Gallatin Pentium 4 Xeon MPs Ready To Roll

by Timothy Prickett Morgan

The second time seems to be the charm for Intel these days. The first generation "Merced" 64-bit Itaniums and the first generation "Foster" Pentium 4 Xeon MPs left something do be desired and they did not sell well. A few months ago, Intel launched the "McKinley" Itanium 2s, making good on the performance promises it made for Merced, and this week, Intel launched the "Gallatin" kickers to the Fosters, which are going to be snapped up by high-end server makers.

While lots of PR and spin has been focused on the 64-bit Itaniums and how Intel and its partners (particularly Hewlett-Packard) are in it for the long haul in pushing a new architecture that breaks free of the confines of the old X86 chip architecture that is still at the heart of the Pentium 4 and Pentium 4 Xeon processors, the fact remains that the 32-bit processors, when implemented with some clever engineering and advanced chip making process technologies, are more than enough CPU for all but the largest customers. This is why Gallatin chips are important. Companies who want to keep their applications without having to recompile them or suffer an emulation penalty want faster 32-bit, X86-compatible processors--and that goes double for software houses.

The Gallatins were initially expected in early 2002, with the Fosters originally expected around mid-2001. The Fosters were delayed time and again, and finally shipped in March 2002. The delay in the Fosters pushed the Gallatins out to the end of 2002 or early 2003. To some ways of thinking, Intel is doing better with its 0.13 micron process and it has been able to move up the Gallatin announcements by several months compared to the most pessimistic expected launch date for the chips. This is progress. Unfortunately for Intel, this is also a rotten economy. And that is why Intel has held the price on the Gallatins current with the Foster Xeon MPs and the Pentium III Xeons (which come in "Tanner" and "Cascades" flavors) that they replace as the high-end, 32-bit processor for four-way and larger Wintel and Lintel servers.

The Gallatins come in three speeds, and like the Xeon MPs and Pentium III Xeons, offer different cache sizes for a number of different clock speeds. The idea is to provide larger caches to the faster chips to allow them to actually make use of those extra clock cycles to do real work. Without the larger caches, the chip can't do as much work. The Xeon MPs had much higher clock speeds than the Pentium III Xeons, but had smaller L2/L3 caches than the Pentium III Xeon L2 caches, and they therefore suffered in performance comparisons. The Xeon MPs didn't look like they offered all that much more performance, and therefore server vendors (the notable exception being IBM, with its xSeries 440) gave them lip service and have by and large been waiting for Gallatin to announced revamped servers to replace their Pentium III Xeon product lines. The Gallatin chips, like the "Prestonia" Pentium 4 Xeon DP chips used in dual processor servers and workstations, support Intel's version of simultaneous multithreading, which it calls HyperThreading. This HyperThreading support, which debuted with the Prestonias earlier this year, uses electronic tricks to present two virtual chips to the operating system, and it can boost performance on some workloads by close to 30 percent.

The table below compares the Gallatins to the Foster Xeon MPs and Tanner Pentium III Xeons. The relative OLTP performance figures are based on Intel data and our own internal estimates. These estimates assume that HyperThreading is turned on.

As you can see from the table above, at least at the processor level, the Gallatins offer significantly better value than the Fosters did compared to the Tanners. However, the vendors who said that the Fosters were not a good value overstate their hands. It might not have been profitable to have to re-engineer and re-certify their machines to use the Fosters in the down economy of last year and early this year, to be sure. But the processors were, by our reckoning, a decent chip. That said, the Gallatins offer a lot more performance for the same money, and they are the kind of chip that server makers were expecting when they started planning for Foster.

Sponsored By ACUCORP

Acucorp is a leading developer of application extension solutions running on over 600 platforms such as Linux. These extend5 solutions include a powerful ANSI COBOL compiler, an integrated development environment, web deployment technology, seamless interfaces to RDBMS, COBOL-based GUI development, distributed processing and client/server technology. For more information, visit www.acucorp.com