Hello, I wanted to report on my experience with the Supermicro PIIIDME. ===================================================================== Abstract : running 2 PIII800E with 1Gb on the the Supermicro PIIIDME I bought proved unstable when making compilations or ssh remote connections on X, from kernel 2.2.14 to kernel 2.4.0-test3 ; although the latter solved some of the stability problems, there were segfaults appearing. Certainly the MTH that has problems with SDRAM (and not only ECC sdram). I changed for the Supermicro P6DBE which, with the same hardware, is stable for me. ===================================================================== I have 2 PIII800E and 4 modules of 256Mb Non-ECC PC100 SDRAM from Venturatech. Well, I bought the PIIIDME in the beginning of June, and had many problems with it, notably a short circuit that wouldn't let me boot the computer. I use the Supermicro SC-760 case. So I had to remove some of the metal clips I had put on the chassis. I then installed SUSE 6.4, with reiserfs support. But then I had random problems with some IRQ vectors arriving in the CPU that blocked it, and had some SCSI resets that blocked the computer too, and some conflicts between the UDMA setting and the SCSI. I sent a mail here, and many people responded (thank you all !), which made me check many things (notably the SCSI termination), and upgrade to kernel 2.4.0-test3 with reiserfs 3.6.10 patch. It then seemed to work well : no more SCSI resets or IRQ vectors randomly blocking the CPU, or UDMA problems. But when compiling the 2.4 kernel, I had had some random segfaults. So I was suspicious. I ran math programs (fourier transforms of images), that took the whole 1Gb memory and even swapdisk, and it looked stable. I then tried to change some settings in the kernel, and had to recompile it, which unleashed segfaults ! I went to the sig11 webpage http://www.BitWizard.nl/sig11/, and found a test that consisted of making multiple kernel compiles and checking for differences in the output. Well, the compilations failed after some random time, even rebooting my computer sometimes ! I also found Doug Ledford's page very useful : http://people.redhat.com/dledford/ I also devised another very good test for the segfault : I ran an ssh remote connection to a server, from which I launched netscape, and after playing around with netscape, I inevitably had a segfault, or a total blockage of the computer. My own programs also segfaulted when compiled. Even gnorpm segfaulted ! So, a small program like ssh, using a small part of memory, can put the i840 on its knees, whereas a program that fills the memory and computes stuff on it might not. Funny. I then changed many settings in the bios, but the segfaults were always there. Also removed all the cards, swapped them, etc. All very depressing. I then replaced the memory with 2 modules of 64 Mb, and it worked well, no segfaults. I tested the 1Gb on another computer, no segfaults (these tests were described in the very useful sig11 page). Of course, I looked up everything I could find on the web concerning the PIIIDME, and found some benchmarks that proved its instability even with SDRAM non-ecc. See http://www.2cpu.com for references. I then decided to change the board. I bought a P6DBE Rev 3.0, that supports the PIII800E. And there was light. That P6DBE is rock stable with my 1Gb, and 2 PIIIDME 800 under kernel 2.4.0-test3 with reiserfs 3.6.10 patch and my 120 Gb of disk. No compilation problem, no segfault, happiness ! That's all, I hope it proves useful to someone ! Cheers, Yann Le Du ------------------------------------------------------------------------ Dr Yann Le Du E-mail: [EMAIL PROTECTED] Theoretical Physics Web : http://cdfinfo.in2p3.fr/~ledu/ 1, Keble Road University of Oxford Oxford, OX1 3NP Phone : (44) (0)1865 273 989 United Kingdom Fax : (44) (0)1865 273 947 ------------------------------------------------------------------------ - Linux SMP list: FIRST see FAQ at http://www.irisa.fr/prive/dmentre/smp-howto/ To Unsubscribe: send "unsubscribe linux-smp" to [EMAIL PROTECTED]
