Message boards :
Number crunching :
New Faster BOINC 4.13 client for LINUX available now
Message board moderation
Author | Message |
---|---|
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
Hi, I have new significantly faster optimised boinc clients available for linux. These were compiled with gcc-3.4.2 and give 10% improvements on Whetstone and >40% improvements on Dhrystone scores compared to optimised clients compiled with gcc-3.3.3 and approximately double the benchmark scores from the standard downloaded client from Berkeley. They are available from the downloads section of the link in my signiture. These latest boinc clients now actually produce benchmark scores slightly higher than the same hardware running Windows :) Further details on these optimised boinc clients is available here: http://forums.pcper.com/showthread.php?t=364551 Regards, Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Sir Ulli Send message Joined: 21 Oct 99 Posts: 2246 Credit: 6,136,250 RAC: 0 |
Thanks Ned for your Great Work here. Greetings from Germany NRW Ulli S@h Berkeley's Staff Friends Club m7 © |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
No problem Sir Ulli. Let me know if you have any problems with them - you know where to find me :) Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Joe Rhodes Send message Joined: 4 Oct 02 Posts: 24 Credit: 1,288,238 RAC: 0 |
Ned, I downloaded your optimized seti for linux and it improved my integer benchmark by about 70%. I'm using it on 3 linux boxes. Thanks a bunch. |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
> Ned, > I downloaded your optimized seti for linux and it improved my integer > benchmark by about 70%. I'm using it on 3 linux boxes. Thanks a bunch. > > Glad you like it. It's great us Linux users can now finally compete on a level playing field with our Windows cousins :) Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
THESPEEKER Send message Joined: 3 Apr 99 Posts: 168 Credit: 48,990 RAC: 0 |
Hi Ned, does this effect your credit? will it be less or more? |
xi3piscium Send message Joined: 17 Aug 99 Posts: 287 Credit: 26,674 RAC: 0 |
I'm not very linux savy, can someone kindly tell me or give me the command to create the link in usr/lib to run the optimized client? thanks in advance, Joe |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
> I'm not very linux savy, can someone kindly tell me > or give me the command to create the link in usr/lib > to run the optimized client? > > thanks in advance, > > Joe > Simple method is to just drag and drop the file in konquorer and select "make link". From the shell, use the ln command (See 'info ln' for more info). Here's the shell command run as root: ln -s /usr/lib/libstdc++.so.6.0.3 /usr/lib/libstdc++.so.6 Alternatively, just renaming the file to libstdc++.so.6 would work but this isn't deemed particularly good practice as it can cause problems if the library file is later updated to a newer version. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
> Hi Ned, > does this effect your credit? > will it be less or more? > Yes, it will affect your requested credit, increasing it more in line with what Windows users currently request. Thus, your average awarded credit probably will increase as a result. By making it more in line with what windows users request, there's more change of the credit you're awarded being closer to what you're actually requesting. Ultimately, everyone should request about the same credit per unit - this brings us a step closer to this ideal. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Charles Dennett Send message Joined: 29 Apr 00 Posts: 27 Credit: 18,785 RAC: 0 |
Thanks! I'm already running an optimized 4.13 but I thougth I'd try this one anyway to see how it worked. Downloaded the athlon-xp version and the library and installed. Still missing a gcc library. When I tried running this newer version I get this: boinc -run_cpu_benchmarks boinc: /lib/libgcc_s.so.1: version `GCC_3.3' not found (required by /usr/lib/libstdc++.so.6) I checked my version of gcc and it's 3.2.2. In /lib I find this: ls -la /lib/libg* -rwxr-xr-x 1 root root 30324 Feb 25 2003 /lib/libgcc_s-3.2.2-20030225.so.1 lrwxrwxrwx 1 root root 28 Apr 26 2003 /lib/libgcc_s.so.1 -> libgcc_s-3.2.2-20030225.so.1 Looks like I'll have to update my gcc in order to get the newer library for that. Charlie |
Ken Phillips m0mcw Send message Joined: 2 Feb 00 Posts: 267 Credit: 415,678 RAC: 0 |
> Thanks! I'm already running an optimized 4.13 but I thougth I'd try this one > anyway to see how it worked. Downloaded the athlon-xp version and the library > and installed. Still missing a gcc library. When I tried running this newer > version I get this: > > boinc -run_cpu_benchmarks > boinc: /lib/libgcc_s.so.1: version `GCC_3.3' not found (required by > /usr/lib/libstdc++.so.6) > > > I checked my version of gcc and it's 3.2.2. In /lib I find this: > > ls -la /lib/libg* > -rwxr-xr-x 1 root root 30324 Feb 25 2003 > /lib/libgcc_s-3.2.2-20030225.so.1 > lrwxrwxrwx 1 root root 28 Apr 26 2003 /lib/libgcc_s.so.1 > -> libgcc_s-3.2.2-20030225.so.1 > > Looks like I'll have to update my gcc in order to get the newer library for > that. > Charles, If you actually succeed with the major miracle of updating your libraries, could you please post back here where you got the rpms from, which were needed, and just where they are supposed to end up, because, I've just given up trying to update mine (3.2.2 on rh7.3, old I know, but not sure if newer version will run on machine without overwhelming it), I've been trying to install 3.3.something for about 3 hours now, with absolutely no joy whatsover. I'm trying to persevere with linux in spite of the fact that I detest it with a passion, because, everything is so hidden, complicated, and cryptic, and I'm just fed up and getting nowhere!!!! I've reverted back to a non-optimised client for now. Bye for now, Ken Phillips BOINC question? Look here "The beginning is the most important part of the work." - Plato |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
> Thanks! I'm already running an optimized 4.13 but I thougth I'd try this one > anyway to see how it worked. Downloaded the athlon-xp version and the library > and installed. Still missing a gcc library. When I tried running this newer > version I get this: > > boinc -run_cpu_benchmarks > boinc: /lib/libgcc_s.so.1: version `GCC_3.3' not found (required by > /usr/lib/libstdc++.so.6) > > > I checked my version of gcc and it's 3.2.2. In /lib I find this: > > ls -la /lib/libg* > -rwxr-xr-x 1 root root 30324 Feb 25 2003 > /lib/libgcc_s-3.2.2-20030225.so.1 > lrwxrwxrwx 1 root root 28 Apr 26 2003 /lib/libgcc_s.so.1 > -> libgcc_s-3.2.2-20030225.so.1 > > Looks like I'll have to update my gcc in order to get the newer library for > that. > > Charlie > Thanks for the update Charlie. Updating your version of gcc is quite a major operation. It's difficult for me to test for compatability against older distro's as the oldest I'm now running is FC1. OK, seems we're still missing some libraries. The best thing I can do is release a statically linked version of the boinc client containing the necessary libraries built in. That should run without problems for you although it will be a fraction slower than the dynamically linked version you just downloaded. Still, it's way faster than before. I'll post statically linked versions in the next day or so, so please check back. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Shaktai Send message Joined: 16 Jun 99 Posts: 211 Credit: 259,752 RAC: 0 |
> Thanks for the update Charlie. Updating your version of gcc is quite a major > operation. > > It's difficult for me to test for compatability against older distro's as the > oldest I'm now running is FC1. OK, seems we're still missing some libraries. > > The best thing I can do is release a statically linked version of the boinc > client containing the necessary libraries built in. That should run without > problems for you although it will be a fraction slower than the dynamically > linked version you just downloaded. Still, it's way faster than before. > > I'll post statically linked versions in the next day or so, so please check > back. > > Ned Thanks Ned. I also tried to get it running as well and ran into the same problem. Thought I found the correct libraries online, but still can't get it to work. Running Suse Linux 9.1. At least I didn't break anything, but too much a newbie for some of it to make sense. The statically linked version will work fine, even if it isn't 100%. Thanks for all the hard work, and for putting up with newbies like me. Team MacNN - The best Macintosh team ever. |
simplify / aki Send message Joined: 30 May 99 Posts: 61 Credit: 2,801,095 RAC: 0 |
hi i downloaded and compiled gcc 3.4.3 to test the better optimization. i compiled a statically linked version and it is a bit slower than your dynamically linked version (still much faster than gcc 3.3.3) so if there are problems with missing library feel free to download this version (signature) |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
Hi everyone, OK, I've updated my site. The dynamically linked versions have been removed (if anyone still wants these, they can contact me directly). I've posted statically linked versions that should have greater compatability across a greater range of linux distros. They are almost as fast as the dynamically linked versions and are MUCH faster than all previous versions :) Again, please let me know if you have any problems with these. @ simplify / aki - Great that you found the same thing. I wonder if it would be worthwile us getting together to test some further optimisations so we're not just duplicating each others efforts? Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
Charles Dennett Send message Joined: 29 Apr 00 Posts: 27 Credit: 18,785 RAC: 0 |
I tried looking for some gcc RPMs last light and could not find any. I downloaded the soruce for the latest gcc. It compiled easy enough. It took about an hour to build on my system with an AMD XP2600+. Additional software is requied to test the build (DejaGnu if I recall). I stopped short of installing it as I was a bit hesitant. Now that I see Ned has posted statically linked versions so I'll try those. Besides, I burned the CDs for Fedora Core 2 a while ago but never got around to upgrading. Now that FC3 is out, I should upgrade to there one of these days. In the meantime I use synaptic to check for updates to what I have installed. (Nice utility.) I don't have the benchmark numbers from before I compiled my own optimized 4.13 CC a few weeks ago from instructions found via one of Ned's previous posts, but they about doubled the numbers from the unoptimized version downloaded from the project site. I'll report back with the new numbers from the latest optimized version. Charlie |
Charles Dennett Send message Joined: 29 Apr 00 Posts: 27 Credit: 18,785 RAC: 0 |
The latest binaries are a bit faster. My testing shows a 9.0% increase in the Whetstone benchmark and 2.8% increase in the Dhrystone benchmark. Using my previously optimized version, the numbers were 2004/4205. Under the same conditions using Ned's latest binaries the numbers are 2184/4323. This was done on an AMD XP2600+. Thanks, Ned, for all that work! Charlie |
Bernard Pidoux, F6BVP Send message Joined: 28 Jan 02 Posts: 14 Credit: 6,369,782 RAC: 0 |
> I tried looking for some gcc RPMs last light and could not find any. I > downloaded the soruce for the latest gcc. It compiled easy enough. It took > about an hour to build on my system with an AMD XP2600+. Additional software > is requied to test the build (DejaGnu if I recall). I stopped short of > installing it as I was a bit hesitant. There are RPMs for GCC 3.4.1 and libstdc++6-3.4.1 on the mirror sites of Mandrake Linux 10.1 distro (based on RedHat) ftp://nephtys.lip6.fr/pub/linux/distributions/mandrake/current/i586/media/main/gcc-3.4.1-4mdk.i586.rpm ftp://nephtys.lip6.fr/pub/linux/distributions/mandrake/10.1/i586/media/main/libstdc++6-3.4.1-4mdk.i586.rpm Bernard |
Bernard Pidoux, F6BVP Send message Joined: 28 Jan 02 Posts: 14 Credit: 6,369,782 RAC: 0 |
I made some observations about the performance depending on library libstdc++ version used. I have two PCs with the same CPU (Intel Pentium III Katmai at 451 MHz) giving 897 BogoMIPS says Linux kernel) PC#1 runs 2.4.25 kernel (Mandrake Linux 9.2 distro) with libstdc++.so.5.0.5 PC#2 runs 2.6.8 kernel (Mandrake Linux 10.1 distro) with libstdc++.so.6.0.1 Here are the results of CPU benchmarks with the pentium3-pc optimized boinc_4.13 PC#1 450 W / 928 D PC#2 489 W / 1007 D ----- ----- +8.6% +8.5% And here are some additional comparizons between boinc optimized 4.13 versions BOINC optimized version PC#1 PC#2 boinc 4.13 i686 pclinux (GCC 3.2) 412 W / 562 D 490 W / 860 D boinc 4.13 i686 pclinux (GCC 3.4) 489 W / 917 D Conclusion : +6.6% performance improvement is given by the GCC 3.4 compiler (at least for Integers), +8.5% performance improvement in number crunshing is added by libstdc++ version so.6 A total of +17% on integer MIPS is obtained when using the appropriate compiler optimization (i.e. PIII versus i686), and the more recent versions of GCC and libraries. And final remark, boinc_4.13_pentium4 will crash during benchmarks with SIGILL illegal instruction that seems rather normal with a pentium III CPU. Bernard |
Ned Slider Send message Joined: 12 Oct 01 Posts: 668 Credit: 4,375,315 RAC: 0 |
> > And final remark, boinc_4.13_pentium4 will crash during benchmarks with > SIGILL illegal instruction that seems rather normal with a pentium III CPU. > > Bernard > Yes, that's correct, I see the same thing attempting to run the P4 client on Athlon XP processors. The P4 version was compiled with -march=pentium4 and will generate code that's scheduled specifically for that architecture with no regard for compatability. Using -mcpu would schedule code for a given architecture whilst retaining i386 compatability, but would probably be less optimised as a result. Interestingly, I did note that the PIII code appears to run on Athlon XP's, at least for the benchmarks, albeit slower than the athlon specific code. Finally, the speed improvements afforded by gcc-3.4.2 on the boinc benchmarks are not mirrored for the seti client compile. Ned *** My Guide to Compiling Optimised BOINC and SETI Clients *** *** Download Optimised BOINC and SETI Clients for Linux Here *** |
©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.