The BIOS is as new as possible, I think from 28.01.2005. memtest had been
If you have the "memory hoisting" or "memory remapping" or similar option in the BIOS enabled try to disable it. If that doesn't help see below.
running for a couple of hours but found nothing. I tried different boot-options for no acpi, iommu=force/noforce and played around with the BIOS options and so on but it still freezes. We also download SLES9 and installed it on a second hdd but the system freezes also. We watched some processes to be killed sometimes, so we have to cases: the system freezes or our test processes die suddenly. We test the system with one of our own binarization-program wich needs lots of RAM, vgstudiomax (also lots of RAM) or the 64bit-setiathome. I got a kernel oops only one time saying a process running at CPU3 tried to adress RAM of CPU0. I couldn't reproduce this oops but we are now checking the RAM-Modules. The system is currently running at 8GB (2G at each processor) for testing. We got 8GB of Infinion and 24GB of Samsung-Modules. Maybe mixed RAM don't work correctly somehow. Its stable now for a couple of hours. I rebuilded the smp-kernel without NUMA-support and it was stable for 8 hours or so. Normaly the systems freezes after a few minutes if all cpu's are fully loaded.
This sounds very much like a memory or other hardware problem (especially the rebuilding without NUMA helps bit - this just changed the memory access patterns). Unfortunately memtest86 cannot catch them all because it only runs on a single CPU and also doesn't do any IO load. When memory is unstable installation can randomly hang too. I would contact your hardware vendor. -Andi