4 ms·
The benchmark that's blown me away most so far: Marco Arment on Accidental Tech Podcast says the new Macbook Air is significantly faster to compile Overcast (a
by mcintyre1994 6y ago
The benchmark that's blown me away most so far: Marco Arment on Accidental Tech Podcast says the new Macbook Air is significantly faster to compile Overcast (a pretty big iOS app) than his 10 core $6000+ iMac Pro with 64GB RAM and a 160W CPU. https://overcast.fm/+R7DWXqqm0/1:08:11 https://overcast.fm/+R7DWXqqm0/1:08:11
- brundolf 6y agoI/O-bound, maybe? Could be misleading but still impressive
- kllrnohj 6y agoCompilation tends to benefit heavily from bigger caches & faster memory, yes. It's an area that Zen 2 & 3 have been absolutely fantastic at thanks to their huge caches as well. The iMac Pro's relatively slow 2666MHz RAM can't be saved by the also relatively small (these days) cache in the Xeon-W it uses.
- dragontamer 6y agoXeon W is 7 years old at this point, and the Apple M1 is a pretty darn good modern chip. Today's desktops are as good as servers from 5 years ago. Today's laptops are as good as desktops 5 years ago. Today's phones are as good as laptops 5 years ago. --------- The Apple M1 makes a leap with a huge L1 cache, very wide execution, at a very low clock rate (2.8 GHz for power-efficiency). But when you pair such a leap with a 7-year technology jump, yeah, things look quite silly.
- mikkelam 6y agoYeah something is fishy here. Could be the new RAM which is way faster
- FireBeyond 6y agoOne thing to factor in here is that the MBA isn't cross-compiling. It's a RISC compiler writing RISC code. To be clear though, it doesn't matter in the end if the _result_ is more productivity. But it is worth noting that under the hood, the comparison is not entirely apples to apples.
- tpush 6y agoWhy would it matter whether you emit x86 code or ARM64 code? That code that produces either doesn't get magically faster when executed on their respective archs.
- hawk_ 6y agoEdit: as had been pointed out here there is no x86 code generation happening so this doesn't apply. Yes I have seen claims from people along the lines of 'x86 code has more optimization paths in mainstream compilers for historic and other reasons'. Not sure about that, certainly can't imagine such a significant difference.
- klelatti 6y agoIt's not identical but no obvious reason why emitting ARM64 code from a compiler running on x86 hardware should be more onerous on ARM64 hardware. Plus ARM64 is not really RISC in a meaningful way - it'll still be executing microcode under the hood.