Posted 10 July 2005 - 05:08 PM
Well, at this point in time, I think the folding it outpacing Moore's Law. However, that's only because the number of people running the folding is increasing, and their hardware is roughly increasing with Moore's Law, so the folding is staying ahead of Moore's Law by combining both increases.
As for what you can purchase for $50K, I think that what we should be concerned about is performance/price. You can get an AMD 2200+ for $200 or less, and you wouldn't need a monitor other than for setup. Just get a 16-port ethernet hub, some NAT software, an Internet connetion, and 16 cheap desktops, and just one monitor to hook up to one PC at a time to check in on ones that aren't working or need to be scanned for viruses or spyware or whatever. The whole setup should cost about $4K plus tax. Buying 16 PCs, you can probably get a bulk rate to get the price down even further, so you might be able to get this setup for $3K. Compare that to a high-end $3K system, which will definitely not have the same performance, and I'd be willing to bet not even half the performance.
For $50K, you can get literally hundreds of PCs, though at that point, you'd need a "real" WAN router, costing you a thousand bucks, within a factor of 3 (I haven't priced one in nearly two years, so I forget how much they cost), but you should be able to get a used one for fairly cheap.
Of course, all this presumes that you're spending all this money just to do folding. I prefer to let folding run in the background, so I can actually use my PC for games, programming, etc., some fraction of the time. I don't have a particular use for 16 PCs, let alone hundreds, so I wouldn't buy such as setup even if I had the money.
Once distributed folding participation peaks and starts declining, it'll follow more in line with Moore's Law. In other words, in about five to six years, it'll be going about 10 times faster, assuming a doubling every 18 to 20 months. In ten to twelve years, it'll be going 100 times faster. And that's just the distributed folding. Dedicated supercomputers can outpace Moore's Law through a variety of techniques, such as just using more processors, better software, etc. Just as the price to sequence a genome is falling much more rapidly than Moore's Law, the effectiveness of protein folding will also outpace Moore's Law, at least for the foreseeable couple decades. During that time, protein folding should not only get faster, but more accurate.
