+- +-
Say hello if visiting :) by Mike
23 Apr 2021, 08:55:37 am

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: Linux port of Alex Kan's?  (Read 25836 times)

Offline sunu

  • Alpha Tester
  • Knight who says 'Ni!'
  • ***
  • Posts: 771
Re: Linux port of Alex Kan's?
« Reply #45 on: 18 Mar 2009, 08:54:26 pm »
The science done by optimized apps is the same. You don't miss anything.  AKv8 is the fastest one.

Offline Josef W. Segur

  • Janitor o' the Board
  • Knight who says 'Ni!'
  • *****
  • Posts: 3112
Re: Linux port of Alex Kan's?
« Reply #46 on: 18 Mar 2009, 09:59:36 pm »
Are there any disadvantages in using AK v8 or 2.4L compared to the default 6.03? Are they able to use all work units that 6.03 can use and do they perform the same analysis? Or does 6.03 have a newer, better analysis and with using an older piece of software I might miss signals?
So what made the version numbers rise from 5.x to 6.x?

The graphics for stock builds. 6.x are built with a separate graphics app in order to match what BOINC 6 had to do to work within Windows Vista guidelines.

Note that 6.03 doesn't run on Win95, and Eric Korpela's advice was to use 5.27 with an app_info.xml file. As far as the signal analysis is concerned, any version of setiathome_enhanced back to the 5.1x versions is fine. Credit claims can be enough different to annoy other users, though.

The CUDA app (both stock and the builds included in Raistmer's packages) has some things missing so in some cases it bails out with an "unsupported function" error (though BOINC doesn't know that's what the -12 exit code means so it calls it "unknown").
                                                                             Joe

Lysia

  • Guest
Re: Linux port of Alex Kan's?
« Reply #47 on: 19 Mar 2009, 09:06:42 am »
So it's safe to use 2.4L (SSE only) and pretend having 5.27 ?

Offline Richard Haselgrove

  • Messenger Pigeon
  • Knight who says 'Ni!'
  • *****
  • Posts: 2819
Re: Linux port of Alex Kan's?
« Reply #48 on: 19 Mar 2009, 10:00:42 am »
So it's safe to use 2.4L (SSE only) and pretend having 5.27 ?

Yes

Lysia

  • Guest
Re: Linux port of Alex Kan's?
« Reply #49 on: 19 Mar 2009, 03:45:51 pm »
Credit claims can be enough different to annoy other users, though.

It annoys me, too. Credits claimed dropped by roughly a factor of 3, computation time dropped by a factor of 2-2.5. So in the same time I do more work for less credits than with the default app. I guess one needs SSE2 or SSE3 to have enough speedup to compensate the lesser credits with the increased number of WUs.

So I have to decide if I care most about credits, than I use the default app again, or if I care most about science and helping the project, then I stay with the optimized app.

Perhaps I should just try AP, although I guess it will take weeks to figure out if that pays off.

Offline sunu

  • Alpha Tester
  • Knight who says 'Ni!'
  • ***
  • Posts: 771
Re: Linux port of Alex Kan's?
« Reply #50 on: 19 Mar 2009, 04:17:53 pm »
Perhaps I should just try AP, although I guess it will take weeks to figure out if that pays off.

I think right now AP 5.03 gives more credits that MB, but it takes a while to pick up.

Offline Josef W. Segur

  • Janitor o' the Board
  • Knight who says 'Ni!'
  • *****
  • Posts: 3112
Re: Linux port of Alex Kan's?
« Reply #51 on: 19 Mar 2009, 08:30:55 pm »
Credit claims can be enough different to annoy other users, though.

It annoys me, too. Credits claimed dropped by roughly a factor of 3, computation time dropped by a factor of 2-2.5. So in the same time I do more work for less credits than with the default app. I guess one needs SSE2 or SSE3 to have enough speedup to compensate the lesser credits with the increased number of WUs.

So I have to decide if I care most about credits, than I use the default app again, or if I care most about science and helping the project, then I stay with the optimized app.

Perhaps I should just try AP, although I guess it will take weeks to figure out if that pays off.

Actually, the claims from 2.4L versions are fine, in those builds Crunch3r had implemented the code to comply with the credit_rate in the WU header. The low credits you've seen so far are because you got some Very High Angle Range work (aka 'shorties').

I concur that if you're patient optimized AP_v5 work should give higher credit/time than MB work. The patience is needed not only to get the work done but also because wingmates will often abort or error out or simply fail to complete the work, so the credit grant can be delayed a lot.
                                                                                Joe

Lysia

  • Guest
Re: Linux port of Alex Kan's?
« Reply #52 on: 20 Mar 2009, 05:37:34 am »
Actually, the claims from 2.4L versions are fine, in those builds Crunch3r had implemented the code to comply with the credit_rate in the WU header. The low credits you've seen so far are because you got some Very High Angle Range work (aka 'shorties').

I now had a WU that gave nearly the credits of the WUs before (42.6). It seems as if the speed improvement for SSE-only CPUs is in a region of about 5%, which isn't even statistically significant with only one WU done. Perhaps the version with fftw is better, but these 5% are not worth the effort of finding and installing the optimized app.

I concur that if you're patient optimized AP_v5 work should give higher credit/time than MB work. The patience is needed not only to get the work done but also because wingmates will often abort or error out or simply fail to complete the work, so the credit grant can be delayed a lot.

One week to go...

 

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: 59558
Total Topics: 1672
Most Online Today: 12
Most Online Ever: 983
(20 Jan 2020, 03:17:55 pm)
Users Online
Members: 0
Guests: 7
Total: 7
Powered by EzPortal