Re: Irix performance

guy albertelli (albertel@cis.ohio-state.edu)
Sun, 4 May 1997 17:22:26 -0400 (EDT)


> > Adding a bit to what I said earlier, I'm getting about 480 Kkeys/s on a
> > 250 MHz R4400 running Irix 6.2, and 820 Kkeys/s on an R10000, also
> > running 6.2. I'm wondering if the much worse performance on the 5.3 box
> > (~180 Kkeys/s) is because of the OS or because it only has a 1 MB L2
> > cache, vs. 2 MB on the 4400 running 6.2 (plus a little extra for 250 vs
> > 200 MHz).
>
> Just guessing, but I'd say it's PRIMARILY from the OS revision. The 6.2
> code is backward compatible but was still compiled with intentions of
> running on systems using more advanced features I guess.

Nope this isn't it at all, it is most definately at least the MHz change
and then the L2 cache, there might also be slight differences in the chips
internal design. SGi has made lots of little variations on the R4k design.
Likely that there maybe a small change in the amount of On-chip cache,

run hinv on the SGIs, that will let you know if there are any other
hardware differences. But the code was compiled as an old 32-bit client,
rather than the new 6.X Irix 32--bit or 64-bit versions.

(And to head off complaints, this was done because it was the fastest
version. There has been alot of expirementation with command line options
to compilers, on many different SGIs)

-- 
Guy Albertelli II   albertel@pilot.msu.edu | "And God rested, chuckling at
O- http://www.cis.ohio-state.edu/~albertel | His own little play on words"
--------------------------------------------------------------------------
    A good scapegoat is nearly as welcome as a solution to the problem.