The floating-point execution speed of
AMD's K6-III line is not inherently
faster than that of the K6-2. The main
performance improvement in the K6-III is achieved through the
incorporation of an on-chip, full-speed CPU cache.
Most processors use at least two levels of CPU cache memory to improve
their execution speed: 'level 1' (L1) storage built into the chip and
'level 2' (L2) storage that normally resides within the supporting chip
set on the motherboard. L1 cache runs at the same speed as the CPU, but is
costly and thus small. Both K6-2's and K6-III's have 64kB of L1 cache. The
L2 cache in the K6-2 series is built to run only at the speed of the
system bus (either 66MHz or 100MHz, depending on the system). In the
K6-III, 256kB of L2 cache is brought right into the CPU and can operate at
the full speed of the chip (currently up to 450MHz). The K6-III will also
employ any cache storage in the chip set as a 'level 3' or L3 cache, hence
their touted Tri-Level Cache design.
The speed improvement can be quite dramatic (check our CPU
Comparator to see the difference). However, the performance increase
applies equally to both integer and floating-point operations. AMD
provided a real floating-point improvement with their Athlon
design.
