, Draft Blog Entries

% fortune -ae paul murphy

Dempsey, Woodcrest, and the G5 Mac

Last week I reported on what I thought was a simple comparison between Apple's high end desktop competitive position two years ago and its position now. Unfortunately a lot of readers thought was I doing something else - either comparing system performance and ignorantly equating Intel's latest "Woodcrest" CPU line to some of its older Xeons or claiming some kind of price or performance parity between base configurations. I wasn't, and the response surprised me because I thought the comparison between Apple's competitive position then and now was pretty clear.

Guess not; but an assumption many of the discussants seemed to be making: that Intel's new core architecture represents a technology breakthrough justifying Apple's move to MacTel, seems worth following up on anyway.

The answer is that it's not - not a technology breakthrough, and not a justification for Apple's decision.

The Core products are generally faster and more power efficient than previous Intel product generations, but the surprise is that they're already obsolete. In fact, I think they represent little more than a slightly modified 2006 implementation at 65nm of Intel's 90nm based 2000/1 plans for 2003/4 - plans that were set aside in 2001 to support three alternative ideas: Itanium, HyperThreading, and megahertz marketing, all of which have since failed.

Itanium should have succeeded on merit, but missed its market window by five years or more, suffered from the loss of focus that went with HP's failed attempt to take over Compaq (Look at HP today, and what you see is Compaq with a printer division, not HP with a PC division), got hobbled by Microsoft's technical indifference or inability, and seems now to have fallen to Sun's strategy of using AMD to force Intel to meet demand for 64bit products via the x86, not the Itanium, architecture.

Hyperthreading was an idea that should have succeeded too - but the initial implementations weren't that good and the Microsoft community was no more ready for multi-threading than it was for Itanium. As a result this was an Intel innovation that even Anandtech failed to hyper-ventilate over. This, for example, is from their April 2002 review:

Enabling Hyper Threading turns one physical CPU into two logical CPUs, each with their own architectural states (including registers, program counters, etc...). So a single 2.2GHz Xeon with Hyper Threading enabled appears to be two CPUs under Windows 2000 Server. Enabling Hyper Threading boosts the performance of the single Xeon CPU by just over 12%, giving it the same transaction processing power as the single Athlon MP 2000+.

[Lots of omitted benchmark stuff after which the best he can say is:]

The other thing to take away from these tables is that Hyper Threading does have a great deal of potential. In our most stressful database server test, it improved performance by 17%. For a feature that takes up such a small percentage of the overall die size, it's not bad at all. These tests also show that the performance improvement of Hyper Threading is not always predictable. In the Web and Ad DB tests, the performance change varied from an increase of 3% to a decrease of 5%. You'll remember from our descriptions that both of these tests were heavy on the selects. The more varied benchmark was the Forums DB test and that's where we saw the largest overall improvement that Hyper Threading provided; it also happened to be the benchmark that the Xeons with Hyper Threading disabled did the worst in.

Hyperthreading failed, the Itanium seems to be on life support, power needs and competition from AMD brought realism to the megahertz madness, and now the new core micro-architecture is a back to the future move: something that originated almost co-temporaneously with AMD's Opteron but didn't make it to production until pulled off the design shelf in a mid 2005 panic for release about a year later. As a turn-around feat that's actually very impressive -but bear in mind that the situation was caused by Intel's failure to see where AMD would get by releasing their 64bit, dual core, x86 architecture machine on everybody's original schedule.

Right now the Opteron is getting long in the tooth and appears to be at a relative performance disadvantage -something that seems to have caused more than the usual blind rejoicing among Wintel fans with Anandtech, for example, deliriously proclaiming Intel a world technology leader once again.

They've got to know better - at the very least you'd think they could count the pins on the Socket F series - because Intel is getting most of its power and performance gains relative to Opteron from its ability to manufacture the part at 65nm; and when AMD matches that late this year they'll be a year ahead of Intel in banging up against the memory bottleneck again.

So is "woodcrest" today superior to the dual 3.4Ghz Xeons in the Dell Precision machine I compared the Mac Pro to in August of 2004? Sure, but it's hardly revolutionary and not here for the long haul.

Paul Murphy wrote and published The Unix Guide to Defenestration. Murphy is a 25-year veteran of the I.T. consulting industry, specializing in Unix and Unix-related management issues.