Are companies like IBM and Intel "cheating" in using multichip module (MCM) packages to create the latest advancements in multicore processors? Or is insisting that those next-generation devices be manufactured using a single monolithic design such as those by Advanced Micro Devices just gamesmanship?When InformationWeek published a story on Wednesday detailing how IBM had extended its "quad-core" Power 5+ processor-based server line to address wider markets, including entry-level deployments most often associated with x86 processors, some readers where quick to point out that IBM didn't have a "real" quad-core design for its processor.
"IBM has been claiming achievements beyond reality," wrote one reader. "Aggressive skepticism of vendor claims is a necessary component of good IT reporting. Simply repeating what a vendor says is the role of a vendor-hired PR firm, not the press... (The Power 5+) is a module with two dual-core chips, not a quad-core chip."
"A [MCM] acts like a coherent unit, but in reality is just four discrete chips," wrote another reader. "IBM and Intel are not in the same league of sophistication as AMD. AMD has by far the more advanced of these solutions."
This issue came up last year when Intel introduced its first dual-core processors as well. For those first offerings, Intel combined two single-core processors in a single package to create its offering, while AMD's first dual-core offerings were indeed single monolithic designs, or what AMD likes to call a "native" multicore implementation.
And Intel has already announced it will use the MCM approach again later this year when it introduces its first quad-core processors. Those devices will actually combine two existing dual-core processors inside a single MCM.
Analysts Dean McCarron and Nathan Brookwood say multicore processor designs that utilize MCMs are no less "true" multicore offerings than the monolithic, or "native," designs from AMD.
"What it comes down to is there is a processor socket, and a single processor package plugs into that socket," McCarron of Mercury Research says. "Then you can count the number of processors in the device to determine the total number cores. I think it's a bit of gamesmanship to say one approach isn't valid."
"This isn't a computer science project where the guy who does the more elegant design gets extra credit," Brookwood of Insight 64 says. "If the chip has four cores, you can say it's a quad-core design."
That's not to say there can't be performance advantages to one approach over another, or other manufacturing considerations that come into the fray.
Brookwood says Intel's upcoming quad-core design amounts to "jamming two processors together" without much thought to such important features as the ability of the cores to share memory resources.
McCarron points out that using an MCM approach will allow Intel to get to the market with its quad-core design faster than AMD, as well as provide Intel with up-front manufacturing savings. With the vast majority of Intel's manufacturing line already dedicated to producing dual-core processors, Intel can get higher yields of usable processors, and therefore higher margins, by using two dual-core processors in a single package.
AMD on the other hand continues to try and demonstrate the technological advantages of its processor designs over Intel as a way to further increase its market share versus the much larger Intel. Although AMD's quad-core processor is expected to be six months or more behind Intel to market, it could likely be able to create a higher-performance alternative by taking advantage of the tighter interconnect capabilities of a monolithic design.
Intel understands the improved performance capabilities a monolithic design can bring to a processor. That's why after introducing its first MCM-enabled dual-core processors, Intel later followed up with an enhanced monolithic dual-core that's at the heart of its mainstream product line today.
Intel will use the same approach with its first quad-core design. After getting into the market quickly and efficiently with its MCM-enabled quad-core design later this year, Intel will eventually move to a monolithic quad-core offering, probably later next year as the company transitions to the smaller and more advanced 45-nanometer manufacturing process it demonstrated earlier this year. That process will shrink the size of the transistors on the chip from the 65-nanometer process currently used, providing Intel with more flexibility in creating a monolithic quad-core processor.
And there are already some pretty advanced monolithic multicore processors on the market, like Azul Systems' 24-core Vega processor and Sun Microsystems' eight-core UltraSparc T1 (code-named Niagara).
Azul has already announced plans to push its Vega chip to 48 cores next year. And Sun recently disclosed that its UltraSparc T1, which currently has eight cores that each operate with four independent threads to provide a total of 32 processing elements, will be updated with Niagara 2 to have eight cores with eight independent threads, providing a total of 64 processing elements in a single chip.
"Companies take different approaches because of what value they think it can bring to the end product," McCarron says. "There can be a lot of variables in how the cores are attached. Customers will continue to look for all the usual things: performance, value, and reliability."