Supercomputing on the Cheap

In my "modest" opinion, supercomputers should be parallel, with thousand of boards populated with Xilinx FPGAs, say large Spartan or Virtex chips, each with millions gates.

In nature all brains are parallel, with hundred thousand millions of neurons, and each neuron is like a computer connected to a large network, and this is true starting with the brains of insects like bees and other sophisticated flying predators, they are capable to recognize anything on the fly because they process visible and infrared images and ultra sounds in parallel, not like Von Newman and Harvard machines.

In my opinion Xilinx FPGAs with dominate absolutely all aspect of fast math algorithms processing in real time. There is no possible comparison, I say this by experience, inside an FPGA you can build complex math modules, in floating point, that respond almost instantly, at the nano second speed time resolution.

Another thing is about programming. The future of programming parallel processes is in graphics tools, not textual but graphics, because text is intrinsically understood as a sequential process, even if text can represent a parallel process, it does not help to visualize a complex parallel process as easy a graphics. For the moment, some programmers with sequential software "alma mater" have been trying to dominate the scenario in FPGAs with their obsolete sequential concept text tools, but at last, graphics programming will dominate all the scenario of FPGA programming.

When you drive in a big city looking for an address, what do you prefer to use, a large GPS all-text address book or a colorful screen with an illustrated map with some little GPS text on the margins?

This is my "modest" opinion based exclusively in experience.

Jaime Soto Figueroa

http://www.matharts.cl/