Intel opened the kimono a bit on its future Larrabee chip this week, and much commentary has ensued. The early consensus is that Larrabee will (at least at first) be a neither-fish-nor-foul failure, although Ars Technica seems to disagree. But even if the consensus is correct, the Larrabee discussion highlights some points well worth reflecting on.
1. Users are of two minds as to what they think about interfaces. On the one hand, they want the best, snazziest, prettiest, graphics and animation. On the other hand, they seem happy with iPhone-sized web UIs, as per the much-bandied figure that iPhone users consume 30X the data other cellular users do. And of course …
2. Users who want great graphics need nontrivial client machines.
3. As Ars Technica points out, to use Larrabee for close to maximum performance, you need optimization of or for your specific program, probably with the help of Intel engineers. But that’s not just Larrabee; it’s also true of Intel CPUs today. In general, I believe small software companies are a greater threat to large ones than many people recognize. But being a priority for Intel to partner with is one of the nice advantages larger companies enjoy.
4. High-performance computing guru Bill Blake (then at Netezza, now at Interactive Supercomputing) once claimed to me that game chips would take over general-purpose computing from current styles of CPUs (presumably in no small part because they would be cost-effectively manufactured in higher volumes for the consumer market). Larrabee can be seen as a marker in that direction.
Related links
- Peter Glaskowsky calls Larrabee a science project that got out of hand
- Hacking Cough thinks Larrabee will fail, but its successors (from Intel) will eventually win
- AnandTech offers a very detailed Larrabee analysis (which looks great, but I must confess to not having read all the way through)
I recently wrote about similar considerations in connection with Sun’s Rock chip, and a great comment thread ensued.
