A new version of Gimps is out at ftp://mersenne.org/gimps/mprime2411.tar.gz This version yields an average 15% speed improvement for 32-bit Athlon processors over 23.8. An Athlon64 optimization was found for the 32-bit version. You'll get about a 15% performance boost. Still not as fast as a similarly clocked P4, but it is much closer. Reproducible: Always Steps to Reproduce:
The linux executable contained in the archive above is dinamically linked with the libstdc++.so.6 library, which AFAIK is only provided on systems with the GCC 3.4.x compilers. For x86 (which still uses GCC 3.3.5) it would be better to use if we the statically linked executable contained in ftp://mersenne.org/gimps/sprime2411.tar.gz, which has no other dependencies. Still waiting for a native 64-bit executable for amd64, though. Reasons why it's not provided are in the following discussion: http://www.mersenneforum.org/showthread.php?t=3934 Maybe someone can help.
Yet a newer (still beta) version is out at ftp://mersenne.org/gimps/mprime2412.tar.gz (dynamically linked version, fine for amd64 and ~x86 with gcc 3.4.x) and ftp://mersenne.org/gimps/sprime2412.tar.gz (statically linked, OK for x86). Performance improvements for SSE2 machines.
Version rc2 of the Gimps client for linux machines is online at: ftp://mersenne.org/gimps/mprime2412.tar.gz (dinamically linked, works if gcc >= 3.4) ftp://mersenne.org/gimps/sprime2412.tar.gz (statically linked works with other compilers). I know the filenames are the same of the previous comment but the files inside have several bugfixes and improvements on the previous ones.
24.13 in portage, closing a stale bug.