Going Faster, Sideways

I <3 Ars…

Now, however, we’ve entered a period of system integration for integration’s sake, because the clockspeed increases have slowed down and integration is what we have left. This new era of “integration first, then clockspeed” is one for which the industry is relatively unprepared. In particular, it’s now much harder to sort out what should happen next.

One artifact of the previous decade’s focus on steady processor clockspeed increases is that everyone got used to knowing what the future of computing would look like. Things would be much as they are, we figured, except processors would run faster, and programs would run faster as well. And for the most part, this view of the future held true right up to the moment where the semiconductor industry hit the power wall and the clockspeed party ended relatively abruptly.

Now the future of computing is substantially murkier, and both industry and academia are still in the midst of a collective “now
what?” moment. Everyone is clear on one fact: the future is less about clockspeed and more about die-level integration. But beyond that, every aspect of the way that computer systems are built and programmed is up for grabs.

Beyond the Teraflops: Why Intel really put 80 cores on a single chip, Ars Technica

I’ll write something later…