NJ5 said:
@goddog: If the performance per cycle gets changed, we're not talking about simple die shrinks but redesigned chips (no matter how slightly). This is by definition. There are plenty of die shrinking and clockrate increases going on in all CPU families all the time, but the performance per cycle doesn't change if the transistor pattern is identical.
|
from a pdf paper on the 970, and 970 fx
Early PowerPC 970 chips were constructed with a 130 um and silicon-oninsulator
(SOI) manufacturing process with a die size of 121 mm2 [1]. The 970FX,
essentially version 2 of this chip, was updated and manufactured using a SOI and strained
silicon 90um process with a die size of 65mm2 [2]. Both versions of chip contain
approximately 58 million transistors [2]. The processor contains 32 64 bit general
purpose registers, 32 64 bit floating point registers, 32 128 vector registers for use with
the Altivec instructions (a SIMD multimedia instruction set), and a number of special use
registers [3]. Clock speeds for the 970FX currently run at speeds up to 2.5 Ghz [2].
IBM PowerPC 970FX RISC Microprocessor User’s Manual, V1.5
The major change, was production on the chip, with lower power drain, and less heat generation. Now without changing chip design reducing heat can improve performance through prediction paths being less likely to err. on these guides there is no listing f a major change outside of the production method. I do not know if the change in production method could cause this. but the FX did see a speed enhancment over the original.
on a sad side note none of the over 2ghz chips are still made, though the lower speed chips are. its kinda cool though that this family was a base for the PPE that was developed into both the cell, and the 360s chips, the highly customized powerpc chips that they are