Hacker News Reader

I think they focused more on real-world performance that wouldn’t have been represented so well.

Things like lots of floating point ability for JavaScript (which has no integer types) or optimizing ref counting (obj-c and swift do this a lot).

I imagine those things wouldn’t impact traditional benchmarks.

View on HN

That's the interesting thing; they don't seem to have.

If they'd gone for ref-counting benchmarks, that would have shown something more like 5-10x; it's really good at that. The graphs they showed seem to be reflective of performance in general-purpose benchmarks (SpecINT et al).

I'm saying they focused on real-world performance in the chip and therefore couldn't show benchmarks that would've represented well. Not that they should've had more specific benchmarks. In no way would this audience be able to understand something like reference counting.

Reply on news.ycombinator.com