How do you virtualize an FPGA? A GPU can be virtualized and shared between several processes. It takes waaay too long to flash an FPGA to implement time-divisio

How do you virtualize an FPGA? A GPU can be virtualized and shared between several processes. It takes waaay too long to flash an FPGA to implement time-division multiplexing on it. All you can do is "slice up" the FPGA and give different processes different portions, and even that is a security vulnerability waiting to happen unless the FPGA is designed to keep these "blocks" separately contained. How do you balance the need for high-performance communication between the FPGA and the rest of the computer with the inherent inability to trust whatever configuration has been loaded into the FPGA? Software security is hard enough without infinitely reconfigurable devices lurking inside our machines. How does a possible FPGA configuration work optimally across a wide variety of FPGAs (which can have various "built-in", non-reconfigurable components) and sizes that can only be determined at "flashtime"? What exactly do regular users need an FPGA for that isn't already handled by dedicated silicon with much greater performance and much less power usage than an FPGA would? EDIT: I'd like an FPGA card for my PC just for running FPGA "emulators," because you can recreate a SNES or whatever in a tiny fraction of the gates that you need to make a CPU capable of emulating it with cycle accuracy at full speed. I seriously doubt there's much demand for that, though: if you're going to go to the trouble of buying special hardware for emulation, you might as well just buy the console and a flash cart.