Message boards :
Number crunching :
Linux benchmark ?!?!?!
Message board moderation
Author | Message |
---|---|
WerK Send message Joined: 30 Jun 02 Posts: 26 Credit: 221,390 RAC: 0 |
OK I'm running Linux-2.6.7 so i DLed the CLI client 3.18 for linux, works ok. I have AMD Barton 3200+ processor and the benchmark results are ~1100 floating, ~2400 integer. Then I tried to run the 3.19 windows boinc binary boinc_cli.exe ( the boinc_gui.exe was screwing up ) using wine emulation facility. I was very very surprised about the results, because even it was emulating windows environment i was getting around 2600 float and 4600 integer ( you can have a look @ http://setiweb.ssl.berkeley.edu/hosts_user.php?userid=100170 , trust me, its really the same machine ). So I'm asking : why Linux machines are getting 2x lower benchmark results ??. Is it compensated (will be?) or Linux users are really getting almost 2x lower credit? P.S. : I'm getting around 25 cobblestones claimed from 1 WU crunched, somebody with same processor as me & windows please post your claimed credit average, thanks WerK |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
Windows is better, more sexy, faster, secure, ;-) OOOPS a little flame war tread ;-))) [/url] [/url] |
WerK Send message Joined: 30 Jun 02 Posts: 26 Credit: 221,390 RAC: 0 |
No problem ;-). If thats right and Berkeley are windows-oriented guys, i will experiment with optimal config and run boinc under wine emulation like I was doing with Seti@Home Classic ( the 3.08 linux was horribly slow - about 3hr/WU, emulated 3.03 windows was doing ~2:20h/WU ). Its proven that wine is faster than native windows because of "cleaner" implementation of API. But i hope that BOINC is really "multiplatform" :))) |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
allright you unterstand a joke have a nice weekend and if i see good stuff for linux on the web i post it on the board here [/url] [/url] |
Zorous Send message Joined: 25 Sep 01 Posts: 7 Credit: 897,254 RAC: 0 |
I am finding the exact thing happening on my P4 laptop that I dual boot with XP Pro and Suse 9.1...the performance (according to the benchmarks) seems to be more than double on XP!! That is messed up. |
Alex Send message Joined: 26 Sep 01 Posts: 260 Credit: 2,327 RAC: 0 |
> I am finding the exact thing happening on my P4 laptop that I dual boot with > XP Pro and Suse 9.1...the performance (according to the benchmarks) seems to > be more than double on XP!! That is messed up. > I found the same thing http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=61773 Machine running windows client under wine. http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=37586 Machine running linux client. Results Linux Measured floating point speed 132.98 million ops/sec Measured integer speed 245.24 million ops/sec Wine Measured floating point speed 315.06 million ops/sec Measured integer speed 524.11 million ops/sec |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
[/url] [/url] |
Grant Send message Joined: 3 Apr 99 Posts: 4 Credit: 9,600 RAC: 0 |
I was skeptical about this until I tried it myself. I'm running on Fedora Core 2, Kernel 2.6.6 on a Pentium 2 333: Linux Command Line Client: http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=59887 Measured floating point speed 172.39 million ops/sec Measured integer speed 359.18 million ops/sec Windows GUI Client (using wine, Win95 emulation): http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=62037 Measured floating point speed 396.51 million ops/sec Measured integer speed 680.06 million ops/sec |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
[/url] [/url] |
markkuk Send message Joined: 9 Oct 00 Posts: 6 Credit: 1,214,242 RAC: 0 |
> I was skeptical about this until I tried it myself. I'm running on Fedora > Core 2, Kernel 2.6.6 on a Pentium 2 333: > > Linux Command Line Client: > http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=59887 > Measured floating point speed 172.39 million ops/sec > Measured integer speed 359.18 million ops/sec > > Windows GUI Client (using wine, Win95 emulation): > http://setiweb.ssl.berkeley.edu/show_host_detail.php?hostid=62037 > Measured floating point speed 396.51 million ops/sec > Measured integer speed 680.06 million ops/sec The Windows client is compiled with Microsoft Visual C++, the Linux client is compiled with GCC 3. GCC sacrifices performance to gain portability. It would be interesting to compile the client from source using the Intel C++ Compiler for Linux to see what kind of benchmarks it gets. |
Chris Bosshard Send message Joined: 5 Jun 99 Posts: 86 Credit: 3,474,583 RAC: 0 |
I absolutely agree.... Already the old client had the same issue. It underperformed when comparing with the same box runing under Windows. The good thing about the new version is that it is open source and optimizations can be done.... Anybody that is working on optimization of the code for better performance? Is there someone that can compile the original code with the intel compiler? I would be ready to do some testing... ;-) Chris Bosshard astroinfo Team http://www.astroinfo.org |
Grant Send message Joined: 3 Apr 99 Posts: 4 Credit: 9,600 RAC: 0 |
its worth a try. I'm gonna go get the source code for BOINC and SETI now and see if i can get GCC to optimize it for my CPU and see if that helps at all. |
Grant Send message Joined: 3 Apr 99 Posts: 4 Credit: 9,600 RAC: 0 |
So far I have the current CVS version of the BOINC client recompiled. In order to optimize the compile i set the following environment variables: export CFLAGS="-O3 -march=pentiumpro -funroll-loops -fexpensive-optimizations" export CXXFLAGS=$CFLAGS then ran: ./configure --disable server I havent recompiled the actual SETI client yet, but I ran a benchmark on just the BOINC client. Here are the stats: Linux Command Line Client (precompiled distribution client): Measured floating point speed 172.39 million ops/sec Measured integer speed 359.18 million ops/sec Linux Command Line Client (optimized and self compiled): 319 double precision MIPS (Whetstone) per CPU (floating point speed) 347 integer MIPS (Dhrystone) per CPU (integer speed) The compile time optimizations greatly increased the FPU speed, but didnt do anything for the integer speed. I'll have to see if I can find any more GCC options that will help optimize that as well. |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
OOPS the dark linux force OOPS ;-) hi freaks have a nice sunday [/url] [/url] |
WerK Send message Joined: 30 Jun 02 Posts: 26 Credit: 221,390 RAC: 0 |
thx much for feedback guys, did some RTFM and got first results : 1946 double precision MIPS (Whetstone) per CPU 2731 integer MIPS (Dhrystone) per CPU using: export CFLAGS='-O3 -mcpu=athlon-xp -march=athlon-xp -mfpmath=sse -m3dnow -msse export CXXFLAGS=$CFLAGS which is much better now :] I'm going to play with flags more and post the best settings |
markkuk Send message Joined: 9 Oct 00 Posts: 6 Credit: 1,214,242 RAC: 0 |
> It would > be interesting to compile the client from source using the Intel C++ Compiler > for Linux to see what kind of benchmarks it gets. An initial result: the CVS snapshot boinc-cvs-2004-07-10.tar.gz compiles with ICC after one correction, but the resulting program crashes with SIGSEGV. That makes the version compiled with GCC infinetly faster :-) |
Grant Send message Joined: 3 Apr 99 Posts: 4 Credit: 9,600 RAC: 0 |
WerK: Try running "-funroll-loops -fexpensive-optimizations" as well. It may help bump up your stats a bit more, and i'm interested in seeing by how much. |
WerK Send message Joined: 30 Jun 02 Posts: 26 Credit: 221,390 RAC: 0 |
> WerK: > > Try running "-funroll-loops -fexpensive-optimizations" as well. It may help > bump up your stats a bit more, and i'm interested in seeing by how much. > > Played with it a bit more and found this setting : export CFLAGS='-O3 -mcpu=athlon-xp -march=athlon-xp -mfpmath=sse -m3dnow -msse -mmmx -funroll-loops -fomit-frame-pointer -ffast-math -s -static' the ' -fexpensive-optimizations' are already included because use of '-O3', you can find some nice info on this page : http://www.freehackers.org/gentoo/gccflags/flag_gcc3opt.html So, with this config I have following results : float - 2558.21 million ops/sec integer - 3172.54 million ops/sec (the floating point is now same as in Windows binary, integer increased a bit, but its still much lower than Windows binary) |
Guido_A_Waldenmeier_ Send message Joined: 3 Apr 99 Posts: 482 Credit: 4,774 RAC: 0 |
Ground Control to Major Tom Ground Control to Major Tom Take your protein pills and put your helmet on Ground Control to Major Tom Commencing countdown, engines on Check ignition and may God's love be with you (spoken) Ten, Nine, Eight, Seven, Six, Five, Four, Three, Two, One, Liftoff This is Ground Control to Major Tom You've really made the grade And the papers want to know whose shirts you wear Now it's time to leave the capsule if you dare "This is Major Tom to Ground Control I'm stepping through the door And I'm floating in a most peculiar way And the stars look very different today For here Am I sitting in a tin can Far above the world Planet Earth is blue And there's nothing I can do Though I'm past one hundred thousand miles I'm feeling very still And I think my spaceship knows which way to go Tell my wife I love her very much she knows" Ground Control to Major Tom Your circuit's dead, there's something wrong Can you hear me, Major Tom? Can you hear me, Major Tom? Can you hear me, Major Tom? Can you.... "Here am I floating round my tin can Far above the Moon Planet Earth is blue And there's nothing I can do." [/url] [/url] |
Chris Bosshard Send message Joined: 5 Jun 99 Posts: 86 Credit: 3,474,583 RAC: 0 |
To compile the source with optimizations shows certainly improvements. Just some critical questions: Will the results from self compiled clients be accepted by the Project when compared with results calculated with the official clients? Is there a way to make sure that it will be accepted? Is there a way to self certify the self compiled clients? What do you guys think? Chris Bosshard astroinfo Team http://www.astroinfo.org |
©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.