Thanks for a quick fix on a Sunday afternoon!
I’m impressed that the M1 can keep up on this SIMD-optimized code, likely at much lower temperature / power use.
And even the Rosetta numbers are pretty decent.
I’m impressed that the M1 can keep up on this SIMD-optimized code, likely at much lower temperature / power use.
And even the Rosetta numbers are pretty decent.
Far inferior becomes....actually superior in many cases, even at SIMD.
It is clearly the case that the M1 CPU/SoC has a significant performance advantage in typical branchy single-core code, but much less advantage if any for certain kinds of heavily optimized numerics. Beyond that high-level summary, it’s good to dive into the details, and spark discussions.
Everyone is just now getting their hands on these chips, learning how to work with them, and trying to figure out how to best optimize for them.