Hacker News Reader

Comment by raverbashing | Hacker News Reader

One thing it's not clear to me is why did ARM was so much power efficient even at the beginning and apparently without being explicitly built as such

View on HN

The ARM chip was low power essentially by accident. The designers wanted to keep it under (I think) 10 watts which would let them use cheap plastic packages for the chips. But they didn't have good power simulation tools so they needed to be very conservative with the design. When they manufactured the chips, they found the power consumption was closer to 1 watt.

Edit: as beagle3 pointed out, I got the numbers wrong; they aimed for 1 watt and came in under 0.1 watt. For more details, see Steve Furber's oral history, page 15. He says "Effectively, we used Victorian engineering margins on the power consumption of the chip." https://archive.computerhistory.org/resources/access/text/20...

May find it later when I am not on mobile, but iirc they aimed for 1W and ended up achieving 0.1W or so - at first they were sure something was wrong about their test rig because it seemed like an unpowered CPU was actually working - and then they realized that it was getting enough power through the data and address lines, even though the Vcc was not connected.

> The power test tools they were using were unreliable and approximate, but good enough to ensure this rule of thumb power requirement. When the first test chips came back from the lab on the 26 April 1985, Furber plugged one into a development board, and was happy to see it working perfectly first time.

> Deeply puzzling, though, was the reading on the multimeter connected in series with the power supply. The needle was at zero: the processor seemed to be consuming no power whatsoever.

> As Wilson tells it: “The development board plugged the chip into had a fault: there was no current being sent down the power supply lines at all. The processor was actually running on leakage from the logic circuits. So the low-power big thing that the ARM is most valued for today, the reason that it's on all your mobile phones, was a complete accident."

> Wilson had, it turned out, designed a powerful 32-bit processor that consumed no more than a tenth of a Watt.

https://www.theregister.com/2012/05/03/unsung_heroes_of_tech...

An incredible story, thanks.

Here's a link to the video of an interview with Steve Furber [1], who worked on the chip's development, discussing this.

[1] https://youtu.be/1jOJl8gRPyQ?t=508

I remember one quote from Hermann Houser (one of the founders of Acorn)... "I gave them two important things to develop the Arm chip. I gave them no money, and I gave them no people".

It's amazing what necessity can accomplish...

That's reminiscent of the famous "janitor" letter written by IBM's president Watson in 1963. He asked why the CDC 6600 was a better computer than IBM's when Seymour Cray's team had "34 people including the janitor" while IBM had a "vast development effort".

Cray's response (probably apocryphal) was, "It seems like Mr. Watson has answered his own question."

https://www.computerhistory.org/revolution/supercomputers/10...

I'd love to see a deep analysis of this, but I think a confluence of a few elements:

- fixed instruction width, small instruction set => simple decode

- instructions wide enough to load many constants into registers in immediate mode

- comparatively large register file (15 general + PC) compared to the 80386 (effectively 8 + PC, plus the segment registers) meant reduced need to fill/spill intermediate values, a tradeoff which gets better over time as the power consumption of bus doesn't benefit from Moore's law since it's off chip

- any-to-any register instructions saves space and time shuffling values within the CPU (again, a hassle on the x86 series)

- conditional execution bits were a nice pipelining assistance trick, not clear how widely used they were as they weren't included in Thumb

- not having the 8086's ridiculous segment system

Which can roughly be boiled down to 25,000 transistors vs 275,000 in the (roughly contemporary) 386.

It might be a consequence of Acorn's aggressively simple design for the CPU. They preferred complex software over complex hardware.

I don't think RISC makes software more complex, rather the contrary. You need to use more instructions to do the same task, but they are simpler, more regular. Especially when writing compilers, RISC is actually a nice target.

It was about packaging. If you run cool, you can get away with much cheaper materials by this saving significant portions of the total manufacturing costs. This was an explicit design goal.

Reply on news.ycombinator.com