Message boards :
Number crunching :
YAOSCW r8.1 (optimized client for Windows)
Message board moderation
Previous · 1 · 2 · 3 · 4 · 5 . . . 10 · Next
Author | Message |
---|---|
[ue] Toni_V Send message Joined: 6 Apr 03 Posts: 52 Credit: 141,788 RAC: 0 |
So.. I crunched 6 packets with IPP 8.1 N-version with my p4-2.66@2.85 cpu. Times: 5344 5419 5429 5411 5509 5532 (avg)5441 I thought the older fftv-version was faster for me, so I switched back. fftv times: 4881 5567 4925 5118 5310 5093 (avg)5149 It's about same speed what I saw earlier - the fftv is couple hundred seconds faster that IPP 7 or 8.1. Now I have 10 packets behind with the fftv. Only two of those packets went over 1:30 (5400sec). Five out of six packets took longer than 1:30 with IPP 8.1 version. Fftv is faster for me - not sure why, but I'm crunching with it. |
Arm Send message Joined: 12 Sep 03 Posts: 308 Credit: 15,584,777 RAC: 0 |
So.. I crunched 6 packets with IPP 8.1 N-version with my p4-2.66@2.85 cpu. On a P4 Willamette 1.7 on a i850E chipset (RDRAM) ver.8.1-N is slower than the v.7-N. So Im back on v.7. |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
It's strange... but you should use what you want :) On my boxes (P4 Northwood/Prescott) YAOSCW's are clearly faster than fftw (5-10%). I don't (cannot) guarantee on other boxes. I'm testing only on my machines. BTW I had tried MKL on Linux also before I built YAOSCW, and found its speed was between fftw and IPP. Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
B-Roy Send message Joined: 4 May 03 Posts: 220 Credit: 260,955 RAC: 1 |
So.. I crunched 6 packets with IPP 8.1 N-version with my p4-2.66@2.85 cpu. same thing for me on an Athlon64 3400+ |
Nyx Send message Joined: 11 Nov 01 Posts: 3 Credit: 774,335 RAC: 0 |
heh, patch sources are always appreciated =D Made my own version using the current seti source, your patches, and played around with the code to get the gui working (i like teh pretty pictures) =D gets nice times =D YAOSCW-P-r8.1.zip - QxP = P4 SSE3 close enough, Tetsuji uses linux, /Qx* are windows switches, linux is -x* For those interested (or bored),looking at intels documentation: 0 - optimisations off (it's /Qd on win) K – P3 & Amd XP procs (mmx/sse) W – P4 & Amd 64 and Opteron procs (mmx/sse/sse2) the rest have code to check for 'non-compatable' cpu's at runtime, which means they are fussy. B – Pentium M N – P4 (adds some extra intel only stuff to the W switch) P – P4 with SSE3 (prescott & EM64T) (mmx/sse/sse2/sse3) you can play with teh switches too: x* makes a targeted build ax* does the same as x* but adds generic ia32 code so that it'll run (without the optimisations) on most x86 procs, obviously this makes the app larger slightly, this can be chained so you can do something like -axKW, which should create a semi-generic (and largeish) client that'll run on almost anything, but still have the optimisations of teh K & W options |
SURVEYOR Send message Joined: 19 Oct 02 Posts: 375 Credit: 608,422 RAC: 0 |
I’m attached to BOINC alpha test some wu with setiathome 4.17 and some with 4.18. Using YAOSCW-N-r8.1.zip for setiathome when trying to download new wu’s I get the following errors 6/17/2005 11:24:58 PM|SETI@home|Requesting 468459 seconds of work, returning 0 results 6/17/2005 11:25:00 PM|SETI@home|Scheduler request to http://setiboinc.ssl.berkeley.edu/sah_cgi/cgi succeeded 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.118 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.117 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.111 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.105 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.119 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.104 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.123 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.122 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.88 in sched reply 6/17/2005 11:25:00 PM|SETI@home|WU refers to nonexistent app_version: setiathome 418 6/17/2005 11:25:00 PM|SETI@home|Can't link workunit 29fe04aa.5344.5106.686074.109 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.118 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.118_0 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.117 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.117_0 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.111 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.111_2 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.105 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.105_3 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.119 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.119_0 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.104 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.104_2 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.123 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.123_0 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.122 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.122_0 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.88 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.88_1 in sched reply 6/17/2005 11:25:00 PM|SETI@home|link_result: nonexistent WU 29fe04aa.5344.5106.686074.109 6/17/2005 11:25:00 PM|SETI@home|Can't link result 29fe04aa.5344.5106.686074.109_1 in sched reply 6/17/2005 11:25:00 PM|SETI@home|No work from project 6/17/2005 11:25:01 PM|SETI@home|Deferring communication with project for 58 seconds 6/17/2005 11:25:01 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.118 6/17/2005 11:25:01 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.117 6/17/2005 11:25:09 PM|SETI@home|Finished download of 29fe04aa.5344.5106.686074.117 6/17/2005 11:25:09 PM|SETI@home|Throughput 47765 bytes/sec 6/17/2005 11:25:09 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.111 6/17/2005 11:25:10 PM|SETI@home|Finished download of 29fe04aa.5344.5106.686074.118 6/17/2005 11:25:10 PM|SETI@home|Throughput 40642 bytes/sec 6/17/2005 11:25:10 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.105 6/17/2005 11:25:17 PM|SETI@home|Finished download of 29fe04aa.5344.5106.686074.111 6/17/2005 11:25:17 PM|SETI@home|Throughput 49819 bytes/sec 6/17/2005 11:25:17 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.119 6/17/2005 11:25:18 PM|SETI@home|Finished download of 29fe04aa.5344.5106.686074.105 6/17/2005 11:25:18 PM|SETI@home|Throughput 46706 bytes/sec 6/17/2005 11:25:18 PM|SETI@home|Started download of 29fe04aa.5344.5106.686074.104 6/17/2005 11:25:25 PM|SETI@home|Finished download of 29fe04aa.5344.5106.686074.119 6/17/2005 11:25:25 PM|SETI@home|Throughput 47963 bytes/sec It work fine until I attached to SETI@Home Beta on the 15 Jun which uses setiathome_4.00_windows_intelx86.exe It looks like I received wu but none made to my computer, but the web page [my account] says I have received them. I looks like boinc is looking to link the seti wu to 4.18 not to 4.11. Do I need to add 4.18 to the app_info.xml file, I not sure about the format. Resulting time increase from public version 4.09 to 4.11 r8 is as follows: P4 3.4 ht two wu’s at the same time 3:34 to 1:58 = 1hr 34 min faster P4 2.53 no ht 3:46 to 1:35 = 2 hr 11 min faster Fred BOINC Alpha, BOINC Beta, LHC Alpha, Einstein Alpha |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
BENCHMARK TEST WARNINGS to everyone I suspect many people should have a wrong thought about benchmark test. If you are satisfied with your results, it's completely up to you which you will use and it's none of my business, but for accuracy, I recommend the following way. Especially I have tested only on my boxes (p4 Northwood/Prescott) and I put what I thins is the best, but many people want other results. (but sorry, my benchmark is not accurate and won't publish) On other boxes, what I think is the best isn't necesarily the best. Sorry I cannot confirmed it by myself....so I hope correct benchmark on several machines will help others. Correct method of benchmark test. Will you use referenceunits.zip for benchamrk? This is the standard workunit provided by the official SETI@home project and I put on my page. Rename reference_work_unit.sah to work_unit.sah, put the cruncher and this work_unit.sah in a folder, and crunch it with the target client, and check the time (wu_cpu_time field) in init_data.xml which is created when crunching finishes, and compare the crunching time. You must not interrupt this crunching until it finishes in order to obtain the correct crunching time. It's the standard "crunch time" everybody publishes. For ex Metod's page. Everybody uses this reference work unit for benchmarks and it's the standard. Otherwise your benchmark cannot be compared with other benchmarks. If you do several benchmark tests on the same machine, someone (probably Metod) will be willingly publich on a homepage and everyone will want to see it. Don't forget to mention what processor/OS(even on Linux this work unit is used) you used. regards, PS: but to tell the truth, I myself don't have so much interest in these benchmark on other machines.....use what you think is the best :) Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
Nyx Send message Joined: 11 Nov 01 Posts: 3 Credit: 774,335 RAC: 0 |
Make sure theres nothing but the client and the reference units in the folder you create. p.s.: if you are running a version with a gui (the pretty graphs and stuff) with the benchmark (such as teh vanilla to calibrate) and don't want the gui slowing it down, then append -nogui to the command, so you'll end up with something like: setiathome_4.11_windows_intelx86.exe -nogui on windows and I'm guessing something fairly similar on *nix If you do tho, be warned that you won't know whats going on or be able to stop it until it finishes unless you look at the process itself (eg taskman in windows). There's prob a command somewhere to get it to run in the old cli mode instead of automatically going into standalone mode, but I'm too busy to look for it at the moment. |
JohnB175 Send message Joined: 15 Oct 03 Posts: 124 Credit: 321,769 RAC: 0 |
Do you have any for linux using IPP4.1 and new Intel C++ compiler 9.0? I currently have yours but it's using IPP4.1 and Intel C++ compiler 8.1. |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
Do you have any for linux using IPP4.1 and new Intel C++ compiler 9.0? I currently have yours but it's using IPP4.1 and Intel C++ compiler 8.1. Here is my own P version. (statically linked) No other versions. This is the fastest on my P4 Prescott. YAOSCL-P-r8.1.tar.gz Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
JohnB175 Send message Joined: 15 Oct 03 Posts: 124 Credit: 321,769 RAC: 0 |
Ah, well I would need the "K" version since I have a P3-800EB. |
michael37 Send message Joined: 23 Jul 99 Posts: 311 Credit: 6,955,447 RAC: 0 |
|
Juan the cruncher Send message Joined: 10 Oct 99 Posts: 12 Credit: 807,165 RAC: 0 |
On my Athlon XP fftv-version client seems to be the fastest solution. (YAOSCW r8.1 and YAOSCW r7.0 give almost the same score and are about 5% slower than fftv) On my P4 the fastest client is YAOSCW r8.1 - giving another 10% or so improvement. Overall running on my 3 CPU optimized clients I went from 500 credit/day close to 1k/day. Thanks a lot Tetsuji for your great work.... |
Metod, S56RKO Send message Joined: 27 Sep 02 Posts: 309 Credit: 113,221,277 RAC: 9 |
Any idea what's the time difference between Linux ICC9.0 and ICC8.1 versions? Metod -- you are an expert in this kind of stuff. I don't have slightest idea. Intel doesn't want me to get the new 9.0 (non-commercial licence). Until they start giving out non-commercial licences, I won't be able to use it. Metod ... |
trux Send message Joined: 6 Feb 01 Posts: 344 Credit: 1,127,051 RAC: 0 |
I just posted about new very fast optimized BOINC clients for several CPU platforms here http://setiweb.ssl.berkeley.edu/forum_thread.php?id=15960 trux Czech National Team |
Shaktai Send message Joined: 16 Jun 99 Posts: 211 Credit: 259,752 RAC: 0 |
Well, I've completed my first two SETI units on a Pentium D 840 3.2 ghz dual core (without HT). http://setiathome.berkeley.edu/results.php?hostid=1100185 (posted so you can monitor future results if your interested) This is using Tetsuji Maverick Rai's optimized SETI app and the optimized BOINC client. Very impressive. (thanks Tesuji) The standard BOINC benchmark 1596 whet and 2614 dry Optimized BOINC client 2036 whet and 4454 dry. The average time for the first two work units was 3869.78 or 64.5 minutes. Since it is dual core, it is running two units at one time or just over 32 minutes per unit. Of course, two units is hardly enough to get a really good feel for it so the link above will show you the continued progress. Mind you I am also running Einstein & Predictor on the same box, so future results will be farther apart. I just wanted to see what it could do out of the gate. Team MacNN - The best Macintosh team ever. |
jrmm22 Send message Joined: 30 Jan 04 Posts: 353 Credit: 24,536,157 RAC: 0 |
Any idea what's the time difference between Linux ICC9.0 and ICC8.1 versions? Metod -- you are an expert in this kind of stuff. I've got the link for the 'Free evaluation trial' if you're interested. AFAIK the free non-commercial license will be available after August... I'm unsure whether you can distribute compiled binaries or not (I think not, I'll check the EULA and tell you later). |
David@home Send message Joined: 16 Jan 03 Posts: 755 Credit: 5,040,916 RAC: 28 |
As setiathome application version 4.18 has now been relased does anybody know 1) If the Berkeley client is optimized in anyway? 2) Will the app_info.xml still work OK with 4.18? 3) Will there be an optimized version of 4.18 from the wizard TMR? |
Heffed Send message Joined: 19 Mar 02 Posts: 1856 Credit: 40,736 RAC: 0 |
2) Will the app_info.xml still work OK with 4.18? The app_info.xml will basically tell BOINC to ignore the new version, and continue to run whatever optimized version you may be running. 3) Will there be an optimized version of 4.18 from the wizard TMR? Probably. |
David@home Send message Joined: 16 Jan 03 Posts: 755 Credit: 5,040,916 RAC: 28 |
2) Will the app_info.xml still work OK with 4.18? Is that a good idea or not? I have put my CC on no new work just in case it is bad for the science. 3) Will there be an optimized version of 4.18 from the wizard TMR? :-) |
©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.