I downloaded Matlab 7.04 hoping that would fix my problem. It didn't. It exhibited the same segmentation fault upon startup as before. Both 7.01 and 7.04 work on my Suse 9.1 single, but fail to work on dual 9.2 and SLES9 machines. Strange because SLES9 uses an older kernel than 9.1. Perhaps it is an smp issue? No matter, as a little googling found me the solution. For 7.01 on AMD64 people said to add "export LD_ASSUME_KERNEL=2.4.1" to the beginning of $MATLAB_INSTALL_DIR/bin/.matlab7rc.sh (notice the leading dot) I tried this and it worked (by worked I mean it ran the bench program a few times). Since 7.04 has a 2.6 build config I also tried 2.6.1 and it worked as well. On both 9.2 and SLES9 boxes. 2.6.3 would load but crashed during bench. 2.6.5, 7, and 8 all segfaulted upon load. Friggin Matlab. I wonder why they would fail to adequately support the most popular AMD64 distro. It's not like people who use Matlab require the most memory and thus be the first to migrate to inexpensive 64-bit platforms. I just can't wait to see how they handle the fact that most (near-term at least) future performance increases will be coming from multiple cores. Joe
SP1 works fine for me as well on SuSE 9.1. Since I've only just started migrating to 9.2 I hadn't noticed, but it segfaults there. Even with -novjm and -nosplash. Fujitsu Siemens desktop with AMD 64 3200+, SuSE 9.2 Pro and 2.6.8-24.11-default.
I've been comparing strace outputs on SuSE 9.1 on Opteron, SuSE 9.2 on Athlon 32 (both work fine) and SuSE 9.2 on AMD64 (which segfaults). The latter strace output is _littered_ with gettimeofday() and time() calls which aren't in the other two outputs.
Googling reported something similar on RHEL 3, where it was a kernel bug. Neither acpi=off nor various perturbations of clock= (tried tsc, pmtmr, and pit) had any effect.
Matlab R14 SP2 is out, I'll have to give that a try to see if it solves the problem - I'm just glad I found out before I went into production with 9.2 on my desktop machines. Thanks Joe.
Bjørn