+- +-
Say hello if visiting :) by Gecko
11 Jan 2023, 07:43:05 pm

Seti is down again by Mike
09 Aug 2017, 10:02:44 am

Some considerations regarding OpenCL MultiBeam app tuning from algorithm view by Raistmer
11 Dec 2016, 06:30:56 am

Loading APU to the limit: performance considerations by Mike
05 Nov 2016, 06:49:26 am

Better sleep on Windows - new round by Raistmer
26 Aug 2016, 02:02:31 pm

Author Topic: AK V8 + CUDA MB team work mod  (Read 134638 times)

Offline Slawek

  • Knight o' The Realm
  • **
  • Posts: 55
Re: AK V8 + CUDA MB team work mod
« Reply #15 on: 26 Jan 2009, 05:39:59 pm »
If Athlon work better on SSE2 .. why not :P

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: AK V8 + CUDA MB team work mod
« Reply #16 on: 26 Jan 2009, 06:03:53 pm »
Well, it seems new ICC have even worse support of AMD chips than it has before so SSE3 AMD compatible build will take more time than I expected...  :-\

Kiva

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #17 on: 27 Jan 2009, 05:50:14 am »
@ Raistmer

Your hard work is much appreciated - thanks :bow:

regards
Kim Vater, Team Norway.

Offline Marius

  • Knight o' The Realm
  • **
  • Posts: 84
Re: AK V8 + CUDA MB team work mod
« Reply #18 on: 27 Jan 2009, 07:17:03 pm »
Good news, and i'm very interested in this release! (A8v8 is the best!)

Forgive my ignorance in this matter, but what is the best way to install this while keeping the current queue (if that's possible). Currently i have a seti queue of nearly 700 workunits for my quad. Normally i would use "no new tasks" and let it run until the queue is empty, then copy the new files into BOINC\projects\setiathome.berkeley.edu, then allowing new tasks. But it would take several day's to flush the queue and start the AKv8b/cuda. Any tips?

popandbob

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #19 on: 27 Jan 2009, 11:01:11 pm »
As expected it only uses 1 GPU on multi GPU setup...
So far so good though...

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: AK V8 + CUDA MB team work mod
« Reply #20 on: 28 Jan 2009, 03:25:26 am »
@ Raistmer

Your hard work is much appreciated - thanks :bow:

regards
Kim Vater, Team Norway.

Thanx.

AMD-friendly version now available (see attachment at first post). It's restricted in SSE level to SSE3 that some Athlon 64 and all AMD's x2 and Phenoms should support.
File: Raistmer's_opt_package_V8a_CPU_GPU_team_SSE3.rar
Additional changes: improved elapsed time representation.
« Last Edit: 28 Jan 2009, 04:43:52 am by Raistmer »

The Naja

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #21 on: 28 Jan 2009, 06:20:43 am »
Just installed your version on my Q6600 + 8600GT, taking the opportunity of the lack of new MU from S@H (servers overloaded)

- i configured CPU to 5 in my prefs
- I authorized GPU usage in my prefs

Before, was only using ak_V8 and APr103, without GPU

Now, I'm waiting some WU to be D/L, crossing fingers...

Don't hesitate to tell me if you see something wrong in my setup... ;-)

Cheers

The Naja

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #22 on: 28 Jan 2009, 07:58:27 am »
Hello,

I just received some WU... I would like some confirmations.

Are CUDA units tagged differently in the 'application' column of BOINC Manager ?

The point is:

- 4 WU are processed in parallel
- all of them are marked as 'setiathome_enhanced 6.08'
- the CPU cores are at 75%, meaning only 3 of them are used
- my screen refreh rate being reduced (laaaag), for sure the GPU is used
- I even put '6' in the CPU to be used on the BOINC prefs website, and I made a refresh afterwards...

Is this running as 'spec' ? How to have 5 WU processed in parallel ?

Thanks for all your tips...

Cheers

EDIT: I confirm that GPU is used: http://setiathome.berkeley.edu/result.php?resultid=1137360552 for instance

So it really means that only 3 CPU are used even if I put 6 cores (4 CPU +1 GPU +1) in the website prefs...
« Last Edit: 28 Jan 2009, 08:04:05 am by The Naja »

Offline Richard Haselgrove

  • Messenger Pigeon
  • Knight who says 'Ni!'
  • *****
  • Posts: 2819
Re: AK V8 + CUDA MB team work mod
« Reply #23 on: 28 Jan 2009, 09:19:05 am »

So it really means that only 3 CPU are used even if I put 6 cores (4 CPU +1 GPU +1) in the website prefs...


You need to put the extra CPU (5, in your case) in a cc_config.xml local file. The website can only limit CPU numbers downward - it doesn't go up to 11.

The Naja

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #24 on: 28 Jan 2009, 09:22:44 am »
Thanks for your answer and for the clue,

Would yit be possible to have a link to such a template, and also in which folder to put it ?

Cheers

Offline Richard Haselgrove

  • Messenger Pigeon
  • Knight who says 'Ni!'
  • *****
  • Posts: 2819
Re: AK V8 + CUDA MB team work mod
« Reply #25 on: 28 Jan 2009, 09:57:25 am »
Thanks for your answer and for the clue,

Would yit be possible to have a link to such a template, and also in which folder to put it ?


There's plenty of discussion on the main board, but here goes:

Code: [Select]
<cc_config>
<options>
<ncpus>5</ncpus>
</options>
</cc_config>

(from BOINC Client configuration)

As stated on that page, it goes at the root level of your BOINC data folder.

Once in place, you can load and activate it from the advanced menu in BOINC Manager, 'Read config file'. When you change the number of CPUs this way, expect BOINC to re-run benchmarks.

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: AK V8 + CUDA MB team work mod
« Reply #26 on: 28 Jan 2009, 08:02:15 pm »
And now test on SSE3-capable Athlon 64 (Venice)

WU : PG0009.wu
AK_v8b_win_SSE3.exe : 1082.844 secs CPU
AK_v8b_win_SSE2_GPU_CPU_team.exe : 1035.328 secs CPU
Speedup     : 4.39%
Ratio       : 1.05 x

WU : PG0395.wu
AK_v8b_win_SSE3.exe : 1012.813 secs CPU
AK_v8b_win_SSE2_GPU_CPU_team.exe : 986.641 secs CPU
Speedup     : 2.58%
Ratio       : 1.03 x

WU : PG0444.wu
AK_v8b_win_SSE3.exe : 871.516 secs CPU
AK_v8b_win_SSE2_GPU_CPU_team.exe : 824.000 secs CPU
Speedup     : 5.45%
Ratio       : 1.06 x

WU : PG1327.wu
AK_v8b_win_SSE3.exe : 946.563 secs CPU
AK_v8b_win_SSE2_GPU_CPU_team.exe : 875.781 secs CPU
Speedup     : 7.48%
Ratio       : 1.08 x

As you can see SSE2 build performs better than SSE3 one on SSE3-capable (early) AMD. Maybe on latest Phenom SSE3-situation was improved ?
Could someone of our pre-testers or regular users try to run KWSN bench for AK_v8b_SSE3 and AK_v8b_SSE2 on new Phenom CPU to shed light on current situation with AMD SSE3 support quality ?



Offline Jason G

  • Construction Fraggle
  • Knight who says 'Ni!'
  • *****
  • Posts: 8980
Re: AK V8 + CUDA MB team work mod
« Reply #27 on: 28 Jan 2009, 08:20:21 pm »
It is quite possible the SSE3 build, being a generic P4 switch (QxO or QxP?) , is assuming the presence of hardware prefetchers that those generations of AMD don't have, and of course long Pipelines (Along with the IPP libraries too).  It's difficult that the two architectures are so different at that period.  I'm leaning more and more towards a preference for isolating the core functions, maybe even into DLL's one day, but starting simply with a delay loaded choice of FFTs might be a start.

The Naja

  • Guest
Re: AK V8 + CUDA MB team work mod
« Reply #28 on: 29 Jan 2009, 06:59:10 am »


As stated on that page, it goes at the root level of your BOINC data folder.

Once in place, you can load and activate it from the advanced menu in BOINC Manager, 'Read config file'. When you change the number of CPUs this way, expect BOINC to re-run benchmarks.

Haselgrove: thanks a lot for your explanations.

I ran a few tests yesterday based on your message.

It went fine on configuration point of view, but my computer was nut usable anymore: screen refresh was horrible: an alt-tab between 2 applications was taking at least 8 seconds.

Graphic drivers: latest from Nvidia website from last Saturday.

I had to roll back to AK_V8 + AP r103 package I was using before, no GPU used.

Hope this can give some clues for you guys, thanks again...

Offline Raistmer

  • Working Code Wizard
  • Volunteer Developer
  • Knight who says 'Ni!'
  • *****
  • Posts: 14349
Re: AK V8 + CUDA MB team work mod
« Reply #29 on: 29 Jan 2009, 06:46:47 pm »
New combo for AMD CPUs with SSE3 support is available.
At least for early AMD CPUs with SSE3 support (x86 mode) AK_v8 SSE2 version works faster than SSE3 one.
So this build aimed for such CPUs (SSE3 AP and SSE2 AK_v8).
If new Phenoms don'r show such speed degradation on SSE3 instruction set I would like to know it ( with benchmark results posted of course).
Package name: Raistmer's_opt_package_V8a_CPU_GPU_team_SSE3_AMD.rar
Attached to first thread post.

 

Welcome, Guest.
Please login or register.
 
 
 
Forgot your password?
Members
Total Members: 97
Latest: ToeBee
New This Month: 0
New This Week: 0
New Today: 0
Stats
Total Posts: 59559
Total Topics: 1672
Most Online Today: 40
Most Online Ever: 983
(20 Jan 2020, 03:17:55 pm)
Users Online
Members: 0
Guests: 37
Total: 37
Powered by EzPortal