Re: Using >1GB RAM
Noah Beck (noah@ecn.purdue.edu)
Thu, 15 Jan 1998 15:00:32 -0500
Joel Jaeggli wrote:
>
> On Wed, 14 Jan 1998, Perry Harrington wrote:
> >
> > Curious thing, I took Intel's benchmarks and did some calculation. The PPro
> > gets faster when you increase the Mhz; the performance per Mhz goes up. The
> > P 2 gets slower ther faster it runs, performance per Mhz. My calculations
> > suggest that a PPro running at the speed a P 2 (300Mhz) runs at, it would be
> > faster than a P 2. This doesn't account for the fact that a PPro is a 64 bit
> > processor, since their benchmarks are 32 bit only.
>
> This can be accounted for in the l2 cache designs, since the ppro's l2
> cache runs at the internal clock rate of the cpu and the current
> incarnation of the pentium-II runs the l2 cache at a constant 2x the bus
> speed or ~133mhz so when you speed up a pentium-II the l2 cache doesn't
> get any faster. whereas on the pentium pro the l2 cache performance
> should increase in lock step with an increase in the cpu clock. Hence the
> promise of merced since they're promising 100mhz bus speeds along with
> everything else, even so if they still maintain the l2 cache as some
> multiple of the bus speed slower than the cpu clock you'd expect an
> increase in performance modula the pentium-II and not the pro.
>
> joelja
Where'd you get that info? I could have sworn that the datasheets for
the P-II said the L2 runs at half of the internal clock. (pause while
I look to see if I still have it, I do.) "Transfer rates between Pentium II
processor core and L2 cache are one-half the processor core lcock frequency
and scale with the Pentium II processor core." (from the Developer's
Manual.)
My guess for the performance per MHz increase in the PPro is their branch
prediction algorithms; they were trying lots of them at that time, I think,
and it could be that the higher-MHz generation got a better algorithm.