Comment by acqq | Hacker News Reader

Comment by acqq | Hacker News Reader

Parent Full thread

https://techcrunch.com/2020/11/17/yeah-apples-m1-macbook-pro...

WebKit compile time:

"2020 13-inch MacBook Pro with Intel CPU: 46 minutes and 10 seconds

M1 MacBook Pro: 20 minutes and 43 seconds"

And even more fascinating:

"2019 Mac Pro (Intel 12-Core 3.3GHz CPU): 20 minutes and 11 seconds

M1 Mac mini: 19 minutes 32 seconds"

Some other measurements of some other compilations show the advantage of many core Intels but with more cores and much more electricity spent:

https://build2.org/blog/apple-m1-compilation.xhtml

"time sh ./build2-install-0.13.0.sh --local --yes ~/install

163s"

"As the first point of reference, let's compare it to my workstation with an 8-core Intel Xeon E-2288G (essentially i9-9900K plus ECC). It does the same build using vanilla Clang in 131s. While E-2288 is faster, M1's performance is nevertheless impressive. Especially when you consider that during the build the workstation is belching hot air and screaming like an airplane about to take off while M1 is whisper-quiet with barely warm air coming from its exhaust."

It seems that in the comparable power envelope M1 remains better:

"CPU CORES/THREADS TIME

Apple M1 4+4 163s

AMD Ryzen 4700U 8/8 222s

Intel Core 1165G 4/8 295s"

Well, the next question is: is clang doing anything different for arm than for x86? Why is the difference so big if that is not the case? Memory, cache and latency might be the big thing, and there is nothing saying that intel or AMD couldn't compete with that.

Clang is definitely not better for ARMv8 yet, in fact clang doesn’t even have an optimization profile for the M1 yet.

Not spending as much time in the target optimizer could explain some of it.

Right, and on the flipside significant effort has been invested in Clang performance when compiling for Intel CPUs. Much of that is likely to the benefit of all backends, but some of it surely is specific to Intel.

IMO this bit right here is worthy of highlighting again:

> Especially when you consider that during the build the workstation is belching hot air and screaming like an airplane about to take off while M1 is whisper-quiet with barely warm air coming from its exhaust.

I have never run a significant amount of compilation on any machine that didn't hit heat issues. So either the M1 is doing very well at managing heat, or Clang is doing incredibly poorly at exploiting the full system. In either case this makes the M1 look like something special.

armv8 is one of clang's most important targets, and the optimization profiles don't impact compile time.

Reply on news.ycombinator.com