8 ms·
No. Intel worked out it needs to open its production capacity to other vendors. They will end up another ARM fab with a legacy x86-64 business strapped on the s
by hughrr 4y ago
No. Intel worked out it needs to open its production capacity to other vendors. They will end up another ARM fab with a legacy x86-64 business strapped on the side. That's probably not a bad place to be really. I think x86-64 will fizzle out in about a decade.
- astrange 4y agoI don't feel like ARM has serious technical advantages over x86-64 as an ISA, although it is cleaner and has more security features which is good. Isn't the main advantage just that it's easier to license ARM? Once enough patents expire all ISAs are eventually equal, I'd think.
- hughrr 4y agoSpend some time looking at optimised compiler output on godbolt on both architectures. ARM has some really nice tricks up its sleeves. I’ve been using ARM since about 1992 though so I may be biased.
- saagarjha 4y agoSo does Intel ;)
- hughrr 4y agoYes and no. Look at some of the loop optimisations possible on ARM compared to x86-64. I've had x86-64 run 8 instructions that ARM does in 1 instruction.
- saagarjha 4y agoThis applies equally well, or dare I saw even better, on x86. (Arm tends to catch up because of higher IPC.)
- astrange 4y agoI remember PPC and its rlwinms and co. My ARM isn’t that good, though I can read it. But some of those x86 instructions take 0.5 cycles and some of them take 0 if they’re removed by fusion or register renaming. It has worse problems, like loop instructions you can’t actually use but take up the shortest codes.
- smoldesu 4y agoI feel much the same way. I've used both pretty extensively at this point, and I'm not sure if I'm a believer in either mentality. I'm hoping that RISC-V will be the one to blow my mind, though.