Would anyone care to comment on their experience with regards to the following questions I have about this?
1. Are you using an FPGA implementation of OpenCL, or are you synthesizing a specific piece of OpenCL code into the FPGA?
2. How does performance compare between OpenCL running on a high-end GPU and a high-end FPGA?
3. How does performance compare between implementing someting in FPGA directly and implementing it in OpenCL and running that on an FPGA?
4. How does working with OpenCL for FPGA compare to working with the FPGA directly? Iow, what are the advantages and disadvantages you find of OpenCL as opposed to Verilog or VHDL (or other ways you program an FPGA)?
5. Based on the above, if you had X amount of dollars that could buy you either a number of GPUs or a (probably different) number of FPGAs, which would you choose? Consider also the electricity cost of running these as I've heard that FPGAs generally waste less electricity because they produce less heat.