Reciprocal Quantum Logic
Another potential major computing advance.
Imagine an AI data center that fits into a shoebox.
OK, that’s a bit extreme, because it would need a cryocooler sufficient to keep liquid helium in a liquid state. A thousand watts of compute will need 300kw of power into the cryocooler. That’s roughly equal to the power output of a new Corvette. So this will come down to an operations per watt calculation.
But if it can be made to work … things that take up acres today will fit into cubic feet in the future.
There is a LOT of physics in this Asianometry piece. I liked it. Most of you will probably not.
Speculation:
Recently we have covered …
And now we have a fourth possible epochal change in Snowcap’s effort to commercialize Reciprocal Quantum Logic.
Thermodynamic methods are new, the other three are already used in various devices, and have been for a long time. They’re just all ripe for major innovation that displaces the things we’ve used for my entire life.
This is an exciting time to be alive. Last night I was reading about Cerebras and Groq, both of whom build enormous, specialized chips just for AI. If what they are doing can be made smaller, cooler, and much faster …
What a brave new world we are facing.
The assumptions made here are bassakwards. Computer processor chips over time use smaller and smaller sized transistors. As the transistor size goes down and and the binary voltage goes down, power decreases over the years. Today’s processors have multiple cores on one chip and the main factor in energy consumption is how the software utilizes all those cores. What we attempt to compute today is far more than in the past. Power for CPUs are limited by the size and thermal path away from the active layer. This puts all chips in roughly the same energy limit. Most chips are less than an inch square and the fan that you clip onto the chip can only do so much cooling via the thermal grease. The energy used is more about how many CPUs a data center is running and the AC units used to cool all those computers back down closer to room temperature. The environment will be fine if we stop polluting with our gasoline or diesel cars and comically oversized pickup trucks. The data center pollution could also include the IT guy’s drive to work in his beater.
The assumptions made here are bassakwards. Computer processor chips over time use smaller and smaller sized transistors. As the transistor size goes down and and the binary voltage goes down, power decreases over the years. Today’s processors have multiple cores on one chip and the main factor in energy consumption is how the software utilizes all those cores. What we attempt to compute today is far more than in the past. Power for CPUs are limited by the size and thermal path away from the active layer. This puts all chips in roughly the same energy limit. Most chips are less than an inch square and the fan that you clip onto the chip can only do so much cooling via the thermal grease. The energy used is more about how many CPUs a data center is running and the AC units used to cool all those computers back down closer to room temperature. The environment will be fine if we stop polluting with our gasoline or diesel cars and comically oversized pickup trucks. The data center pollution could also include the IT guy’s drive to work in his beater.