Looks very similar to what Transmeta did. Hard to know if there are any real gains without hands on the actual hardware. 4 instructions per clock as 2.8GHz is not exactly earth shattering performance by modern standards.
While I can imagine niche applications I cannot see this becoming mainstream soon. Usually software needs to be rewritten to take advantage of the FPGAs - often using different algorithms.
5 comments
[ 2.6 ms ] story [ 10.8 ms ] threadThey presented some high level stuff at ISC 2020. https://www.tachyum.com/assets/img/Tachyum_20ISC20.pdf
The image on slide 7 looks almost exactly like this design (including the 8x8 core matrix)
https://en.wikipedia.org/wiki/TILE64
Which targeted the same market. The problem is, this is actually a pretty hard architecture to program for.
Earlier attempts: https://www.anandtech.com/show/9790/ibm-xilinx-sc15-collabor...