Jed Rothwell wrote: > From the website blurb: > > "NVIDIA Tesla Personal Supercomputer > > Get your own supercomputer. Experience cluster level computing > performanceup to 250 times faster than standard PCs and > workstationsright at your desk. The NVIDIA® Tesla™ Personal > Supercomputer is based on the revolutionary NVIDIA® CUDA™ parallel > computing architecture and powered by up to 960 parallel processing cores."
I'm not familiar with that one, but from what I've seen of vaguely similar hardware it may work a bit like a vector processor. In other words, it's decidedly not SSI, and it's not symmetric; it runs as a specialized attached processor, and you DMA stuff into its dedicated memory and then DMA the results back out. (That's how Sony does it with the Cell processors, for instance.) How they're managing the memory with that many processors on a single board is, however, quite beyond me. Sounds pretty magical. The answer may be rolled up in the precise meaning of the world "streaming", which doesn't convey much to me in this context. How much hand holding you need to give it, and how much the software does for you automagically, is something you'd need to determine by a close reading of the documentation. So far nobody's told me to port our debugger to this beast (and I rather hope they don't). > > > 960 cores! Amazing. > > - Jed > >
