What to do with Transputer Equipment?

On Monday (05/30/2011 at 08:27AM -0700), Chuck Guzis wrote: So, there are some people who still want such things-- and some people who will still build them. I know a thing or two about this system, http://www.nas.nasa.gov/hecc/resources/pleiades.html as I did the architecture for the first version rolled out in 2007-- at 487 TFlop in 100 racks. They've since scaled out to 168 racks, adding higher-end CPUs and can now cross the 1.0 PFlop line with LINPACK. I'm no longer with SGI but this product line continues and has been updated to QDR Infiniband, 10 GbE and other enhancements to the interconnect. I can assure you that the performance of the interconnect is still the bottleneck ;-) which is one reason why we put two parallel IB fabrics into this thing. A node in this machine is pretty much a standard Intel Xeon based server without any disks or power supply. It's built on a slightly different PCB form-factor but is otherwise a standard reference design for such a platform. The whole point was to leverage that commodity stuff so that you could A LOT of it together and scale it out at lower cost than totally proprietary stuff. Chris -- Chris Elmquist