Thursday, October 30, 2008

Hot New Technology That Will Change Everything?

Everyone likes lists of stuff.  PCWorld has a list of 15 Hot New Technologies, one of them being massively multicore CPUs.  Missing from their write-up is some idea of what all these processors will do.  If we're talking two, maybe four, possibly even eight, then yeah, there's probably something for them to do at least part of the time.  But 32?  64?  More??

And new technology?  I guess the 16-processor Atari PC from the 80's doesn't count.  Nor do the dozens of other massively parallel machines that have gone down the tubes.  I guess a technology is new if it's failed for decades.

Instead of having more cores, why not small scale lithographs of Andy Kaufman on the chips?  They'll be at least as useful, more resistant to device failure, consume less power, and would be a great way of saying to consumers that "the joke is on you."

Sunday, October 19, 2008

Everything We Know Is Wrong

An interesting post over at the Economist, noting that published results can be wrong -- and that the more selective a venue is, and the "hotter" the area, the more likely it is that something is amiss. This makes a lot of sense; in complex research, things go wrong; there's no way we can eliminate all errors. If the errors skew things in a "positive" direction, reviewers are impressed, and off it goes.

To my knowledge, I've only had bad numbers in one paper -- we were measuring balance of horizontal and vertical wiring from different placement methods, and while the general trend was right, the relative percentages were off (lots of details on how this went wrong; we detected it after the conference camera-ready went in, so we announced the error at the actual talk, and included a note about the error in the journal version). Safe to assume that I've probably screwed up elsewhere.

Fortunately, I got tenure a few years ago, and can now take my sweet time with papers. I've got a few in the pipe, and I'm going to do my best to get everything right.