Seti@Home optimized science apps and information

Optimized Seti@Home apps => Windows => Topic started by: Simon on 13 Nov 2006, 10:42:59 pm

Title: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 13 Nov 2006, 10:42:59 pm
Hi folks,

read up at http://setiathome.berkeley.edu/forum_thread.php?id=31810&nowrap=true#457400 about the new release, please :) It's 4:41 AM and I'm too tired to type it again.

The post title says it all - check the Downloads section for the new apps!
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: BenHer on 14 Nov 2006, 12:05:00 am
You da 'sleepy' man Simon.  8)
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 14 Nov 2006, 03:44:56 am

Hello, or sweet dreams now Simon! ;)

Thanx for the new apps!

The apps are now faster, too?

Greetings! (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)


Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: pepperammi on 14 Nov 2006, 07:21:25 am
Done a complete test/comparison run for the new Rev2 apps.
SSE, SSE2 Generic, SSE2 Intel and SSE3 Intel on my Pentium D 830 @ 3.2 Ghz. I've done them with the added -bench command if maybe there's any extra info for you. All working perfectly and looks like the adaptive optimisation I think you called it working perfectly too. All strongly similar :D

Fantastic work everybody ;D

Only thing I notice is the stderr output is still saying "(R-1.3" In boinc. This could be just because it was started with the previous app though.

[EDIT] added cleared up times and stderr output

[attachment deleted by admin]

[attachment deleted by admin]
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 14 Nov 2006, 12:14:16 pm
Hello!

I had crunched a WU 2/3 with Crunch3rs... then the rest 1/3 with Simons app.
I stopped the crunching one times... (because of the doubled entry in "stderr_txt"...)

Now I cannot see that Simon and the other good guys had made the app.
This is what the optimizers wanted or is this an error?
It was running an my AMD K7 Athlon 600Mhz.
(SaH_5.15_KWSN_MMX_Ben-Joe_2.0_A)

Greetings! (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)


--------------------------------------------------------
Windows optimized S@H Enhanced application by Crunch3r
Version info: Windows MMX V5.12 by Crunch3r

Work Unit Info:
...............
WU true angle range is : 0.4265

ChirpData--[unrolled]: 29520293 (chosen)
  GetPeak--[hand_opt]: 3512 (chosen)
        f_sum--[unroll4]: 72519 (chosen)
GetChiSq--[hoisted+abs(]: 24435 (chosen)
Bench Time: 9.92 seconds
work_len=1048576
ChirpData--[unrolled]: 29284930 (chosen)
  GetPeak--[hand_opt]: 3107 (chosen)
        f_sum--[unroll4]: 72231 (chosen)
GetChiSq--[hoisted+abs(]: 24403 (chosen)
Bench Time: 9.94 seconds
work_len=1048576

Spikes Pulses Triplets Gaussians Flops
     0       0       0           0      16119038856279
--------------------------------------------------------


Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: BenHer on 14 Nov 2006, 01:24:49 pm
Dirk,

The simon app does not display any "tag" information on WUs when it is started beyond about 1% of WU crunching. 

This way, when other projects, or shutting down the computer, and restarting cause the program to be loaded a 2nd or 3rd time for the same WU, it wont have a series of "Crunched by KWSN" or similar in the stderr.txt file.

Any new WU started by it will have the tag lines.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: rgathright on 14 Nov 2006, 02:16:03 pm
I am seeing good results so far on my Xeon 5120.  Thanks!   :o
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 14 Nov 2006, 02:48:44 pm
You da 'sleepy' man Simon.  8)
Yeah, good thing I could take half a day off today cause I definitely couldn't have survived on 3 hours sleep.

I did get some fellow KWSN members to test the apps I couldn't test myself (MMX+SSE only) yesterday, so I got the whole bunch uploaded and released after all. Ni!

Regards,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: BenHer on 14 Nov 2006, 02:49:46 pm
Rgathright - link to your computer info on seti please.

Also - 5120 is a Core duo DP xeon (woodcrest).  And during our testing we found that Simon's older 1.3 was faster than the new release on his woodcrest.

Please confirm.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 14 Nov 2006, 02:52:21 pm
Ben's correct,

the 1.3 and 1.41 Core 2 / SSSE3 (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=30) apps are actually quicker (at least, they were in our testing) than the 2.0-based ones.

So we're of course interested how it's working out for you :)

Please post a host link to this computer to satisfy our curiosity.

Regards,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: rgathright on 14 Nov 2006, 03:21:29 pm
http://setiathome.berkeley.edu/show_host_detail.php?hostid=2722436

 I am getting some strange WU times...
2:28:28 67% for 09jn03aa.22751.29217.786080.3.203
2:27:53 63% for 09jn03aa.22751.29217.786080.3.198

I hope these are just really long WU's.

I used an old version of "kwsn-test" (7/5/2006) and saw a gain of 40 seconds on this computer vs the other results.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 14 Nov 2006, 04:13:07 pm
Hello!

After the installation of the new app...
I got now new work...

In the overview in Boinc I have now WUs for S@H V5.17...
(Before V5.15 with Crunch3rs V5.12)

In the "app_info.xml" of the new Simon- app: "5.12, 5.13, 5.14, 5.15 and 5.17"
I thought S@H V5.17 is now only for S@H- beta...

I let run now V5.17 WUs for main, but they are for beta?


Greetings!  (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)


EDIT:
When it's an error...
The new app is ready to cruch S@H V5.17- main- WUs?
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 14 Nov 2006, 04:17:56 pm
Hi Dirk,

I added 5.17 to the default app_info.xml files because otherwise people have lots of trouble when they run SETI Beta and normal SETI alongside. BOINC 5.4.x has a problem otherwise for them.

No matter what BOINC says, the app itself is based on 5.15 code, albeit heavily modified.

Also nochmal auf Deutsch - ist nur weil einige Leute ziemliche Probleme haben, die gleichzeitig SETI Beta und normales SETI crunchen. 5.4.x BOINC Versionen haben da jedenfalls ohne den 5.17er Eintrag grob 100e WUs mit Fehlern wieder rausgeworfen.

Egal was BOINC sagt, die App selber ist ne 5.15er, obwohl stark modifiziert.

HTH,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 14 Nov 2006, 04:25:25 pm
Hello Simon,

thanx for reply!

And many thanx to you and your guys for the good work! :)


So if in main S@H V5.17 will come, your app cannot crunch them...?
Or, I cannot download new WUs from the server?

So, if I let run only main, then I can modify my app_info.xml? ;)

Greetings! (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)


Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 14 Nov 2006, 04:38:59 pm
The problem with modifying your app_info.xml when you have already downloaded WUs BOINC thinks are for 5.17 is that it will not find the entry and error them out.

So before you edit your file, please run your cache dry with "no new tasks" "keine neuen aufgaben zulassen".

When you have crunched all WUs, then shut down BOINC and edit/remove the entry. Then start it up and it will be fine.

HTH,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Geek@Play on 14 Nov 2006, 05:25:15 pm
Congratulations to Simon and all who have contributed to the new release 2.0.  I have been using it since last night and it seems to be a bit faster so I am very pleased with it.  Note that I am running with a cache and so the outage last night at Seti was a non event for me but I'm still trying to return the results that I crunched.

Thanks to all of you for your hard work! ;D
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Urs Echternacht on 14 Nov 2006, 07:55:24 pm
The problem with modifying your app_info.xml when you have already downloaded WUs BOINC thinks are for 5.17 is that it will not find the entry and error them out.

So before you edit your file, please run your cache dry with "no new tasks" "keine neuen aufgaben zulassen".

When you have crunched all WUs, then shut down BOINC and edit/remove the entry. Then start it up and it will be fine.

HTH,
Simon.

That is one way. The hard way is a bit faster. You would have to edit app_info.xml, and client_state.xml and client_state_prev.xml But this is not recommended, because it can cause failure of BOINC if you do a mistake.

I have put the new Seti-app on some of my machines today and crunched a few wus. I see a 12%-13% increase on my Pentium M (that cuts another 1100secs from 61cr wu) compared to the previous version Rev.1.3.
Nice work and a big thank you to all you optimizers over here.

Two points of criticism:
a. rough output at stderr is bad readable (very minor issue)
b. where is the actual source code? (GPL)
 
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: BenHer on 15 Nov 2006, 01:28:04 am
Simon will post the code soon I imagine, it is allways our intention at KWSN.

However, if you want to pester someone about Open Source go ask Seti-USA...they are Allways happy to supply their source ;)
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 15 Nov 2006, 01:52:16 am
Simon will post the code soon I imagine, it is allways our intention at KWSN.
However, if you want to pester someone about Open Source go ask Seti-USA...they are Allways happy to supply their source ;)
Yes, you are right! ;)



Hello!

Can somebody please explain what does this all mean in the "<stderr_txt>"?
That the people know, what they have done... ;)
In this thread, or maybe better, somewhere on the homepage?


Greetings! (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)



<stderr_txt>
ChirpData--[TrigArray]:  17202812 (chosen)
 GetPeak--[hand_opt]:      2978 (chosen)
   f_sum--[original]:     27790 (chosen)
GetChiSq--[hoisted+abs(]:     36081 (chosen)
IPP FFT SSE2(64K)[original]:   5774918 (chosen)
Bench Time: 2.41 seconds
work_len=1048576
Optimized Windows SETI@Home Enhanced application

Optimizers: Ben Herndon, Josef Segur, Alex Kan, Simon Zadra
  Version: Windows SSE2 32-bit based on seti V5.15  'Chicken Good!'
      Rev: (R-2.0|QxB|FFT:IPP_SSE2|Ben-Joe|)
    CPUID: 'AMD K8 Athlon 64 (Newcastle)'
     cpus: 1 cores: 1 threads: 1   cache: L1=64K  L2=512K L3=0K
 features: mmx 3Dnow 3Dnow+ sse sse2 
    speed: 2211 MHz  -- read megs/sec: L1=12107, L2=5882, RAM=2512

Work Unit Info
True angle range:  7.932342

Spikes Pulses Triplets Gaussians Flops
    2       0        1          0       3227178515016
</stderr_txt>
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: BenHer on 15 Nov 2006, 02:49:53 am

     ChirpData--[TrigArray]:  17202812 (chosen)
        GetPeak--[hand_opt]:      2978 (chosen)
          f_sum--[original]:     27790 (chosen)  }- optimized functions - many versions of each [ <name> ] = fastest
   GetChiSq--[hoisted+abs(]:     36081 (chosen)
IPP FFT SSE2(64K)[original]:   5774918 (chosen)

Bench Time: 2.41 seconds  - Time to complete all testing

work_len=1048576   leftover debugging

Optimized Windows SETI@Home Enhanced application

      Rev: (R-2.0|QxB|FFT:IPP_SSE2|Ben-Joe|)  Revision - Compile option - FFT used - Who incorporated whos code

read megs/sec: L1=12107, L2=5882, RAM=2512 - How fast memory test could read  from L1, L2 caches and RAM in Megabytes per second
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: rgathright on 15 Nov 2006, 09:07:23 am
I ran the V2.0 all night, but unfortunately it slowed down my processing.  You can follow the link I posted above and easily see my WU results.

I do like V2.0 for one reason, I notice I had maybe 200,000 page faults per WU (comparing active results when I looked this morning).  When I started the old client up again, 25 minutes later I have 5,400,233 page faults for one WU.  Yet for some reason the old client is faster.   :(
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 15 Nov 2006, 09:13:17 am
Like Ben and I posted above, Reuben, the 2.0 release is not meant for Core 2 / Woodcrest / Merom systems (I did note that in the original announcement  ;)). For whatever reason (we're not 100% sure yet either), the 1.3 and 1.41 apps perform more quickly on those systems.

Go to the downloads section and grab the Intel Core 2 SSSE3 optimized version from the 1.3 section for now instead, please.

HTH,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Dirk Sadowski on 15 Nov 2006, 03:35:58 pm
BenHer thanx for explanation! :)



BTW. :)
AMD K7 Athlon 600:

True angle range: 0.421940

Windows optimized S@H Enhanced application by Crunch3r
Version info: Windows MMX V5.12 by Crunch3r
CPU time 97124

Optimizers: Ben Herndon, Josef Segur, Alex Kan, Simon Zadra
Version: Windows MMX 32-bit based on seti V5.15 'Chicken Good!'
Rev: (R-2.0|QxB|FFT:IPP_SSE2|Ben-Joe|)
CPU time 77395

5 h 29 min. faster!


Thanx to all optimizers!
Very good work!



Greetings!  (http://www.cheesebuerger.de/images/smilie/froehlich/a010.gif)



EDIT:

AMD K8 3200+

True angle range: 0.426463

SUSE Linux 10.1 -64Bit
Linux optimized S@H Enhanced application by Crunch3r
Version info: Linux x86_64 SSE2 64 bit V5.12 by Crunch3r
CPU time 12555

WinXP -32Bit
Optimizers: Ben Herndon, Josef Segur, Alex Kan, Simon Zadra
Version: Windows SSE2 32-bit based on seti V5.15 'Chicken Good!'
Rev: (R-2.0|QxB|FFT:IPP_SSE2|Ben-Joe|)
CPU time 11330

20 min. (+25 sec. ;)) faster!

Unbelievably !!!!!  :o
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 15 Nov 2006, 09:12:55 pm
Yeah, it's pretty quick...

Thank Ben Herndon, Joe Segur and Alex Kan ;)

Regards,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: MikeK on 16 Nov 2006, 03:36:02 am
Thank you guys.

The new app runs around 20% faster on my X2 4800 standard core speed.
With 1.3 ~ 12000 sec. with 2.0 ~ 9000 seconds.
Now will see with overclocked speed.

regards Mike

Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Kna on 16 Nov 2006, 05:16:14 pm
Another big thanks to all the guys which have allowed to this new optimised revision to exist   :)

But a strange behaviour with my Intel SSE only CPUs ?
Rev 2.0 seems to be on average 10% slower than rev 1.3 or Crunch3r SSE app, at similar AR.

K7 CPUs seems not to be affected. Due to different architecture, or DDR instead of PC133 SDRAM ? Mystery..

Does anybody notice the same thing ?

Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Simon on 16 Nov 2006, 05:21:12 pm
Not really, Franck -

though I have to say that I didn't test on SSE-only systems a lot (though other people did and reported good success). Thing is, they all used AMD CPUs, so I really lack results using P3s.

If the new apps don't work as quickly as the old ones, the logical step would be to downgrade ;)

Regards,
Simon.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Kna on 16 Nov 2006, 05:55:11 pm
Yes Simon, downgrade allready done, even if I don't have entensive measurements of this slowdown.
It was simply not nice to see   :)

Old PIII and Celeron desktops would be upgraded to P4 soon, the last survivors will be some dual PIII servers which suffice for their tasks for a few times.

I've downgraded to Crunch3r SSE, allready the fastest app for Intel PIII familly if I'm not mistaken, but Monday I put previous Chicken everywhere.
After some recent Crunch3r's post in Berkeley forum, his apps are banned from my rigs. But it's another story  ;D

Regards,
Franck.
Title: Great!
Post by: Furex on 16 Nov 2006, 10:55:44 pm
Great stuff!

Hope to see R2.0 for linux soon  ;)
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: ntorcida on 17 Nov 2006, 05:48:14 pm
I am having Amd Athlon 64 3000+ (venice core). Should I use SSE or SSE2 Generic? Curently I am using SSE2 Generic....
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Furex on 17 Nov 2006, 07:16:16 pm
Apps named "Intel" only work on Intel CPUs, so SSE2 generic is indeed the correct choice for AMD64 processors.
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: Arnulf on 19 Nov 2006, 03:12:21 am
Hi!

I am very pleased with the new 2.0 'Ni!' app!

Checking and comparing the speed against other computers have revealed that the 2.0 'Ni!' app running on my Opteron 265's at 1,8GHz are faster than the 1,3 app running on an X2 4800+ at 2,4GHz.
Link: http://setiathome.berkeley.edu/workunit.php?wuid=99273403

The RAC of my computer have climbed through 1400 and are still rising.  ;D

Arnulf
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: MikeK on 19 Nov 2006, 05:10:54 am
Hi

My sons AMD 64 3500+ @ 2.46 Mhz beats most of the core2 duo with 2.0.
He is really amazed.

Mike
Title: Re: New Apps based on Code Revision 2.0 'Ni!' have been released
Post by: msattler on 24 Nov 2006, 07:39:36 am
The RAC of my FX60 rig had been averaging around 950 running on your first optimized app.  Since I switched to your latest release, it has slowly risen to 1205 as of today.
Nice boost!  Nice work!  Thanx to all involved.