When Intel went to the on chip L2 cache it was not to increase
performance, it was to save money. They cut the size of the cache from 512
Kb to 256 Kb. My old Pentium 2 L2 cache tested at half of the L1 cache
speed, which it is supposed to be. With my Pentium 3 (1.26 Ghz with 512 Kb
L2 cache) the L2 cache tests at 60% of L1 cache speed, not 100% as Intel
would have you believe.

The *frequency* at which the cache works in the on-die chips is actually
100% of clock speed. However, as you point out, that doesn't work out to
100% performance, as there are a lot of factors involved. However, in
apps where cache latency is a factor, the chips with on-die cache do come
out ahead.

Perhaps the largest limitation to the P3 line is it's limitted FSB, at
133 MHz. I've used a good number of dual-p3 servers, and found that once
the CPU gets up to about 866, you've got pretty much all the performance
you're going to get - a faster CPU doesn't do anything for you. Once you
start talking about the "-S" chips (the later P3's with 512k on-die cache,
and clock speeds of 1.13 to 1.4 GHz), you get an initial performance boost
from the doubling of the cache, but little to no returns from increasing
the CPU speed along the 1.13- to 1.4-GHz line.

steve