"100x improvement" in _memcpy_ performance?
On a benchmark ("Matched Parallel Algorithms") where a Xeon is 10x _slower_ than an i7?
Their page that should have details is equally vague: https://flow-computing.com/technology/
For instance, the "VIRTUAL ILP/LLP" section sounds very similar to operand forwarding, a technique roughly as old as pipelined execution: