wait, the hard part of quantum is now a normal computer?

by Marco Adebayo

reading the write ups from IEEE quantum week in toronto and one bit stuck with me. apparently the bottleneck for error correction isnt the qubits anymore, its the decoder, the ordinary classical hardware that reads the errors and sends back a fix. it has to do that in something like 10 microseconds or the next round of errors piles up. people were running it on nvidia gpus, others had dedicated decoder partners genuine question from someone who fixes laptops for a living. if the classical side is what everyone is racing on, is the quantum part basically sorted and its just engineering now? or is this just where the next wall happens to be also slightly funny that after all the hype the thing holding it back is latency lol, same as every ticket i get

favorite 2 comment 3

Comments

Daniel Cohen 4 days ago

Marco, good question, and the honest answer is the second one. The decoder race is real and it's a healthy sign. You only fight over a microsecond budget once the qubits are good enough that correcting them is worth doing in real time. A couple of years ago most demos decoded offline, after the run, which is fine for a paper and useless for a computer. But "sorted" is a stretch. Fast decoding buys you logical qubits that survive longer. You still need far more of them, at lower error rates, for long enough to run something useful, and that's mostly fabrication, wiring and cooling. Each of those is its own wall. So I'd read it as: the problem stopped being only physics. That's progress, just not the finish line.

Bernadette Kelly 4 days ago

I used to teach parity checks to fifth years with a grid of playing cards, flip one while they look away and they find it. Never imagined it would end up racing a clock in Toronto. Lovely to read.

Ava Engel 3 days ago

Marco, Daniel's right, but one thing from the lab side, because "latency, same as every ticket" is closer to the truth than you'd think. The decoder has a hard deadline because errors keep happening while it thinks. Miss the window and you aren't just slow, you're wrong, and the backlog compounds every round. That's why it went from an offline afterthought to the main event in a couple of years. A while ago this meant racks of FPGAs or GPUs next to the fridge. Groups are now showing real-time decoding on fairly ordinary CPUs at small sizes. The open question is whether that holds as the code distance goes up, because the work grows faster than the qubit count. So not sorted. But it's a wall far more people know how to climb.