Opteron Benchmarks (22 Jul 2003)
- The League of Extraordinary Gentlemen and the public domain [via Lessig]
- A new addition to the common links at the top of the page: Whiterose. (no relation to my code of the same name)
These results are completely unfair and shouldn't be taken as gospel in any way. Lithium is the dual Operton (it's specs are in a previous post), lithium32 is Lithium running 32-bit code (Mandrake 9.1) and Loch is a dual 2.66Ghz Xeon with only 1GB of memory (Lithium has 4). Lithium only has an ATA disk, while Loch is LVD SCSI. I tried as much as possible to allow both systems to keep everything in buffer cache.
Everything was run with GCC 3.3, but keep in mind that Lithium (in 64-bit mode) is actually building a slightly different kernel (x86-64, not i386).
make 2.6.0-test1, default configure, no -j lithium: 4m33 lithium32: 5m26 loch: 5m41 make 2.6.0-test1, after make clean, -j4 lithium: 2m27 lithium32: 2m48 loch: 2m57 make 2.6.0-test1, after make clean, -j8 lithium: 2m28 lithium32: 2m48 loch: 3m08 md5sum of 512M zero file (in buffer cache) lithium: 3.1s lithium32: 10.6s loch: 3.4s stimpy: 16.8s
Stimpy is another Mandrake box because I didn't quite believe the result for Lithium in 32-bit mode. It seems that Mandrake 9.1's md5sum just sucks, so ignore lithium32's result in that.