CHIPMAKER Intel said its Xeon Phi accelerator cards are not just rebadged Larrabee designs, although they were inspired by the graphics processor architecture that Intel unceremoniously canned in 2009.
Intel's Xeon Phi accelerator cards and its Many Integrated Core (MIC) architecture have often been compared to Larrabee, which the firm demonstrated and then shut down in 2009. Rajeeb Hazra, VP of Intel's Architecture Group told The INQUIRER that the MIC architecture grew out of a multi-decade Intel research initiative and that it incorporates lessons learned while developing Larrabee.
Hazra said, "It is certainly inspired by many of the things we learned while doing Larrabee, but its heritage goes back before Larrabee as it was inspired by a large amount of research done in the labs in Intel ever since the late 1990s on what many core architectures look like. Even though the world at that point was just about entering multiple core, the labs are chartered to look well beyond. They were starting to look at what multi-core systems look like, what should cores look like, what should the interconnect look like, how do you evolve things like cache coherency and other models."
Hazra admitted that when Larrabee's future as a graphics accelerator was foreclosed, the firm turned its focus to making use of at least part of Larrabee in other markets. "Larrabee came as a product aspiration out of that research. When we didn't continue with Larrabee we saw the wonderful potential for that kind of somewhat mature approach to many core to become very useful for high performance and technical compute, and that's where we took it. It's not just Larrabee, it's inspired and learned from and multiple innovations both in the hardware and the software stacks that run on it to make it the best possible product for technical compute."
When Intel showed off Larrabee, all the talk was of whether Intel could finally compete with AMD and Nvidia, but as both firms claim, graphics technology is hard to master. AMD in particular likes to claim that graphics intelectual property is not just expensive but time consuming to develop. After all, AMD had to buy ATI in order to get its hands on advanced graphics technology.
Hazra also nodded at Intel's process capabilities as being one of the factors in making MIC and eventually Xeon Phi a viable prospect. It could well have been that Intel decided that Larrabee as a graphics product would not be commercially viable to produce at the 40 to 60 per cent product margins Intel likes to have, and by that stage its graphics performance would not be competitive against AMD and Nvidia. Instead the firm stuck to what it does best, turn it into an multi-core x86 chip and wait for the manufacturing side of its business to make it a commercially viable product. µ