Hail Montecito

Another day, another Intel CPU code-name

IDF So the year is 2004 and Intel launches Montecito, its next-next-next generation 64-bit server CPU. That's the code name and that's the plan for the latest processor to appear on the company's roadmap, announced yesterday by Mike Fister, Intel's top server guy, at IDF.

The CPU is to be built on a 90nm process and it will mark an 'evolution' of Intel' 64-bit architecture, with "amazing things on the die". Sounds like a different core from McKinley, but that's all Intel is prepared to tell us right now. This is not so surprising, as the company has to launch first McKinley, the next version of Itanium, sometime this summer; and second, Madison and Deerfield (the dual processor version), in 2003. Here's another teaser supplied by Fister in his keynote presentation - Madison will have up to 6MB on-die L3 cache and will be pin-compatible with McKinley.

Fister was, as one might expect far more forthcoming about the company's latest server CPU launch, the new Xeon, the first 32-bit chip in its portfolio to incorporate hyperthreading, a technology in which one CPU behaves like two to deliver perhaps a 30 per cent performance boost. (Incidentally, Intel CEO Craig Barrett showed off a 3GHz P4 desktop in demo yesterday. This showed a 3D box filled with three broadband streams - implying to us that hyperthreading will make its way to the client side in the not so very far away future). Add in Intel's Netburst fat bandwidth architecture, featured for the first time on Xeon, then turf in the company's spanking new Xeon chipset, the E7500 (formerly codenamed Plumas for those of you who care about such things) and you get something that performs a whole heap faster than its PIII-based antecedents. Fister sought to show just how much faster - with a colleague conducting a demo in real-time, by way of online links to two server racks. One was based on 1GHz PIII Xeons and one was built around 2.2GHz P4 Xeon sisters. As is the way of such things, the link to one rack broke down - but the graph on the overhead project would have shown an 80 per cent speed increase for the 2.2s, Fister told us.

That's the problem with server demos - running gazillions of transactions in the background does not make for compelling onstage visuals. But do not let that stand in the way of IBM, which yesterday show off a McKinley beast working with an new new Xeon P4 server box. The demo featured a simulated ecommerce benchmark in action, called ECPerf, and it was, we were informed, suitably blindingly fast.

It is clear though, that Intel, has - at last - with McKinley some serious armoury in its assault on the proprietary Unix server market. Three trends in the server market are all swinging the pendulum more and more in Intel's direction, according to the Fister worldview.

The first trend is segmentation: blade servers, pedestal servers, carrier class beasts... the list goes on. Increased segmentation places strain on the manufacturer - no-one can do everything on their own, says Fister. You have to pick the right alliances, make the right outsourcing decisions. The inference is that the proprietary players find this more difficult to achieve than Intel.

The second trend sees enterprise customers voting with their feet. The proprietary Unix market has "been in decline for five years", and nothing is going to reverse this, Fister says.

And trend three, Fister terms as "scale up by scaling out". Arrange your boxes into clusters and you get some bloody big machines pumping out some awesome TPC-C benchmarks. In other words, you don't have to buy really Big Iron to have a really big server.

And Intel can produce server chips cheaper than its rivals: it has volume manufacturing economies of scale, it has the R&D budget to make serious improvements, and it has already succeeded in knocking out HP and Compaq as competitors, replacing them, although not quite yet, as Intel-only enterprise server customers.

That leaves Sun and IBM as serious competitors for the long-term Intel is today gunning for Sun. Fister in his presentation, referred to a customer comparison between Sun's (discontinued!!!) E450 platform and various Intel server CPUs. The company's database application ran twice as fast on Xeon, ten times faster on ported code to Xeon, 16 times faster on uncompiled code on Mckinley, and 20x faster when recompiled. The battle is on. ®

Sponsored: Minds Mastering Machines - Call for papers now open

Biting the hand that feeds IT © 1998–2018