How do you really target FPGA and GPU when these are so different?
e.g. what I find appealing about FPGA is (1) very low latency and (2) arbitrary precision data paths, like if I want 6-bit data paths I can have them. The GPU on the other hand is throughput oriented and you get the data types that were baked in.
e.g. what I find appealing about FPGA is (1) very low latency and (2) arbitrary precision data paths, like if I want 6-bit data paths I can have them. The GPU on the other hand is throughput oriented and you get the data types that were baked in.
I remember looking at
https://www.intel.com/content/www/us/en/developer/tools/onea...
and thinking "how could that possibly work?" and the fact that it's had basically zero uptake suggests to me that my hot take was the right take.