This is getting seriously off-topic for alternative energy (or maybe not!)...
But a techno-Geek Vortician sent me info about the availability NOW of teraflop desktop supercomputers and servers. If you have the buck$ for a parked Beamer, say, but would rather have a 24/7 internet screamer, go for it! Actually the price of entry into terabyte computing has dropped in the past 4 years from $25 million(minimum) to less than $5,000, and will likely continue to exceed Moore's Law for a while. Nevertheless, any of us can still afford to wait a few years, since the best use would be full speech recognition with parsing (as opposed to voice-to-text only) and this is not ready yet. A few months ago, NVIDIA, a company noted for graphics processing, not CPUs - figured out a way to combine multithreaded parallel graphics chips to do some incredibly powerful and versatile computing- of the very same kind which the human brain also does best. Their well-named "Tesla" processors can be linked together as blade servers. The Tesla S870, has a retail price of $12,000, and will be packed with four x8 series GPUs and will do 500 gigaflops per GPU. With this server, you get 2 teraflops for not all that much, but in two years, when the novelty has worn off, look for the same thing for under $2,500. Marvin Minsky, at one time claimed that the human brain is a one teraflop-equivalent analog computer, but he caught so much flak from Roger Penrose and others, that he raised his estimate (so as not to be too embarrassing to humans?). Something similar happened when the "horsepower" was designated as a comparative unit. I copied a Wiki entry on James Watt and the naming of the HP below, mainly for its historical value to word-phreaks. However, I agree with those in AI who opine that a doctorate level of human brain-power will likely require 10 teraflops, once the necessary software is available. This fall, a graphics card for any computer, called the Tesla C870 will cranks out 500 gigaflops will sell for $1499. But as we all continue to lament (especially those of us with deficient typing and spelling skills), the best use for this kind of computing power, outside of acedemia, would be to dispense with the keyboard altogether; yet speech, or even accurate voice recognition, is not perfected, and parsing the words into true actionable "meaning" is even further away. Jones History of the "horsepower" (paraphrased from Wiki) ... straight from the 'horse's mouth', so to speak. The term "horsepower" was coined by James Watt to help market his improved steam engine. He had previously agreed to take royalties of one third of the savings in coal for this engine, but that scheme did not work with customers who used horses instead. Watt determined that a horse could turn a mill wheel 144 times in an hour. The wheel was 12 feet in radius, therefore the horse travelled 2.4 × 2π × 12 feet in one minute. Watt judged that the horse could pull with a force of 180 pounds. This all was rounded to 33,000 ft·lbf/min. "Engineering in History" recounts that Smeaton estimated that an average horse could produce 22,916-foot-pounds per minute over time. Desaguliers increased that number, but Watt standardized the figure at 33,000. Put into perspective, a healthy human can produce about 1.2 hp briefly, in a sprint - and sustain about 0.1 hp indefinitely; and trained athletes can manage up to about 0.3 horsepower for a period of several hours. Most observers familiar with horses estimate that Watt was intentionally optimistic and wanted to "over deliver" with his replacement; and that few horses can maintain the one HP effort for long. Regardless, comparisons of machines to horses proved to be an enduring marketing tool.
