ASCI White, Lawrence Livermore Lab

The construction of ASCI White (pictured right) was completed in June 2000 and is currently the worlds most powerful supercomputer.

The hardware:
The peak performance of the computer is 12.3 teraflops. This means it is capable of computing 12.3 trillion operations per second. This extremely high performance is achieved through its massively parallel design. The processors used are IBM RS6000 SP Power3's which run at 375 MHz. There are 8,192 of these processors in the core compute system. The total amount of RAM is 6Tb. The system is housed in over two hundred cabinets and fills a large room with an area the size of two basket ball courts.

The computer runs IBM's AIX operating systems. User applications can be developed using the wide variety of languages including FORTRAN and C/C++.

Human brain comparison:
It is possible to make a very rough estimate of the computing power of the human brain. The human brain contains about 100 billion neurons, each of which is connected to about 1000 of its neighbours through synapses. This equals a total of 100 trillion connections, all of which operate in parallel. If we estimate each connection to be capable of 100 operations per second then this gives a total of 10,000 teraflops for the whole brain.

Note that this really is a rough estimate. Others rate the brain at between 100 and 100,000 teraflops. But one way or another the computing power of the human brain is of the order of hundreds of times greater than ASCI White. Although the human brain is much more powerful, it is not unimaginably more so. It is believed that supercomputers will increase in power by a hundred fold within the next decade. This would mean that supercomputers would be on a par with the human brain by the year 2010.

Lawrence Livermore National Laboratory:
The ASCI White supercomputer is owned by the Lawrence Livermore National Laboratory in California, USA. Most of the work using ASCI White is related to nuclear weapons research. Apparently it has never been used to simulate an artificial neural network or for any similar artificial intelligence applications.

Links:
ASCI White, Lawrence Livermore Lab: www.llnl.gov/asci/platforms/white