X

Intel to add memory controllers, threading in 2008

Nehalem processors will use hyperthreading tech found in the old Pentium 4 chips and some of AMD's design tactics.

Tom Krazit Former Staff writer, CNET News
Tom Krazit writes about the ever-expanding world of Google, as the most prominent company on the Internet defends its search juggernaut while expanding into nearly anything it thinks possible. He has previously written about Apple, the traditional PC industry, and chip companies. E-mail Tom.
Tom Krazit
3 min read
SAN FRANCISCO--Intel on Wednesday confirmed plans to integrate critical system components and reintroduce hyperthreading technology in 2008, when it unveils a new chip blueprint.

In the past, Intel executives have spoken in broad terms about integrating components such as the memory controller and direct links between processor cores. But those technologies are on tap for Nehalem, the code name Intel has assigned to a chip family it will start producing in 2008, Pat Gelsinger, senior vice president and general manager of Intel's Digital Enterprise Group, said during a briefing for reporters here.

"We view (Nehalem) as the first dynamically scalable microarchitecture," Gelsinger said. What he means by that is that Intel chip designers will be able to pick and choose from a wide variety of ingredients to build chips for different types of computers, from powerful servers to small notebooks.

Chips based on Nehalem will have between one and eight cores, and will be capable of handling two independent software threads per core. Hyperthreading, Intel's name for the concept, allows a processor to execute two different code streams at pretty much the same time. This was a feature found in Intel's single-core Pentium 4 processors but largely discontinued with the advent of multicore chips.

Intel also plans to build chips with "point-to-point" links that directly connect processor cores with their neighbors, and install a fast link between the processor and memory with integrated memory controllers, Gelsinger said. Those were two design philosophies used by Advanced Micro Devices to break into the server market with its Opteron chip in 2003. Intel has thus far disdained those approaches.

Not every chip in the Nehalem family will come with all those features, but they will be available to Intel's designers, Gelsinger said. Different customers require various types of products for their future PC and server designs, he said.

For example, Intel's newest best friend, Apple, is pushing the chipmaker to develop chips with a lot of integrated pieces, including graphics controllers, Gelsinger said. Apple's focus on industrial design means it's looking at building ever-smaller systems, and integration is one way to accomplish that, he said.

But other customers building PCs for businesses or gaming have different requirements. And some server companies may want an eight-core processor capable of handling 16 threads, while others need a low-power chip for blade servers. "What used to be chipset variations with common cores will be standard chipsets with uncommon cores," he said.

Intel will start producing Nehalem processors--the actual brand has yet to be revealed--in 2008 using its 45-nanometer manufacturing technology. But before then, it plans to introduce Penryn chips as the first processors to use that new manufacturing technology.

Penryn chips will come in several different core combinations, just like the current generation of Core chips. Desktops and servers will have their choice of dual-core or quad-core chips that will run faster than 3GHz, Gelsinger said. Notebook chips will continue to have two cores.

Intel anticipates that the Penryn chips will deliver a 20 percent boost in gaming performance over its current generation of Core 2 Duo processors, and a 45 percent improvement in the performance of media applications. The improvements come from new instructions, such as the SSE4 extensions to the x86 instruction set, as well as larger caches and faster front-side bus speeds.