.

Did I understand correctly: You are doing 3 GPU MB (cuda) and one GPU AP (opencl) at a time on both cards and an additional 4-6 CPU tasks. (4 when runnig AP opencl on both cards)

Now it is time for me to go to sleep. It is 1.00 AM in Finland.

Happy Crunching!


Yep, that's what I do! This is all running with defaults except with the count for the cuda tasks. If anyone knows how to tweak the apps to get more out of them let me know.


I don't buy computers, I build them!!
ID: 1336424 · Report as offensive
1 · 2 · 3 · Next

Message boards : Number crunching : Astropulse CPU vs GPU?


 
©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.