Re: AMD K6-2 / Bogomips problem
On Thu, 21 Jan 1999, Peter Bartosch wrote:
> > On 19-Jan-99, Peter Bartosch took time to write :
> > >> bogomips should only depend on hardware no ?
> > >
> > > only on cpu (and clock)
> >
> > that's what i thought too, but it doesn't seem true in my case
> > as with exactly the same hardware and no change in bios
> > i have
> > 700 bogomips with kernel 2.2.0pre7
> > and 350bogomips with kernel 2.0.36 compiled either as 586 or as 686
> >
> > they are all compiled with same options (as much as possible)
>
> i've read somewhere, that this lies on the size of the cpu's cache (1st
> level) this is IIRC in K6-chips twice as big as in pentiums, but if i think
> over it this couldn't be true/realistic
Actually, the main critique of the real MIPS calculation is that it is
cache-dependent because it is a small loop. While I know nothing of the
bogomips calculation, I assume it's similar. The problem is that the
entire piece of code that calculates this fits inside the cache. Pentiums
(and MMX) came standard w/ 512k, so I doubt the loop would be bigger than
that, so I don't think doubling the cache size would help any. (Just
celerons would really get bad numbers) :)
-Dan
Reply to: