alt Hacker NewsMMIX uses register windows to make stack frame pointer offsets unnecessary when referring to PUSHed arguments. Don is trying to make the algorithms understandable and correct, and by hiding some details that are handled efficiently by compilers (keeping track of FP and offsets), it benefits the human reader.
Don's first computer was the IBM 650 https://en.wikipedia.org/wiki/IBM_650?useskin=vector see also http://ed-thelen.org/comp-hist/KnuthIBM650Appreciation.pdf so MIX was a simplified version of the 650 because, well, it's well-defined and simple -- and Don knew a popular IBM machine very well. And there's this, in Vol 1:
This series of books is affectionately dedicated to the Type 650 computer once installed at Case Institute of Technology, in remembrance of many pleasant evenings.
MMIX is for all you youngsters who think RISC is all the rage ;-) and I think he does an admirable job creating a fully-defined machine that does use more modern hardware techniques. The fact that he fully defines his underlying machine is exactly correct, because it lays the foundation for precisely expressing the algorithms, and for giving Time and Space (runtime) estimates.
I believe it's fundamentally incorrect to think of these abstract machines as 'assembly language' but rather, I think, they define a stable foundation onto which accurately described algorithms can be expressed. You're supposed to 'play computer' and follow along -- step by step -- to understand the deep details of the algorithms.
I mean, a virtual ISA (think PTX, LLVM IR, WASM) is going to do a better job of giving you an abstract machine than something like MMIX. Virtual registers and call arguments/return values as nary arguments rather than fixed registers give you most of what you want, and it's easy to augment it with a large slice of primitive operations that is a superset rather than subset of assembly languages.
(There's another criticism to level at pseudo-assembly language, which is that modern high-performance processors are superscalar with cache hierarchies, which makes the analysis of execution time itself difficult from the kind of first principles that Knuth is working at. I can appreciate why Knuth is working differently from the more traditional big-O notation of typical algorithms classes, but it does need to be acknowledged that it does sap the treatise of its supposedly timeless quality.)