Altera has been known but its so sad. They bought them less than a decade ago for $13B and did nothing with it. They really got themselves into a great position for Heterogeneous computing and then just did nothing.
I wanted to prototype a systolic array coupled with a DRAM slice to hold the weight matrix, but there's not even DRAM blocks distributed on any of the FPGAs I saw. I guess Xilinx has HBM but still not quite what I wanted.