Hi: I am posting here from a suggestion on the Debian amd64 site. My original posting to the mainboard factory/vendor in Europe only resulted in uninteresting suggestions, and they did not answer any more.
My question is directed to the attention of users familiar with multisocket UMA-type mainboards based on 875 dual opteron AMD CPU. My own is Supermicro H8QC8 with chipset nVidia CK804 and AMD 8132, driven by Debian Linux amd64 lenny. One of the CPUs has suddenly lost viability to its 4-slots memory bank (shut down the machine in order, the problem arose on next loading Linux). Still, the CPU cores are OK, hypertransport links are fully working, parallelization to both Amber 10 and NWChem 5.1 is fully provided, but one of the CPUs must be slower, having to borrow memory from the other banks. The hardware status, after a period of complete darkness, is described in the attached lshw_deb64_7Jan2009.txt. As each bank of Kingston DDR1 is filled 2+2+1+1 GB, I identified the faulty bank, removed all slots from there, and replaced the 1+1 GB slots at another bank with 2 + 2 GB from the faulty bank, so that now the computer is at 20GB. The situation is described in the attached lshw_deb64_lessCPU2_scrambling1G_2G_CPU4_7Jan2009.txt. Actually, identification of the CPU (CPU2) related to the faulty mem bank is insecure: I just considered the nearest CPU to the faulty bank. The manual is not helpful to this regard . I understand that, in order to remove non-mainboard causes, I should be certain that a CPU has not lost memory control. Since replacing (I have one spare second-hand CPU) or scrambling, the CPUs is quite troublesome, and risky, in my context (there is very little space around the mainboard in the rack that I engineered to accept the mainboard). Ventilation is excellent, however. Therefore, is it any software way to check if the CPUs are fully in order, including the memory controller? lshw and other software provided only partial help in my hands. Also any other suggestion would be greatly appreciated. Thanks for your kind attention francesco pietra _______________________________________________ Beowulf mailing list, Beowulf@beowulf.org To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf
