[Nokinside]

I have been looking at their patents. Seems well defended and methodological. Someone like Intel, Nvidia or AMD might buy them soon.

Speedup is realistic. Multicore SMP or NUMA are not good for memory access patterns they optimize.

Their thick control flow model that should work for exclusive matrix-addition and log-prefix style memory access patterns. In comparison to the baseline the speedup is 150% in log-prefix algorithm, over 190% in fft-style butterfly algorithm, 50-100% in matrix addition and threshold filtering. silicon area and power consumption are estimated to be low.

Light reading material:

Optimizing Memory Access in TCF Processors with Compute-Update Operations Optimizing Memory Access in TCF Processors with Compute-Update Operations https://ieeexplore.ieee.org/document/9150423

The REPLICA on-chip network https://ieeexplore.ieee.org/document/7792877/

Preliminary Performance and Memory Access Scalability Study of Thick Control Flow Processors https://ieeexplore.ieee.org/document/10305463/

Realizing multioperations and multiprefixes in Thick Control Flow processors https://linkinghub.elsevier.com/retrieve/pii/S01419331230005...

[Firerouge]

Let's say Intel (or AMD/Qualcomm/ARM) buys them today.

What sort of turnaround should people expect before we start seeing CPU designs with this implemented in silicon?

What percentage of that will be R&D and QA vs actual time to create masks and fab an architecture with it?