Linux Benchmark Note- V5.8.17 corrects the problem
10 Mar 2007 12:09:46 UTC
Topic 192517
(moderation:
If anyone is running Linux on other projects, the newest version of Boinc was compiled with the new compiler and makes a great difference in the benchmark scores. My AMD 64 X2 went from 4.0/7.2 to 4.4/10.6 (GFIOps/GIOps).
Linux Benchmark Note- V5.8.17 corrects the problem
)
Really nice, gotta try that out soon. Since my laptop is dual-booted (between Win 2k and Linux) I'll have a rather good indicator of how realistic the scores are. Gonna keep you guys posted.
LOL, well if that proves out
)
LOL, well if that proves out for other Linux hosts all I can say is, it's about time! ;-)
Alinator
RE: If anyone is running
)
Sorry to be a damp squid but a little note of warning.
5.8.17 is an alpha version for testing only. Ohiomike has already found one problem in his other thread.
To give a bit of history, the problems with Linux benchmarks have been down to the way that the code is compiled. MS have extorted billions of dollars out of people like me and can ensure that their compilers are fully backwards compatible. Linux have taken a different course which means that the compilation of Linux BOINC has been held back to include as many versions as possible. Despite work arounds and optimisations the Linux benchmarks have always lagged behind the Windows ones.
The developers have now bitten the bullet and compiled a new version on Glibc 2.4. Initial testing shows that it appears to have worked on Suse 10.1, Ubuntu 6.10 and Fedora Core 5 (with updates) with greatly increased benchmarks but that it has failed testing on earlier distributions e.g. Suse 10.0 which uses Glibc 2.3. Feel free to use it but be aware of its limitations and that it is a test version. Dave
Using BOINC 5.8.17 on SuSE
)
Using BOINC 5.8.17 on SuSE Linux 10.1 my 400 MHz Pentium II Deschutes soared to 404 floating point MIPS and 698 integer MIPS. Wow!
Tullio
Hm, so how are 5.8.17 Linux
)
Hm, so how are 5.8.17 Linux Scores comparing to 5.8.15 Win32 ones ?
From reading the improvements, sounds like they maybe overdid it (going from 5.8.11 or .15 I'd need approx. 14% improvement to be exactly on par)
I tested it on one System,
)
I tested it on one System, and the Benchmark improvement was 47% (!) going from 5.8.15 to 5.8.17.
Now it benches 33% too high compared to Win32 actually, looks like they actually overdid it (even ~15% higher scores than other, more conservatively optimized Clients I tested before to get it as accurate as possible).
Unfortunately, I won't be
)
Unfortunately, I won't be able to test it as Debian Sarge and the new BOINC manager don't really like each other. So, it's still 5.4.11 for me, but I don't mind too much as Einstein and DRTG give fixed credits anyway...
RE: Unfortunately, I won't
)
BOINC core client runs fine on my Debian Sarge. BOINC manager won't run as I don't have installed packages libxinerama and libxxf86vm. I expect that if all the packages are installed, also BOINC manager should run on Sarge.
Metod ...
Yep, I added the packages and
)
Yep, I added the packages and got the new BOINC manager running just fine. However, I must say that while the benchmarks are considerably better than with the old Linux client they are still far from what I get under Windows on this machine. Maybe it has to do with having a "mobile" CPU (older Banias-core Celeron M)? Not complaining, just curious. I prefer fixed credit anyway, but I wonder while this seems to work for many people but not for me. I'm, as I said, on Debian Sarge, 2.6.16 kernel.
Sorry, my bad. I was talking
)
Sorry, my bad. I was talking about 5.8.15, the current stable release. Since 5.8.17 is running into "GLIBC" problems I don't want to risk playing around with it atm, but of course this explains the somewhat lower benchmarks. Sorry about that.