Message boards :
Number crunching :
Optimized BOINC 4.72 core (Windows/Linux)
Message board moderation
Author | Message |
---|---|
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
I built 3 versions (mmx,sse,sse2) w/o boincmgr (please use distributed ones) with Intel C++ compiler. If you like, put them on your homepage. Distribute these for free w/o notice. Use your at own risk. These packages contain boinc.exe and some .dll files. Make a full backup copy of your boinc directory beforehand just in case and unzip all contents. I guess boinc.dll isn't necessary, but it won't harm if it exists. After that, it's like 4.45. boinc-472-mmx.zip (mmx processors) boinc-472-sse.zip (sse processors; P3, AthlonXP..) boinc-472-sse2.zip (sse2 processors; P4, AMD64) No sse3 version because it's (almost) the same as sse2. Happy crunching! Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Hi Thanks again Tetsui. Will give it a try. greetz from Germany Mike With each crime and every kindness we birth our future. |
Spectrum Send message Joined: 14 Jun 99 Posts: 468 Credit: 53,129,336 RAC: 0 |
Hi Maverick, thanks for the effort. |
Mike Send message Joined: 17 Feb 01 Posts: 34258 Credit: 79,922,639 RAC: 80 |
Hi Nice performance boost on Athlon XP 2800+ From 1728 3820 with 4.45 to 1830 4289 MIPS. Great work. greetz Mike With each crime and every kindness we birth our future. |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
|
pindakoe Send message Joined: 4 Jun 00 Posts: 60 Credit: 345,676 RAC: 0 |
Thanks Maverick. On a 1.5 GHz pentium-M benchmark went from 1615/3289 to 1622/3424. Not spectacular, but a few % here and a few % there all adds up. Let me know when you have an optimised boinc out for Linux. Am, currently running with Ned's 4.19.3a version which works well, but is somewhat behind the times. |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
Thanks Maverick. Let me know when you have an optimised boinc out for Linux. Am, currently running with Ned's 4.19 version which works well, but is somewhat behind the times. okay....but I'm not sure I can build boincmgr also. I'll try, but I may ask you to use distributed one (4.45 or something) for boincmgr. It won't taint the performance. Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
pindakoe Send message Joined: 4 Jun 00 Posts: 60 Credit: 345,676 RAC: 0 |
okay....but I'm not sure I can build boincmgr also. I'll try, but I may ask you to use distributed one (4.45 or something) for boincmgr. It won't taint the performance. For me there is no need for boincmgr on linux -- I run it as a daemon and seldom look at it. Under Windows however I do need boincmgr as running Seti or Einstein completely stalls a synchronisation task of Outlook (reported before in the forums and apparently because seti & Einstein do not run at true idle priority; CPDN does not have this issue). |
Ingleside Send message Joined: 4 Feb 03 Posts: 1546 Credit: 15,832,022 RAC: 13 |
SETI@Home/BOINC v4.18 runs on "idle" (priority 1), Einstein@home being built before the change in BOINC-api does not run on idle but low-low (priority 2). But, atleast some of the optimized seti-applications does not run on idle priority but instead on low-low, indicating they've been built with an outdated BOINC-api... Anyway, a problem with synchronication of offline? content in Outlook is really a bug in Outlook, since the thread shouldn't really be running at idle priority. |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
Here's 4.72 optimized Linux clients (statically linked) boinc-4.72-mmx.tar.gz(for mmx) boinc-4.72-sse.tar.gz(for sse processors; P3, AthlonXP) boinc-4.72-sse2.tar.gz(for sse2 processors; P4, AMD64) Sorry, I cannot build boincmgr. Please use one of distributed ones. Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
pindakoe Send message Joined: 4 Jun 00 Posts: 60 Credit: 345,676 RAC: 0 |
SETI@Home/BOINC v4.18 runs on "idle" (priority 1), Einstein@home being built before the change in BOINC-api does not run on idle but low-low (priority 2). The problem appears while being on-line through the office LAN -- this is 100 MBps and sitting within view of the datacenter, so latency and network are OK. Outlook send/receives normally, until after an hour or so (never timed it properly) it does a non-standard sync, which stalls (but does not abort) when seti (Maverick's YAOSCW-B-r8.1 client) or the stock Einstein client are running. Under Einstein something may happen (never waited long enough), but under SETI all traffic is forzen. This does not happen when CPDN is running, although the syncing goes slower than when nothing is running (but not unbearably slow). I tend to use eitehr boincmgr (or boincview, but it is overkill for what I want) to suspend and restart boinc. |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
Thanks. Now I'm tackling non-Intel processor support of Intel C++ compiler and IPP. First Linux, then Windows. Hi Tetsuji, I volunteer as beta tester for your patched client. :o) I'd run it on a Athlon64X2 4400+. Currently I'm using Metod's client. Thanks for that, by the way! Regards Hans P.S: I'd like to test the patched version agianst the unpatched one, so I'll need both |
pindakoe Send message Joined: 4 Jun 00 Posts: 60 Credit: 345,676 RAC: 0 |
Just d/loaded and tried it on my AMD 2800XP (Mandriva 2005LE). It segfaults while running the benchmark. Output is as follows: ./boinc -run_cpu_benchmarks 2005-10-02 21:15:06 [---] Starting BOINC client version 4.72 for i686-pc-linux-gnu 2005-10-02 21:15:06 [---] libcurl/7.14.1 OpenSSL/0.9.7a zlib/1.1.4 2005-10-02 21:15:06 [---] Data directory: /home/keesjeen/Desktop/Downloads SIGSEGV: segmentation violationStack trace (17 frames): [0x808270f] [0x818e0b8] /lib/libc.so.6[0xb7f2ad34] /lib/libc.so.6[0xb7f2b1f1] /lib/libc.so.6(malloc+0x29)[0xb7f2a1a9] /lib/libc.so.6[0xb7f1ad0b] /lib/libc.so.6(fopen+0x1f)[0xb7f1ad9f] /lib/libnss_files.so.2[0xb7fec0e6] /lib/libnss_files.so.2(_nss_files_gethostbyname_r+0x34)[0xb7fec4f4] [0x81b7912] [0x81b776b] [0x8084d02] [0x8084b2c] [0x809ad64] [0x8048231] [0x8187de7] [0x8048111] Exiting... Not good. Could it be that this is conflicting with existing libraries? On a more positive note: the Windows mmx client works, but gives no speed-up versus the mmx version of 4.45 when bencmarked on an old Celeron II 400 MHz. |
Crunch3r Send message Joined: 15 Apr 99 Posts: 1546 Credit: 3,438,823 RAC: 0 |
Its fails because of the static linking. Thats why it will not be able to resolve hostnames on some systems with different versions of the libs it is compiled against. Dynamic linking and excluding standard libs should fix the issue. Join BOINC United now! |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
Thanks. Now I'm tackling non-Intel processor support of Intel C++ compiler and IPP. First Linux, then Windows. Hans, I see you're running Metod's SETI client too. Have you tried my AMD64 client on your X2 as it _should_ be faster on an AMD processor. Regards, Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Hans Dorn Send message Joined: 3 Apr 99 Posts: 2262 Credit: 26,448,570 RAC: 0 |
No, but I'll give it a try tomorrow. I'll post my results in a new thread. Regards Hans. |
Purple Rabbit Send message Joined: 31 Aug 99 Posts: 49 Credit: 5,820,832 RAC: 3 |
I see an interesting side effect with Maverick's v4.72 for Windows. I can't disable work requests with Maverick's optimized version for Windows. Otherwise it works just fine. I can suspend work so the commands are getting through. I didn't try the more destructive commands :-) I put v4.45 back and I could disable work requests. Reinstall Maverick's v4.72 and I couldn't disable work requests. I have been running Metod's client for v4.72 on Linux for almost two months with no problems. Is this an artifact of the optimization or am I missing something? I presume that since I can disable work requests with Metod's Linux v4.72 that the BOINC version level isn't the problem? Maybe it's some Windows thingy? |
Tetsuji Maverick Rai Send message Joined: 25 Apr 99 Posts: 518 Credit: 90,863 RAC: 0 |
okay, i made a dynamic linked mmx version for testing. boinc-4.72-mmx-r2.tar.gz If it's ok I'll make other optimized ones. Now it's 5:44am here. Luckiest in the world. WMD = Weapon of Mass Distraction. Click this table. |
Chilean Send message Joined: 6 Apr 03 Posts: 498 Credit: 3,200,504 RAC: 0 |
|
chrisjohnston Send message Joined: 31 Aug 99 Posts: 385 Credit: 91,410 RAC: 0 |
My X2 went from 2145/4396 to 2878/2143.. Is that normal? - cJ |
©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.