On 16 Feb, Marcelo E. Magallon wrote: > Hi, > > I know this is off topic, but I don't have access to cola (and > newsgroups in general) and I feel more confortable asking here, because I > want Linux specific answers. > > Ten days ago a professor here bought a Pentium II/233 system. He > promptly installed Debian on it, and let me use it for my (thesis) work. > First thing I did was to benchmark the thing using a program of my own. > This program says a Pentium MMX/166 (my old pc) gives about 24 Mflop/s. A > Pentium/100 is about 16 Mflop/s. If the numbers are accurate or not, is > not in dispute now. What's important is the relative speed, and I find the > numbers quoted to be reasonable. (For those curious, it's 3 sums, 3 > multiplications, 1 division) > > The PII says 39 Mflop/s. Over the weekend, I bargained a PMMX/233, which > says 33 Mflops/s. I don't find this this reasonable at all! Taking as a > reference the performance leap from a 486DX4 -> Pentium (same clock speed) > I was kinda hoping something near 80 Mflop/s for the PII (yes, I know, > it's silly to take that as a reference, but one can only hope) > > I know I'm not playing fair comparing the systems this way (different > kernels, memory, chipset, ...) but I was hoping somebody could give better > statistics on this. > > I'd really appreciate if somebody can help me on this one. We are planing > to build a Debian-based compute farm, and the cost difference between > PII's and plain Pentium's could translate into a big difference in the > number of hosts installed. > > Side note: K5/133 = 9; K6/200 = 31; 486/66 = 4; RISC 9000 = 18; VAX > 3000... oops, forgot about it, but it was surprisingly low. > > TIA, >
Which compiler were you using? Programs not specifically optimized for the PPro/PII don't get nearly the performance gain that they could. My experience is something between a 30 and a 50% performance gain possible from using code compiled and optimized specifically for a PPro. I haven't had time to try it yet, but gcc 2.8.0 compiles for PPro/PII, as does egcs (and I think, therefore, pgcc, as that is based on egcs(?) - but its been a while since I checked either of these). -- Stephen Ryan Debian GNU/Linux Mathematics graduate student, Dartmouth College -- TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to [EMAIL PROTECTED] . Trouble? e-mail to [EMAIL PROTECTED] .
