Intel's CEO Paul Otellini said at the firm's fourth quarter earnings conference call that the first samples of Intel's discrete graphics product, the Larrabee, will be ready by the end of this year. Consumers can expect the first products based on the Larrabee in late 2009 or early 2010.
This puts Larrabee further out than I had expected, and makes it a possible companion to Intel's 32nm "Gesher" processor. A 32nm Larrabee that launches sometime after Gesher makes a certain amount of sense, since both products are multicore x86 parts where all the cores are situated on a high-bandwidth ring bus. Larrabee features 16 to 24 smaller, in-order cores, while Gesher has four to eight larger, out-of-order cores.
With Larrabee not arriving until 2010, this gives GPU makers quite a bit of time to catch up to multicore x86 in terms of programmability. And by programmability, I specifically mean the kind of improved exception handling and process switching capabilities that GPUs will need if they're going to compete with a multicore part like Larrabee for the coprocessor spots in commodity-based supercomputing clusters.
Supercomputing is one thing, but when it comes to graphics, the terrain is quite different. It's likely that when Larrabee launches, it will debut at or near the bottom of the performance heap in terms of existing game code. Intel will have to persuade game developers to build their engines around the new part, and that will take time. Indeed, not only will it take time, but there will have to be some kind of extraordinarily compelling eye-candy or physics-related reason for developers to make the jump.