1 comment

[ 3.1 ms ] story [ 9.3 ms ] thread
Slides: http://web.stanford.edu/class/ee380/Abstracts/190605-slides....

Author: http://www.stuffedcow.net/

The design was a PhD project that took a long time (I think 2011-2017, longer than usual for such, I think).

The author now works at Intel.

The design was left unfinished, so no final benchmarks or even final clock rate.

I was impressed watching this, but I have no idea how impressive it actually is.

I wonder what experts int he field think.

I also wonder why there isn't an open source project to do something like this (or finish this project, which I guess is open source?).

Would the techniques apply to an ARMv7 core? How much extra work is it to add floating point and vector instructions (I guess a lot).

Would it be possible to teach a verilog compiler to do things that way instead of the normal way which results in lower frequency?