Intel "hyper-threading" to debut soon
The chipmaker's "hyper-threading" technology for boosting chip performance will debut soon with its first server versions of the Pentium 4 chip.
- Shankland covered the tech industry for more than 25 years and was a science writer for five years before that. He has deep expertise in microprocessors, digital photography, computer hardware and software, internet standards, web technology, and more.
Hyper-threading makes a single CPU act in some ways like two chips, but because there's still only one patch of high-speed cache memory and one connection to the rest of the computer, performance isn't the same as with a true dual-processor server.
Hyper-threading will appear later this quarter on two models of Intel's Xeon chip, a close relative of the Pentium 4 that has various improvements geared for high-end networked server computers. The first Xeon, for lower-end systems, is code-named Prestonia and works in two-processor servers; the second, code-named Foster MP, is for more powerful multiprocessor models, said Shannon Poulin, marketing manager for Intel's enterprise systems.
The chips will debut at the end of March, said Insight 64 analyst Nathan Brookwood, though they're already shipping for workstations and cost $615 for the 2.2GHz model in batches of 1,000.
The Xeon chips will be used in several new high-end Intel servers, including Egenera's BladeFrame, IBM's 16-processor Summit systems and Unisys' 32-processor ES7000 models.
Hyper-threading "is never as good as having an actual dual-processor system," and it's not even generally enabled in workstations, Brookwood said. But Intel hasn't overpromised the technology. "To Intel's credit, they tried very hard when they launched the capability to make it clear it wasn't the same as two processors."
Hyper-threading is one of the major differentiators between the Xeon and Pentium lines, though the Pentium 4 chip has most of the hyper-threading circuitry included but not enabled. It's also a feature intended to keep Advanced Micro Devices from making further inroads into the server market and to keep the pressure on Sun Microsystems for lower-end servers.
In the Pentium III generation, Xeons and Pentiums were less distinct and, indeed, the ordinary Pentium III continues to be popular in many lower-end server products.
Intel also announced an early access program to help software companies take advantage of the Xeon chips, said Melissa Laird, director of services and support for Intel's software and solutions group.
Under the program, customers can try out Xeon systems over the Internet and piggyback on Intel marketing efforts. They also can use Intel programming tools that take advantage of hyper-threading and the 144 new "NetBurst" instructions available to control Pentium 4 chips.
The two Xeons debuting in March come from different manufacturing processes. The lower-end Prestonia models are built with a more modern 130-nanometer (0.13 micron) process, which means smaller chip features and faster circuitry than the 180-nanometer process used to build the higher-end Foster MP models. Smaller process sizes enable chips to be built more cheaply and let more cache memory be squeezed on board.
Intel had planned a two-processor version of Foster for servers, but canceled it in 2001 because server customers prefer to avoid frequent changes in chip designs and the resulting system-testing headaches.
A 130-nanometer multiprocessor Xeon code-named Gallatin is slated for release in late 2002 or early 2003, Poulin said.
The Pentium III Xeon models will be phased out at the end of 2002, he added.