Hacker News new | past | comments | ask | show | jobs | submit login

I was kinda hoping we would leave x86 behind, but whatever.



Why? It's a heck of a lot more fun doing ARM assembler than x86, but the only assembly work being done these days is a few hundred lines in the bootloader to setup the PLLs and get the C environment going. Decoding the instructions is such a tiny part of today's almost billion-transistor chips that the oddities of an instruction set really doesn't matter. AMD64 gives a good register file and fixes the problems most people complain about with x86.

The biggest impact instruction choice makes these days is on memory efficiency, but both ARM & x86 are relatively memory efficient compared to the other common RISC architectures.

Sure ARM uses less power and x86 provides better performance, but that's no longer due to the instruction set, that's almost entirely due to optimization tradeoffs.

I'm willing to bet that it will be easier for the x86 makers to scale down power usage than it will be for the ARM guys to scale up performance.


I wouldn't bet on it. According to Anandtech Cortex A15 will be around Core 2 Duo level of performance, with more or less the same power consumption per chip as today's dual cores and quad cores (Tegra 3).

Dual core 2 Ghz Cortex A15 chips will appear first in Q2 2012 from Samsung (Exynos 5250). Is Brazos or Atom at the level of Core 2 Duo performance yet? Nope. And they still use more power than Cortex A15. In the mean time, ARM chips grow in performance faster than x86 low-end chips drop in power consumption every year.

By the time Atom or Brazos achieves the same power consumption, high-end ARM chips will have first gen i3 or even 2nd gen i3 performance (a couple of years). Atom won't even be fanless until the end of 2013.


I'll take that bet.

Brazos & Atom are still 45nm, and are a lot faster than 45nm ARM parts. They should be, they use a lot more power. Exynos 5250 is 32nm.

In Q2 2012 we will finally be able to compare apples to apples. Medfield vs A15. And we'll definitely see fanless CPUs from Intel next year. I don't think there's room for a fan in this: http://www.anandtech.com/show/4788/intels-medfield-gingerbre...


The node doesn't matter all that much. It gives a 30% boost or so. That doesn't change the fact that ARM has been improving in performance about 2x-2.5x every 12 months. Have you seen Atom improve that fast in either performance or energy consumption lately?


The node by itself gives a 30% boost. It also enables a dramatic increase in the number of transistors, which can be leveraged to also provide a dramatic increase in performance, if there are known gains to be made.

It's far easier to be the second or third company doing something than to be the first. A large part of the increase in performance that ARM enjoyed was obtained by learning from the successes and failures of Intel & POWER & Alpha and all other performance trailblazers.

Now that ARM is getting closer to mainstream performance, that kind of momentum is going to be much harder to maintain.

But mostly I've learned to never bet against Intel. Many have done so, and the landscape is strewn with their failures over the last 30 years. Certainly NVidia & Samsung et al have some very impressive engineers that might be capable of beating them. Eventually Intel will stumble, and it might be this time. But I'm much more willing to bet on Intel than against them.


You can purchase fanless Atom and Brazos boards now. They just have huge heatsinks.


I reckon there might be more NEON that bootloader assembly out there in the ARM world.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: