Morn said:
The Core Duo is based on the Pentium Pro which was out in 1995. Although it's been tweaked an awful lot since then.
The G4 is just a G3 with an added altivec unit. And the G5 is very similar with tweaks and additions, like an extra FPU unit.
The newest microarchitectures around today, the Athlon. And the Pentium 4 which Intel is giving up. New is not necessarily better.
And the G5 is faster than the Core Duo. But the G4 was even slower than the Pentium M, Freescale was not able to keep up with Intel.
The G5 is almost totally unrelated to the G4. All it shares is the instruction set (including Altivec). Different number and types of execution units, different pipeline lengths, cacheline size, out of order depth, number of rename registers, dispatch width, bus type and speed, basically everything. The G5 is descended from the POWER4 line, not the G3 line.
One thing I've been wondering is this:
Why does the 970MP (dual core G5) score SO much better on SPECcpu (which is single threaded) than the 970FX does?
I believe that the key point is that it used version 8 of IBM's compiler, rather than version 6 as the 970FX did. XlC 8 can automatically multithread and vectorize programs in certain situations, so what claims to be a single threaded non-vector benchmark may have ended up being a dual threaded vector benchmark, at least to a certain extent, allowing the 970MP to use some of its second core to help the first core out.
For reference:
2.5GHz 970MP SPECint: 1428
2.5GHz 970MP SPECfp: 2076
2.2GHz 970FX SPECint:1040
2.2GHz 970FX SPECfp: 1241
2.26GHz Pentium-M SPECint: 1839
2.26GHz Pentium-M SPECfp: 1375
note: Pentium-M here refers to Dothan, which is the predecessor to the Core Duo (Yonah).
Sources:
http://www.aceshardware.com/SPECmine/
http://www.heise.de/ct/05/24/018/
http://www.llnl.gov/asci/platforms/bluegene/papers/10mendell.pdf
<edit>
in response to the 64 bit posts: the main point of 64 bit is to allow for more than 4GB of ram.
</edit>