News 12-core Zen 5 CPU beat Intel's 20-core chip and Apple's M3 Max in PassMark — Ryzen AI 9 HX 370 glides past the Core i7-14700HX and M3 Max

The article said:
The Zen 5-based processor showed competitive performance against Apple's top-of-the-line M3 Max CPU, beating the 14-core variant and approaching the 16-core in the single and multi-threaded scores.
Um, so the table shows the Apple M3 Max 16 Core with a single-threaded score of 4,779, while the AMD Ryzen AI 9 HX 370 manages only 4,213. That's a gap of nearly 12%! That's non-trivial and not really "approaching", which I'd infer as a gap in the lower single-digits.

Single-threaded performance is hard, especially in a mobile context, and greatly impacts perceived responsiveness. It should count for a lot more than multi-threaded performance, especially in laptops.
 
Single-threaded performance is hard, especially in a mobile context, and greatly impacts perceived responsiveness. It should count for a lot more than multi-threaded performance, especially in laptops.
Purely out of interest I’d like to see the differences in the instruction execution between competitive architects. What are the tricks? How do they make function x quicker? Wide decode paths.. fine, when I was learning a cpu had an instruction set and for example x86 was directly implemented.

Transmeta Crusoe came with a general purpose execution engine with a decode module that could take an instruction (set) and convert instructions from CISC into RISC. Intel and AMD both implement a similar solution today.

Data was processed in order, no speculative execution. Things were simple, understandable, today the operation of a cpu is a mystery…