BTW, if OP is reading this, I recommend having the baseline in your plots (e.g. std::deque) as the relative 100%, that way the performance improvement is clear.
This might be rendered moot by native microscaling support in Blackwell (MXFP). They've manually done a coarser-grained version of that for Hopper, but with full FP32 scaling factors.
You seem to have confused 1421 with 1491. The latter is generally considered to be a well-researched book while the former is, as you say, merely a grand tale.