Twice a year, a list of the "fastest" supercomputers is released. This list is called the Top 500 list, and it provides vendors and their clients with a chance to pat themselves on the back. In the latest Top 500 list, our local supercomputer, Turing, made it on the list at 66. Other machines of interest are the fastest machine in the world, a Blue Gene machine (made by IBM) at Lawrence Livermore National Lab in California, and the Earth Simulator in Japan, which reigned at the number one spot on the list for nearly three years before being displaced by the Blue Gene phenomenon. The Earth Simulator comes in at number four on this list, and five of the top ten machines are Blue Genes.
The test of speed is how well the machines perform on the LAPACK benchmarking suite. I won't tell you what that means, because it doesn't matter. Suffice it to say that it is a measure of the computer's performance on a limited set of operations. If you want a computer that does the stuff in the LAPACK suite really well, then this list is a dream come true. Most of us, however, have applications that need a different set of operations to run fast.
The list is fun and everything, but it's little more than a marketing ploy. If you optimize the compilation of the test suite, you can improve your performance by a factor of two or more. Some vendors actually spend a lot of manpower tweaking the system just for the purpose of improving the performance of their machines on the test suite, because they know that people who don't know much about supercomputers but who have the money to buy them (e.g. high-level administrators) are impressed by the Top 500 list. In reality, there are a lot more factors that need to be considered when investing in a supercomputer.
Thursday, July 28, 2005
Subscribe to:
Post Comments (Atom)
1 comment:
Whew! Thanks for warning me! Glad I didn't buy that supercomputer yet! ;)
Post a Comment