Curmudgeonly & Skeptical

Cecil replies:

Apples and oranges? Probably, but as a demonstration of the march of progress the explosion of computing speed since the 1960s is hard to beat. The power of a supercomputer is commonly measured in "flops," which stands for floating point operations per second. The Cray-1, the most famous early supercomputer (the first model was installed at Los Alamos National Laboratory in 1976), was capable of 133 megaflops (133 million flops). Early versions weighed over five tons, had a clock speed of 80 MHz, featured the equivalent of 8 MB of RAM, and cost about $9 million. In 1985 the Cray-2 was introduced, which could do 1.9 gigaflops (1.9 billion flops), operated at 244 MHz, had the equivalent of 2GB of RAM, and cost about $12 million. For comparison, a typical PC bought in 2000 or 2001 uses a Pentium 4 processor with a clock speed of 1.5 GHz, benchmarks at around 1.8 Gflops, probably cost under $2,000, and fits under your desk. In short, it's the rough equivalent of a 1985 supercomputer for one-six thousandth the cost.

Don't get smug. Your PC can accomplish only a pitiful fraction of what today's supercomputers can do. The current record holder is the Earth Simulator ultra-high-speed parallel vector computer installed at the Earth Simulator Center in Yokohama, Japan: it's tested at 35.86 teraflops (35.86 trillion flops) and has 10 TB (terabytes) of main memory. As has been true of cutting-edge computers for the past 60 years, the thing is huge, filling not just a room but a building. Whether its performance will be matched 15 years hence by something you can buy for $900 at Wal-Mart remains to be seen. But Gordon Moore, cofounder of Intel and originator in 1965 of Moore's Law (current formulation: computer power per square inch of microchip doubles every 18 months), says we won't haul up against the laws of physics, or at least the limits of wafer technology, until 2017.