The Banana Peel Project

Using blisteringly fast computers for simulations and complex modeling has become so widespread that some semiconductor manufacturers have begun marketing graphics-processing supercomputers that can fit under a desk.  NVIDIA announced this week that the Tesla D870 “deskside supercomputer” will be available in August of this year.  Check out the article on Ars Technica.

Last year, Tyan released a “personal supercomputer” called the Typhoon, but it was essentially just an empty box into which one could stick a whole bunch of processors (sold separately) and then link them up in parallel.  But NVIDIA’s Tesla D870 is a complete computing system for graphical modeling and simulation with 256 individual processors in a box the size of a mini fridge combining for a total of 1000 gigaflops, or one teraflop.  The fastest full-size supercomputer today, IBM’s Blue Gene/L, runs at 280 TFLOPS.

What’s so cool about this not so much the fact that it’s so small, but that it’s small and still qualifies as a supercomputer.  Supercomputers have always shrunk over time.  Ten years from now, personal computers will have the same processing power as supercomputers do today.  By the time the processing power of a given supercomputer is miniaturized and marketed as a PC, a newer supercomputer has already emerged to eclipse and antiquate it.  But the Tesla and machines like it are small while still knocking the socks off of even the most powerful high-performance desktop computers.

Supercomputers are generally designed to be flexible, meaning that they can be used either as number crunchers or graphical simulators.  I think it’s really interesting that what looks like the first “real” personal supercomputer is explicitly designed for graphics.  The machine’s two GPUs (graphics processing units) are each made up of 128 parallel processors whose architecture is specifically created for intensive 3D modeling.

Obviously, if the Tesla deskside supercomputer is successful in the market–and I predict that it will be–it will be indicative of the extent to which our ability to create complex visual simulations has captured our artistic and scientific imagination.

The following image is from a simulation of a black hole run on SGI Project Columbia at the NASA Advanced Supercomputing Facility at the NASA Ames Research Center.