A Future Of Pipes

I recently scored a copy of Maxwell's Demon, a book containing a collection of (among others) Landauer and Bennett's papers. As I was skimming them, I couldn't help but think that if there is a correspondence between heat and information erasure, can that be used to move "heat" in a fundamentally different way?

The idea would be you do some computation at point A, the related information erasure at point B, and the heat associated with it get "transported" faster than it would be by moving pipes around. The hope is that sending the bits along something like an optical network would be faster than laminar fluid flow.

I'm sure HN's more knowledgeable folks will tell me exactly where I'm wrong, but it seemed like an intriguing possibility.

Hanson needs to brush up on his thermodynamics. Reversible computers need not generate any heat. But they do need to grow larger as they are operated (their size scales with the number of operations performed). This means SPACE = TIME, which is a pretty sad state of affairs and probably the reason why all practical computers and known lifeforms are irreversible machines.

It's not a good bet that data centers will need to build out at a constant rapid pace into the future to keep pace with demand. A lot of the work in servers in recent years has been about improving utilization of available resources, and I expect that there is still a lot of headroom left for improvement. It's probably safe to say that across all datacenters, the vast majority of heat is generated while servers are mostly idle and have too little to do.

I don't think that it's safe to anticipate reaching some kind of fundamental limit on power efficiency, either. Nobody has come up with a viable replacement for the silicon transistor, despite decades of trying, and current CPU technology has practically stopped generating any improvements for the past several years. I think the important quantity to measure is the energy that goes into powering up the cpu for one clock cycle, because clock cycles are what drive computation forward and their individual energy usage basically just add together. For decades after the invention of the transistor, this energy dropped rapidly due to shrinking the wires. However, the wires reached a point where their capacitance began to fall off with decreasing size, and there have been no significant energy improvements since. Solving that problem requires either a revolutionary discovery about the physics of conductivity, a completely novel mode of doing computation that isn't based on electronics (photons?), or a fundamentally novel principle of organizing a cpu that somehow avoids being clock-bound (I heard about an effort over ten years ago, but I have not heard of it since). Any of those improvements could have surfaced at any time in the past few decades and been eagerly pursued, but none did, so the odds don't favor the miraculous appearance of one of them in the foreseeable future.

Is it "SPACE = TIME", or is the limitation actually set by the surface area of the volume enclosing the machine? In space-time coding for communications, the limitation on achievable spectral efficiency is related to the surface area of the volume enclosing the antenna array, in units of wavelength. It would seem sensible if this was reflected in a fundamental property of information.

Maybe a fundamental limit of the universe is that the minimum surface area of a volume containing a subatomic structure is related to the information required to define that structure? My maths is a bit fuzzy, but isn't there a fundamental relationship between a volume its boundary, involving a Fourier Transform? That could then mean that a limitation on surface area is equivalent to a limitation of the (4-dimensional?) volume occupied by some other quantity?

The above is speculation on my part, but is there a HN reader our there who can say whether such thoughts are garbage, or whether they have some basis?

>Thus in future cities crammed with computer hardware, roughly half of the volume is likely to be taken up by pipes that move cooling fluids in and out

A lot of this post reads like nonsense.

The author assumes the demand for CPU cycles is elastic in the long term (i.e. as prices of cycles drop, people will continue to use more and more of them).

General-purpose desktops and laptops today often don't come close to using all of the CPU horsepower available to them, and many server applications are limited by memory, network bandwidth, and storage.

It seems conceivable to me that at some point, before reaching thermodynamic limits, CPU cycles will stop being scarce for many application spaces, and future computing investments will mainly be in networking, memory, storage and software.

> The future of computing, after about 2035, is adiabatic reservable hardware. When such hardware runs at a cost-minimizing speed, half of the total budget is spent on computer hardware, and the other half is spent on energy and cooling for that hardware. Thus after 2035 or so, about as much will be spent on computer hardware and a physical space to place it as will be spent on hardware and space for systems to generate and transport energy into the computers, and to absorb and transport heat away from those computers.

The computer almost sounds alive...

Didn't Seymour Cray say that HPC was mostly about plumbing - to the manage the heat

mm maybe I ought to dust off those old steam tables eh.

There's a long way to go before building permanent structures on the Moon is a possibility (meteoroid detection/destruction, understanding seismic activity, etc.), but its far side would be a great place for building data centers.