[mduell]

There's much talk about the dualcore chips from IBM and Intel, so I thought I'd post a summary.

Power consumption
PPC970MP 2.5Ghz: 100W
Pentium D 2.8Ghz: 95W
Pentium D 3.2Ghz: 130W

SPECint
PPC970MP 2.5Ghz: 1423
Pentium D 2.8Ghz: 1331
Pentium D 3.2Ghz: 1490

SPECfp
PPC970MP 2.5Ghz: 2076
Pentium D 2.8Ghz: 1596
Pentium D 3.2Ghz: 1766

Performance per watt
(Sum of SPEC scores divided by maximum power consumption)
PPC970MP 2.5Ghz: 35
Pentium D 2.8Ghz: 31
Pentium D 3.2Ghz: 25

If anyone can find figures for PPC970MP at 2.0 or 2.3Ghz, please post them (and your source).

Sources:
Standard Performance Evaluation Corporation
Intel Processor Spec Finder
Norman Rohrer (one of IBM's top chip designers)

[Catfish_Man]

I find the 2000+ SPECfp number for the MP rather suspicious, but ignoring that, quite impressive results I think it's pretty clear why NetBurst (the P4 architecture) is getting the axe.

[mduell]

Originally Posted by Catfish_Man

I find the 2000+ SPECfp number for the MP rather suspicious, but ignoring that, quite impressive results I think it's pretty clear why NetBurst (the P4 architecture) is getting the axe.

Don't be so modest, I think it's outright bogus (compare to the single-core result of 1241 at 2.2Ghz). But without the actual benchmark results I'm going to take Norman's word for it.

I was surprised that the PPC970MP and Pentium D power consumption was so close; NetBurst is supposed to be "soooo hot, toooo hot," but the G5 is right up there with it.

[Mithras]

How about an Opteron comparison?

[mduell]

Originally Posted by Mithras

How about an Opteron comparison?

I thought about it, but I can't find any SPEC benchmarks for their dualcore chips. I may add the results with their single core chips (since it should be the same result with a single threaded benchmark).

[Todd Madson]

Originally Posted by mduell

There's much talk about the dualcore chips from IBM and Intel, so I thought I'd post a summary.

Power consumption
PPC970MP 2.5Ghz: 100W
Pentium D 2.8Ghz: 95W
Pentium D 3.2Ghz: 130W

SPECint
PPC970MP 2.5Ghz: 1423
Pentium D 2.8Ghz: 1331
Pentium D 3.2Ghz: 1490

SPECfp
PPC970MP 2.5Ghz: 2076
Pentium D 2.8Ghz: 1596
Pentium D 3.2Ghz: 1766

Performance per watt
(Sum of SPEC scores divided by maximum power consumption)
PPC970MP 2.5Ghz: 35
Pentium D 2.8Ghz: 31
Pentium D 3.2Ghz: 25

If anyone can find figures for PPC970MP at 2.0 or 2.3Ghz, please post them (and your source).

Sources:
Standard Performance Evaluation Corporation
Intel Processor Spec Finder
Norman Rohrer (one of IBM's top chip designers)

I'd be curious to see how these match up with the previous FX - what's the
comparison?

[mduell]

Originally Posted by Todd Madson

I'd be curious to see how these match up with the previous FX - what's the
comparison?

SPECint is up about 25% and SPECfp is up 50%. As the other poster said, the SPECfp increase is huge and possibly erraneous.
These SPEC benchmarks are single-threaded benchmarks for speed, so they only use one CPU core; there is a different SPEC benchmark that is multi-threaded for throughput (uses all CPU cores), but IBM hasn't released any PPC970MP numbers for it.

[Eug Wanker]

The numbers for SPEC are what Norman Rohrer said. He's the IBM bigwig design guy for 970MP.

1) New compilers probably. They're on version 8 for XL now.
2) The extra cache will help a lot.
3) It's possible things have improved with the new chipset.