Message boards :
Number crunching :
optimized Enhanced 5.12 app.
Message board moderation
Previous · 1 . . . 6 · 7 · 8 · 9 · 10 · 11 · 12 · Next
Author | Message |
---|---|
Speedy67 & Friends Send message Joined: 14 Jul 99 Posts: 335 Credit: 1,178,138 RAC: 0 |
After a few days of running Crunch3r's 5.12 app on my Linux Ubuntu 5.10 box (Athlon64 3000+ Venice, SSE3), and crunching a few workunits with the same angle range with both the SSE3 and later the SSE2 app, I found that on average a 0.5459 AR workunit needs some 17200 seconds with the SSE3 app, and only 15900 with the SSE2 app. I figure someone will be interested! I might try it myself. :) People who use optimized clients are interested in every second that can be gained.. so in this example, 1300 seconds on a workunit is a huge gain! It might even be possible to make some sort of reference list like this one. Greetings, Sander |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
After a few days of running Crunch3r's 5.12 app on my Linux Ubuntu 5.10 box (Athlon64 3000+ Venice, SSE3), and crunching a few workunits with the same angle range with both the SSE3 and later the SSE2 app, I found that on average a 0.5459 AR workunit needs some 17200 seconds with the SSE3 app, and only 15900 with the SSE2 app. Hi sander, I know of your reference wu list, but it'll be hard to do one like yours with seti enhanced. You need to do testing on the most common AR's this is hard to do. You can write me an email and i let you know what to do ;) Join BOINC United now! |
archae86 Send message Joined: 31 Aug 99 Posts: 909 Credit: 1,582,816 RAC: 0 |
double post error by me--please delete this one if you will. |
archae86 Send message Joined: 31 Aug 99 Posts: 909 Credit: 1,582,816 RAC: 0 |
I figure someone will be interested! I might try it myself. :)I've come to believe recently that my speedup using Crunch3r's 5.12 over the distributed 5.15 is strongly dependent on Angle Range, and sometimes much better than the 20% improvement I guessed very early on. If you don't mind losing some machine time to non-credit computing, then a method posted by Stonelord over on Einstein works just fine to allow recomputation of the exact same Result by more than one program on the same host. Stonelord's captive WU test method Using this method, I just now did a 1.975 AR WU, getting 3710 seconds on crunch3r 5.12, and 5096 seconds for distributed 5.15, for an execution time ratio of .728. The really impressive case is one for which my evidence is less perfect, but I find convincing. I had six results from 07ja99aa.23932.16818.231240.3 to run. All had very, very small angle ranges-- 0.0338. Most of them ran with Crunch3r 5.12 with CPU times tightly clustered around 22500 seconds. The one which ran distributed 5.15 entirely took 51308 seconds. Unless that result was a rogue, the implied execution time ratio is .4385. One additional result of this group ran distributed for the first 5% of reported progress. Rate of advance monitoring for this one substantiated that for this case crunch3r was taking less than half the time. It would be really good to get concrete data from good comparison cases for more angle range cases, especially near the tight central cluster. My box is a Gallatin (some call it P4 Extreme Edition) 3.2 GHz CPU, running hyperthreaded, with an Einstein job on the other side from SETI. I force this with the priority project function in Trux's tx36 client. My hat is off (again) to crunch3r, and I regret contributing earlier to what may have been a significant underestimate of the advantage your 5.12 offers in speed. It also has been running clean. |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
I figure someone will be interested! I might try it myself. :)I've come to believe recently that my speedup using Crunch3r's 5.12 over the distributed 5.15 is strongly dependent on Angle Range, and sometimes much better than the 20% improvement I guessed very early on. Hi Archae86, It really depends on the AR how fast the optimized app runs.( It's allways been that way but didn't matter that much on the normal s@h app. just a few minutes difference to complete the wus) As you've stated yourself there's at least a speed gain of 20% up to 45% over the stock app. depending on the AR. I've got an e-mail from Speedy67 (Sander) and we'll make a reference list on the optimized app compared to the stock app on the different ARs. Hopefully it'll solve some quetestions on how fast the different versions of the optimized apps. are on different ARs. P.S. Damn can't type any more ... must have answered more than 50 mails today ... EDIT P.P.S I've also relased a "stock application" today and notified Eric if he could test that one on the beta project to see if all the c++ exceptions are compiler related due that i know that i didn't seen those on my apps. Hopefully he cosiders it and i wouldn't mind to compile a stock app without cpu specific optimizations to be released to public. But that's not a choice i'll have to make ;) /EDIT Join BOINC United now! |
Saimek Send message Joined: 25 Jan 00 Posts: 121 Credit: 454,423 RAC: 0 |
I'm curious about this SSE2 over SSE3 optimalisation..... if it's true that SSE2 is faster evenave on SSE3 supporting boxes? :P someone who has the possibility to test it on a Venice core and send the results... i would appreciate ;) greets |
Speedy67 & Friends Send message Joined: 14 Jul 99 Posts: 335 Credit: 1,178,138 RAC: 0 |
|
cdr100560 Send message Joined: 12 May 06 Posts: 681 Credit: 65,502 RAC: 0 |
I want to personally thank Crunch3r for providing his optimized app for use on my platform (x86). It's amazing how efficient code can improve existing hardware. I also want to thank all of the posters who answered specific questions regarding the use of Crunch3rs application. I'm no programmer, but it only took two revisions of the app_info file to get things going. I have successfully crunched my first WU and noticed an immediate improvement along the lines of the 20% that have been stated. I can't provide detailed analysis, but I can see from the onset that it's working. It took a little patience to filter through the various threads to find what I was looking for (thanks Jim-R!) but everything is here. I suspect that a lot of confusion is based on the average users lack of general knowledge (myself included!) in the application of supplied info, as there are many posts in different threads just on this subject alone. Again, thank you for the patience required in getting this puppy to fly! Regards, Chris |
Ralf02061973 Send message Joined: 24 Jul 00 Posts: 54 Credit: 9,983,656 RAC: 8 |
here on my pentium-d-820@3600mhz the 5.12ss3 is 3 minutes faster than 5.12ss2 sse2 ~10100sec sse3~ 9900sec greetings ralf edit: wu´s are 5.15 Boinc runs here on: Intel i7-3770K + IntelHD4000 Android-Stick-ARM-Cotex-A17 Sony-Z5C-ARM-Cortex-A53/A57 Nvidia GT-630 / Nvidia GTX-750Ti |
n7rfa Send message Joined: 13 Apr 04 Posts: 370 Credit: 9,058,599 RAC: 0 |
here on my pentium-d-820@3600mhz The WUs are Enhanced. You're just reporting them as 5.15 instead of the REAL version of 5.12. As of my last check, yesterday, Crunch3r's Optimized Apps are only up to 5.12. |
Ralf02061973 Send message Joined: 24 Jul 00 Posts: 54 Credit: 9,983,656 RAC: 8 |
now an update the last 15 setiathome_enhanced-5.19 wu's with optimiced app 5.12sse3 crunchtime 1600sec - 2550sec ar = 1.65xx host = Pentium-D-820 @ 3600MHz tomorrow i will try the optimiced 5.12sse2 with the 5.19 wu's to see if it is also slower than sse3 Boinc runs here on: Intel i7-3770K + IntelHD4000 Android-Stick-ARM-Cotex-A17 Sony-Z5C-ARM-Cortex-A53/A57 Nvidia GT-630 / Nvidia GTX-750Ti |
Geek@Play Send message Joined: 31 Jul 01 Posts: 2467 Credit: 86,146,931 RAC: 0 |
now an update If you are using Crunch3r's optimized app's (version 5.12) then you should be reporting them as version 5.12 and NOT 5.19. Reporting them as being crunched by an incorrect version messes up the master data base. Boinc....Boinc....Boinc....Boinc.... |
Dances with Werewolves Send message Joined: 8 Nov 03 Posts: 489 Credit: 340,188 RAC: 0 |
@crunch3r stderr.txt: SETI@Home Informational message -9 result_overflow NOTE: The number of results detected exceeds the storage space allocated. Is this a problem? I'm seeing it alot. |
[HWU] GHz & CO. - BOINC.Italy Send message Joined: 1 Jul 02 Posts: 139 Credit: 1,466,611 RAC: 0 |
now an update The problem is that the app_info.xml from Crunch3r's site permit to use the optimized client with work for 5.19 client, and BOINC mark and report all WU with version 5.19 of the seti client. http://setiathome.berkeley.edu/result.php?resultid=333555069 application version 5.19 Why this? Maybe a BOINC bug? GHz BOINC.Italy |
Geek@Play Send message Joined: 31 Jul 01 Posts: 2467 Credit: 86,146,931 RAC: 0 |
The problem is that the app_info.xml from Crunch3r's site permit to use the optimized client with work for 5.19 client, and BOINC mark and report all WU with version 5.19 of the seti client. The problem is with the app_info.xml that is supplied by Crunch3r. It allows for versions up to and including 5.19 but since the section that references version 5.19 is the highest version on his list that is the section that is used by Boinc. To correct this...... 1) Set Boinc/Seti to "no new work" and crunch down your cache. Failure to do this will result in the loss of your entire cache. 2) When the cache is depleted, stop Boinc. Replace the information in the app_info.xml file with the following. <app_info> <app> <name>setiathome_enhanced</name> </app> <file_info> <name>setiathome_5.12_windows_intelx86.exe</name> <executable/> </file_info> <app_version> <app_name>setiathome_enhanced</app_name> <version_num>512</version_num> <file_ref> <file_name>setiathome_5.12_windows_intelx86.exe</file_name> <main_program/> </file_ref> </app_version> </app_info> 3) Restart Boinc and set to "allow new work" New work should be downloaded, crunched and reported as version 5.12. Boinc....Boinc....Boinc....Boinc.... |
AlanSerl Send message Joined: 14 May 99 Posts: 1 Credit: 13,845,655 RAC: 0 |
The problem is that the app_info.xml from Crunch3r's site permit to use the optimized client with work for 5.19 client, and BOINC mark and report all WU with version 5.19 of the seti client. The change in the app_info.xml worked exept one thing missing here is that you will need to 1) Stop the service 2) Unzip Crunch3r's setiathome_4.11_windows_intelx86.* files and rename them as setiathome_5.12_windows_intelx86.* 3) Drop the renamed files into C:\\Program Files\\BOINC\\projects\\setiathome.berkeley.edu 4) Restart the service Without these files being renamed, I was finally getting new WU's only to get a application not found error. Many Thanks Alan |
n7rfa Send message Joined: 13 Apr 04 Posts: 370 Credit: 9,058,599 RAC: 0 |
The problem is that the app_info.xml from Crunch3r's site permit to use the optimized client with work for 5.19 client, and BOINC mark and report all WU with version 5.19 of the seti client. Excuse me? Rename the 4.11 executable and rename it to 5.12? The 4.11 executable doesn't handle the Enhanced WUs. |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
Hello, It's not quite easy for me to write this, to honest to you all i'm writing this now with a tear running from my eye. It's been a fantastic time here for quite a while, since things have changed for the bad. That's why i'm taking the consequences and have to say god bye to you all. Take care of yourself and keep on crunching. Good by SETI. Join BOINC United now! |
cdr100560 Send message Joined: 12 May 06 Posts: 681 Credit: 65,502 RAC: 0 |
Hello, I don't know you personally, only professionally from your posts. I am not happy to see you leave. Theres all kinds of solutions to all kinds of problems in these threads. There might be one here for you also. |
Dances with Werewolves Send message Joined: 8 Nov 03 Posts: 489 Credit: 340,188 RAC: 0 |
Hello, Truely, a sorry day for us all and you will be missed. Scott |
©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.