I got a base-spec MacBook Pro 16 inch with the M2 Pro chip from work last week. Before that, I was on my Linux workstation with an AMD Ryzen 3950x, a desktop CPU released 3.5 years ago.
The compile time for a debug build of LLVM and MLIR is basically the same on those machines, the desktop machine only winning by a little bit. Roughly 10 minutes. Yes, of course desktop vs. laptop is unfair, but it's also a 3.5 year old chip.
And if you really want all-out performance, there are still AMD ThreadRippers, which should easily beat an M2 Ultra in almost all multi-threaded workloads.
(Details on the benchmark run: Building LLVM from source using precompiled LLVM 15 downloaded from the official website, debug build, only clang;mlir;lld projects, using mold/sold linker on Linux/Mac respectively, ninja build tool).
The compile time for a debug build of LLVM and MLIR is basically the same on those machines, the desktop machine only winning by a little bit. Roughly 10 minutes. Yes, of course desktop vs. laptop is unfair, but it's also a 3.5 year old chip.
And if you really want all-out performance, there are still AMD ThreadRippers, which should easily beat an M2 Ultra in almost all multi-threaded workloads.
(Details on the benchmark run: Building LLVM from source using precompiled LLVM 15 downloaded from the official website, debug build, only clang;mlir;lld projects, using mold/sold linker on Linux/Mac respectively, ninja build tool).