Comment by misnome

Oh christ, absolutely this. We spent some time evaluating FPGA for our purposes and ended up GPU instead (algorithms we running can be adapted to strength of either).

The concepts are easy enough but learning the toolsets are an exercise in frustration… the documentation/onboarding is either nonexistent or extremely unhelpful, and getting past the stage of “the entire thing doesn’t work because you misclicked a button in the gui several hours ago”. In theory everything can be scripted, usually in TCL, but this is also unstable and seems liable to break every different version of the toolsets.

Alongside Xilinx, we also looked at Altera/Intel OneAPI/dpcpp and this seemed promising until we realised we were encountering so many toolchain/actual compiler bugs that nobody else could have been actually using this, except the oneapi cloud platform that seemed it had been hotpatched to fix some of the issues. In the end, after selling us some compatible cards they dropped the OS and card from support. I guess this taught us not to trust Intel!

We decided teaching to Juniors would be an exercise in frustration unless hiring explicitly for, and decided to go the GPU route.