Talk: PowerPC

From open-encyclopedia.com - the free encyclopedia.

I removed the sentence "One other major problem was that the 88000 (and all earlier Motorola designs) were big-endian, whereas the POWER was little-endian." because I believe it is incorrect. Evidence:

p. 234 of The PowerPC Architecture: A Specification for A New Family of RISC Processors (Second Edition, Morgan Kaufman Publishers, 1994.) indicates that the RS/6000 is a big endian architecture.

The archives of comp.arch contain many references to AIX and RS/6000 being big-endian.

I also re-wrote part of the history section, following the chronology presented in the Forward of The PowerPC Architecture (ibid.), written by Phil Hester of IBM. This appears to be a more reliable source than that of the "so the story goes" version which I replaced.

I removed claims that Apple is/was Motorola's "largest CPU customer". That is very unlikely to be true in terms of volume shipped (consider the embedded market).

Also corrected comment w.r.t. endian switch requiring a reboot. It doesn't. Added expository text on PPC endian modes in linked article. Removed reference to microcode, as it is either ambiguous or incorrect. --Kday

Some suggested changes to the PowerPC article

>>>>The 970 is a 64-bit processor derived from the POWER4 server processor. To create it, the POWER4 core was modified to be backwards-compatible with 32-bit PowerPC processors, and a vector unit (similar to the AltiVec extensions in Motorola's 74xx series) was added.

It probably should also mention the new PowerPC 970FX, which debuted in 2004. The major changes were to switch from a 130nm process to a 90nm process (which allowed the clock rate to be boosted from 1.6GHz to 2.2GHz), and to reduce power consumption from 2.5V to only 1.8V. Oh yes, and the FSB was increased as well (half the clock rate). It's also worth noting that IBM married the PPC 970/970FX chips with AMD's HyperTransport Tunnel in the BladeCenter JS20 blade servers. Interesting combination.

>>>> * 970 (PowerPC G5) (2003) 64-bit implementation derived from the IBM POWER4 enhanced with VMX (AltiVec compatible SIMD extensions), at speeds 1.4 GHz, 1.6 GHz, 1.8 GHz, 2.0 GHz and 2.5 GHz

I don't believe 1.8/2.0/2.5GHz versions of the 970 were ever shipped (although I could be wrong). I think it stops at 1.6GHz (maybe 1.8). Also, my understanding is that VMX isn't merely "AltiVec-compatible". VMX, AltiVec and Velocity Engine are the same thing, just marketed under different names by the three companies.

You might add another line for the 970FX, such as:

>>>> * 970FX (PowerPC G5) (2004) 90nm version of the 970, at speeds 1.8 GHz, 2.0 GHz, 2.2 GHz*, 2.5 GHz

Note the addition of a 2.2 GHz speed. This is the chip used in the current IBM JS20 blade.

Take care.

Mark (mchapman@us.ibm.com)