Hacker News Reader

are we actually going to see application-specific hardware, or are we going to see something like an FPGA as a coprocessor in consumer hardware?

View on HN

FPGAs really wouldn't help much. The theoretical gains sound good, but frankly, the practical speed of a cpu is measured by how fast it can run crappy code. FPGA in a cpu would just be a pointless waste of silicon that almost nobody would program for.

Application-specific hardware is already being added to the cpus -- the newest Intel ones have a hardware RNG and hardware video encoders and decoders.

"already" ? GPUs are almost hardware implementations of directX/openGL APIs and it is not new.

There's a team at UC-Berkeley which is working on compiling C++ to a combination of CPU/GPU/FPGA code right now, which might change things. I think you need a special architecture to take full advantage of it though.

In the ray tracing example I mentioned above, FPGAs got a 10x improvement. Going to silicon got another 10x.

Reply on news.ycombinator.com