On Thu, Oct 08, 1998 at 04:51:40PM +0200, Ricardo Galli wrote:
> We compared in our University a single process (a chess player) running is a
> Octane 2xR10000 against a UP Linux on a PII 233. The program is totally
> programmed in ANSI C. We compiled using O2 option on Linux and all 64 bits
> optimization on the Octane. In the same period of time, the Linux version
> did 180.000 operations, Octane about 100.000 (take in account is a single
> threaded process).
64bit are in general slower than 32 bit because you'll have a signifincatly
lower TLB and cache hit rate and higher memory use. Depending of code, cache
sizes and size of working set this alone can already impact the performance
significantly.
MIPS typically performs quite a bit better on fp operation and somewhat
worse on integer compared to Intel CPUs of the same class. Good results
on MIPS really want SGI's compiler. It's often a bitch to use but it beats
the living shit out of GCC in terms of performance.
In fact Mips own compiler from '93 still makes better code than today's
egcs :-( Not that I would recommend anyone to touch RISC/os with anything but
a long pole ...
> Finally, SGI is going to change radically in few months, they are going to
> sell NT machines with Intel processors. MIPS factory has been sold months
MIPS neither owned factories nor has been sold. For quite some time MIPS
CPUs will still the engine to power non-desktop SGIs which later will
change over to Merced. NT only for SGI's toy class of machines. You may
hate IRIX but for the next time it's still the OS that solves quite a
number of jobs best.
Ralf
