Chuck Moore: Programming a 144-computer Chip to Minimize Power [video] | Hacker News Reader

Chuck Moore: Programming a 144-computer Chip to Minimize Power [video]

infoq.com

78 pointsqznc13 comments

I've been looking forward to this one! I have a soft spot for eccentric inventors that challenge the status quo simply by doing things their way and not blindly following conventional wisdom. Even if you disagree with them it's good to have your assumptions challenged, if only because it makes you aware of the fact that you have them.

The incredibly low-power usage of the chip is interesting, as is his perspective of counting the energy of each operation.

You can run the chip full-throttle with a solar panel the size of your hand, but where it shines is when it's not fully utilized. This is for (odd) general-purpose computation, not ASICs.

Not sure how many people remember, but back in the PowerPC days of Macintosh, the OpenFirmware chips actually contained a Forth interpreter implemented in hardware. It always struck me that, should the need arise, you could use the Forth interpreter to write a bootstrap C compiler, then use that to build a proper C compiler and from there compile and load a minimal kernel. In other words, starting with bare metal you could, in theory, get to a fully functional desktop computer...

...one of these days maybe I'll actually have the time to find out if that really works in practice.

The OLPC XO machines also have it, as well as SPARC-based Sun workstations.

If only a geek could buy the latest version of the OLPC laptop...

If only a low-income parent could buy the latest version of the OLPC laptop for their child....

The thing that really stop you is to find good documentation about your hardware. Displaying a character on screen or sending a packet over the network without using system or BIOS calls will require you to pages after pages of very technical and not always well written documentation.

That said you can pick a very old PC with a functional floppy drive, and write a Forth interpreter under DOS in assembler. Then, you "just" have to replace the system call to DOS to read/write blocks to/from your blockfile with BIOS calls to read/write floppy disk sectors. You can even do that under DOS, provided you don't have anything valuable on your hard disk. This DOS-based Boostrap Forth should feature at least an editor, a mini-assembler and a self-compiler. Once you have that, you can make a bootable disk that boots on your standalone Forth interpreter (which has therefore been promoted to the rank of 'OS' ;-). With a very old 5MHz 8086 "laptop" featuring an half-broken single density floppy disk drive, one can boot, make a change in the kernel, recompile, then reboot and test the change within a minute. Those machines were power monsters. It's a pity that CGA cards didn't have shaders ;-).

Chuck Moore is definitely one of the great computing heroes, especially if you're an idealist.

The problems with the GreenArrays product seem to be that it's too separated from the rest of the world, and in order to successfully interface peripherals (even SRAM, as covered in this video) to it requires completely grokking this alternate way of doing things, in addition to understanding the low level details of absolutely every interface.

Maybe I'm in the wrong circles, but right now if you can't shoot UDP packets out of something it doesn't really meet the minimum baseline for connectivity. If you're going to head off on your own (and certainly for energy purposes there would be good reason to do so) you need to have a simple bridge back to more standard land.

Two of the GA144 evaluation boards on my bench here in my office both shoot UDP packets out to learn the date and time when they boot. UDP over IPv4 over bit-banged 10baseT. You may ping one of them at g144.minerva.com.

That's on the standard eval boards? You've managed to get me interested in this rabbit hole again!

We have a little piggyback board with transformer, 10 MHz fox osc, and a dual op-amp to let us swing enough V to make the spec for twisted pair. Haven't started offering them publicly yet, but the code has been running two years. We gave an overview of the software defined NIC at SVFIG Forth Day 2012; video and powerpoint available thru SVFIG. The code is in the current arrayForth distribution. Still waiting for time to finish converting TCP to run on the 16-bit polyFORTH model but IP, ICMP, ARP, UDP all work. TCP only knows how to send a proper RESET.

I completely agree on Chuck Moore being one of the computing heroes of the last 4 decades or so. As far as I can see it, he (or Forth or GreenArrays) needs proper communication/marketing/lobbying staff that ensures that his ideas are being understood by the politicians, decision makers and investors.

Power consumption and processors has always been an issue, they get to warm and need cooling, server parks consume huge amounts of energy, batteries wear out to quickly etc... Wouldn't it be nice if the surface of a tablet (I am talking about 2015-2020) contained a 10" solar panel, and the processors inside the tablet would need only the energy provided by this solar panel? that would be a nice start.

I have an interest in forth, the insights of Chuck Moore, and GreenArrays, which is why I was (1) so excited to watch this talk and (2) so disappointed in my in-ability to follow it. I'm not sure what kind of background you need for his talk to make sense, but whatever it is, I don't have it, which causes me to wonder how many people do. I'd guess less than 1 in a million.

Simply put: this speech was a marketing disaster...

If Chuck Moore really wants to sell these things, he needs to go back to "this is colorForth" and start at the beginning.

I hate to say it, but right now GreenArrays looks like an awesome solution in search of a problem.

Reply on news.ycombinator.com