Google Willow

The folks at Google came up with this new chip and error correction package they named Willow.

They cannot really be blamed what happened after they released information on that... because what the journalists and analysts made of it is a prime example of what happens if somebody with dangerous half-knowledge writes an article about something that he or she think is the bees knees.

Or maybe - if you are an investor - what happens if you ignore some facts and laser-focus on something else to make things look waaay better than they really are.

Most of what has been written in newspapers and magazines - some of them even in the "science" category - latched on to one of the statements made by Google that just sounds fantastic: That Willow can do in 5 minutes what traditional computers could do in 10 septillion years (10^25) years.

Except, that is not what is written in the Google press release.

You can blame Google that they didn't aggressively set the record straight. Or maybe not... after all, they didn't lie in their Google Willow press release... they did make some claims in text but mentioned that those 5 minutes vs. 10^25 years is a thought experiment in a table somewhere at the bottom of that blog post.

Here is what Google wrote:

They are a tiny bit less dramatic in the table showing performance results:

See that? The word "Estimated".

Let's ignore the fact that the benchmark is designed specifically to be almost impossible to run at anywhere a decent speed with traditional computers. It is specifically designed to measure (theoretical) performance of quantum computers. Even some traditional benchmarks use problems that are not really practical to show a CPU/GPU/SDD/... performance. That is not the issue here. Let's focus instead on the real issue.

Willow runs for an average of under 100 microseconds. Let's make this an even 100 microseconds. It won't matter that much.

To do that calculation for real, Willow would need to run for 5 full minutes. That means the runtime would need to improve by the factor of 3 million.

Let that sink in.

In case you missed the point: Increasing the runtime T1 by 3 million is anything but simple. It is not like Google simply has to provide 5V to that chip for 5 minutes and that is it. There are various reasons why the average runtime is measured in microseconds. Lt. Data would say "The quantum field collapses rapidly."... with "rapidly" being in average less than 100 microseconds.

Which means that no, Google Willow cannot calculate 10^25 times faster than traditional computers can. Not. Even.Close.