Hacker News Reader

> On a CPU intensive app [...] the emulated software is roughly 50% of the native speed. While this number can still be improved [...] this probably won’t change much in the future (it will probably top at maybe 60%)

This is a shame if true. What's the difference between this and something like rosetta 2 that allows the latter to achieve greater performance?

View on HN

Unless I am mistaken M1 chips have hardware dedicated to making translations faster.

To be specific, the M1 supports a stronger memory model which makes emulation easier. There is no real dedicated hardware though.

But the original benchmarks weren't multithreaded, which should make that a nonissue?

(Is it possible to generate code assuming a weaker memory model, and then invalidate that upon creation of a thread? Might get dicey with shared memory, though...)

The hardware itself has dedicated instructions

Possibly M1 hardware optimisation for x86 emulation?

Reply on news.ycombinator.com