Hacker News Reader

Comment by dmitrygr | Hacker News Reader

Besides go dropping the ball on aarch64, it looks like the days of Intel are numbered in every place where perf per watt matters.

That, or Intel has to actually wake-up their R&D department that's been in hibernation for the last decade of their desktop and server de facto monopoly.

View on HN

Intel is addicted to insane markups on chips. I'm sure their R&D department knows where to go, but the Intel of today cannot subsist on the relatively humble royalties that ARM charges.

On the very high end, they used to have no competition, so they could charge whatever. But that's being eaten by AMD, nVidia and ARM. On the low end, they can't compete with the margins they like.

Yes, in some parts of the market at least. The Cloudflares of the world have the resources to get spun up with new architectures while mom and pop shops might have a harder time. But then again the Googles and Facebooks of the world could see that those Go libraries get optimized for ARM if they needed them to.

I don’t think so. Intel has world leading fab capabilities. Once they shrink the process, perf per watt ratio should improve quite a bit. It will be a really hard slog for everybody else to roll out 7nm and beyond.

There is only so far you can shrink. Also Intel’s architecture is way more complex, due to legacy support. Qualcomm for example do not support ARM 32 bit instructions at all, only the new 64-bit ones. If Intel could do the same, they would be better off. Alas it’s too late. I think if they went this way with Itanium, 64-bit, fixed sized ISA, instead of VLIW it would be wise. Just thinking about the complexity of the decoder that jas to cope with intructions ranging from 1 byte to 32 byte!

Intel might be a year or two ahead of everybody else, at least in performance if not density, but with Moore's law running out of steam the value of a year keeps going down and the performance competitiveness of everyone else keeps getting closer.

Reply on news.ycombinator.com