Sizing Up Multicore Processors

Sizing Up Multicore Processors

By Tom R. HalfhilltomH.jpg
If two processor cores are better than one, can we evaluate multicore chips simply by counting cores? Unfortunately, we can’t. That would be as misleading as judging single-core processors solely by their clock speeds, and everybody knows how that turned out.

All multicore processors are not created equal. They have profound differences that greatly affect their performance. Before the marketing schemers at AMD and Intel corrupt your mind with propaganda, consider some important aspects of multicore designs.

One factor is the microarchitecture of the cores replicated on the chip. For now, AMD and Intel are using processor cores originally designed for single-core chips, because that’s all they’ve got. Multicore PC processors are so new that neither company has had time to create entirely new cores. I think Intel has a slight edge, because its latest multicore processor (Yonah, expected in January) has two cores derived from the low-power Pentium M (Banias) microarchitecture. AMD’s cores aren’t quite as power-optimized.

Another factor is whether the multiple cores are homogenous or heterogeneous. Homogenous means all the cores are the same, and so far, that describes all the multicore chips announced by AMD and Intel. Someday we may see heterogeneous designs with different cores optimized for different software workloads.

Another very important factor is how the cores communicate with each other. There are numerous options. AMD’s first dual-core processor, the Opteron 800, has a sophisticated on-chip link called a crossbar switch. In comparison, Intel’s first dual-core processor, the Pentium EE 840, looks like a duct-tape job. Under pressure to match AMD’s multicore introduction, Intel crammed two Prescott Pentium 4 cores on the chip and routed their communications through the external chipset. Intel’s Yonah is a much better multicore design.

Cache matters, too. Because AMD and Intel derived their first multicore processors from single-core chips, each core has its own L2 cache. Like jealous siblings, they don’t share. Yonah takes a step forward by using a shared cache, which is more efficient when one program needs more cache than another does.

There are many more differences among multicore processors, and they always reflect engineering trade-offs—there’s no absolute best design. Weighing those trade-offs will require better benchmarks and knowledgeable reviewers.



+ Add a Comment

Log in to MaximumPC directly or log in using Facebook

Forgot your username or password?
Click here for help.

Login with Facebook
Log in using Facebook to share comments and articles easily with your Facebook feed.