Wednesday, October 26, 2005

IBM releases details of G5 970MP

Yesterday, IBM's Norman Rohrer presented the details of their new 2.5 GHz dual-core G5 970MP at the Fall Processor Forum in San Jose.

It looks like they've gotten the power utilization down quite nicely in these chips, as the new 2.5 GHz puts out only 100 Watts maximum with both cores active. Estimates are that when the 2.5 GHz 970FX first came out it was somewhere in the range of 80-100 Watts, and that's for only a single core chip, with less cache, and also on a 90 nm process. The 2.5 GHz 50 Watt per core power spec for the 970MP likely means something like a 2 GHz low power single-core laptop G5 or a lower clocked dual-core laptop chip could be produced. In fact, Rohrer states that "low power, high performance" dual-core G5s will be available at 1.2 GHz and above. However, it is very possible that Apple won't release such a machine and will wait until dual-core Yonah or dual-core 64-bit Merom chips are available from Intel before moving away from the G4 in their laptops. As for the Power Macs, one suspects that Apple will wait until the Woodcrest Xeon chips are available before switching away from PowerPC. These dual-core chips are the server version of Merom, and are planned to have a maximum power utilization of about 80 Watts.

Rohrer also released SPEC CPU2000 scores yesterday for the 970MP. While the integer score of 1438 in mediocre, the floating point score of 2076 is quite impressive. Given the excellent fp scores, it's likely that the improvement isn't due only to the L2 cache boost. Other factors are probably also contibuting to the better fp performance (including for example possible improvements in IBM's XL compilers). In comparison, the current 2.26 GHz Pentium M 780 scores 1812 int, a significant improvement over the G5. However, Pentium M's fp performance suffers in comparison. A 2 GHz Pentium M Dothan has an fp score closer to 1100. Fortunately for Apple, Yonah and Merom are to get significant improvements in fp performance, according to Intel.


(Edited 2005-11-07 to correct Dothan fp score.)


Anonymous said...

> It looks like they've gotten the power utilization down quite nicely in these chips

When the 970MP was still a rumored chip the Application Note was partially leaked at and the summary mentionned some design enhancements. "Clock dithering (New feature for 970FX DD3.0, enhanced in 970MP)", "Additional Dynamic Power Management", yadda yadda. The information from the Master of dreams was really genuine.

> and also on a 90 nm process.

C'mon IBM, we need a 65nm chip just for the boasting rights. 90nm is so last week. :-)

Anonymous said...

Eug, are you aware that the SPEC FP2000 submission for a 2.26-GHz Pentium M has a score of 1375?

I believe that this is the highest score for the Pentium M.

Eug said...

You're right. I think I may have been mistakenly looking at Dothan's SPECint score. Dothan 2 GHz is closer to 1100 for SPECfp.

That's definitely not ├╝ber impressive, especially against the G5, but at least it's a major improvement over the G4 (which Pentium M will replace in laptops). Also, supposedly Yonah will get a nice fp performance boost. (No Macs will use Pentium M Dothan.)