At first glance, Larrabee seems to come out on the winning end of comparisons with its direct competitors. More parallel than the Cell and more flexible than a GPU, it seems to have everything going for it. But let’s keep our excitement in check. A product that exists only on paper necessarily has all qualities and no defects. For a long time, the Itanium seemed to be the future of processors before one of its flagrant faults became painfully obvious: while it’s not easy to reorganize a program’s instructions dynamically on the processor, it’s just as complicated to do it statically in the compiler.
So it’s not a good idea to blindly accept each announcement of a new technology. In certain forums, for each question concerning rendering or a particular algorithm, somebody will eventually mention Larrabee as the ideal solution, which is ridiculous for the time being. Larrabee won’t suddenly solve all problems of real-time 3D, though it should make some good progress possible.
Larrabee undeniably has the benefit of positive hype, and Intel needs to take care that that doesn’t result in blowback if the first generation of products doesn’t live up to expectations (the Merced syndrome).
In a few days, we’ll be giving you a look at Larrabee’s thorny software side, so stay tuned.