every once in a while I like to just sketch out future computing hardware. I recently got an nvidia 9500GT card. Soon I’ll likely try to get a GTX card (because those CUDA cores could rapidly accelerate my renders).
At least 32GB of RAM. Intel (at least an 8 core xeon, maybe more cores). Probably running raring.
4TB of spinning disk. 4TB of SSD storage.
The compute architecture of this hypothetical machine is going to be a little strange:
an nvidia tesla card of some sort
an nvidia gtx/quadro card for display
an amd firepro card
and an intel xeon phi card.
three of these cards are for computation purposes only. I can see how someone might balk at this, but hear out my reasons for this strangely architectured machine: individually, I’m not sure how much faith I want to put in a specific vendor. Yes, they’re all very fast, but my goal is not bitcoin mining or mmorpg playing, but doing mathematics visualizations. The heterogenous compute architecture in this hypothetical machine is insurance against hardware quirks by each vendor. One would like to verify computations on different hardware in the same way that it used to be the case that one would test supercomputers by instructing them to compute many pi to many a decimal place, because then the output could be tested against previous computations, and if there were mistakes, tracing them to problems of machine design or implementation was the thing to do.