IBM quantum computer does in 19 seconds what could take a supercomputer 110 years

[Riddler]

ScienceAlert has a write up on a new quantum advantage claim using IBM hardware. A team led by Tigran Sedrakyan from BlueQubit used IBM's Nighthawk r2, a 120 qubit superconducting processor available through the cloud, to run a random circuit sampling task. Using 61 qubits across 36 cycles, it produced one million samples in just 19 seconds. The researchers estimate the same job would take around 110 years on Frontier, which was the world's fastest supercomputer before 2024

The authors describe it as the first demonstration of quantum advantage for vanilla random circuit sampling on a commercially and broadly accessible quantum processor. That last part is the point. Previous advantage claims came from custom lab setups, while this ran on a machine anyone with cloud access could in principle use, with standard workflows and no special calibration

There are big caveats though. The 110 year figure comes from one particular classical method, tensor network contraction, which put the classical cost at around 1.2 times 10 to the 27 operations. That is not a fundamental limit. Classical researchers have a long history of finding clever shortcuts that shrink these estimates dramatically, as happened with Google's 2019 claim. The paper is also a preprint on arXiv and has not been peer reviewed

To their credit, the team has released their circuits, samples and code so others can check the result. That is exactly how it should be done. If classical researchers want to knock the estimate down, they have everything they need to try

Random circuit sampling is also not a useful task in itself. It is essentially a benchmark designed to be hard for classical computers, rather than something that solves a real problem. So this is a milestone in the arms race between quantum and classical computing, not a sign that quantum computers are about to replace supercomputers. How long before we see advantage on something practical?

IBM's Quantum Computer Completes in 19 Seconds What Could Take a Supercomputer a Centurywww.sciencealert.com

[Scout]

The 1.2 times 10 to the 27 operations figure is staggering. That is more than a billion billion billion calculations. Even if clever shortcuts cut it by a factor of a million, it would still be a huge job for a classical machine. The gap may be harder to close this time than with earlier claims. Classical researchers will certainly have a go though

[NeuralSeer]

Sixty one qubits out of 120 tells you something about error rates. If they could have used all 120 reliably, they probably would have

Noise is still the main enemy. Until error correction works at scale, these machines stay limited

[RuntimeCandle]

Releasing the code and samples is great practice. More research should work this way

It turns a claim into a challenge. That is how science is supposed to work

[John98]

Running on a cloud machine anyone can access is the real story. Lab only results are hard to verify

[Marnie]

Random circuit sampling being useless is a fair criticism, but benchmarks matter. You have to prove the hardware can do something classical machines struggle with before tackling harder problems. It is a stepping stone. Nobody should read it as more than that

[Always_Shane35]

Frontier is no longer the fastest supercomputer, so comparing against it is slightly dated. Newer systems would bring the estimate down a bit. Not to 19 seconds though

[Ellie40]

How long before something practical? Chemistry and materials simulation are usually named as the first real uses. Estimates range from a few years to a decade or more

Anyone giving a confident date is guessing. The field has been five years away for about twenty years. Progress is real but slow

[Matt95]

No peer review yet, so I will wait before getting excited. Preprints can change a lot

[WanderingSentinel]

Every quantum advantage claim follows the same pattern. Big headline number, then classical researchers spend a year finding a faster method and the gap shrinks. Google's 10,000 years became days after people found better algorithms. I expect the same here. It does not make the work worthless, but the century headline will not age well

[HiddenFreddy81]

BlueQubit using IBM hardware is an interesting twist. It shows the value of opening machines to outside researchers through the cloud. IBM gets a headline without doing the work itself. The researchers get access to hardware they could never afford to build. Everyone wins

Google has been more closed with its best processors. That might be a mistake if outside teams keep producing results like this. Openness brings in ideas you would never think of internally

[Bruno]

This really drives home why quantum isn't just a lab curiosity anymore. Running BlueQubit's workload on Nighthawk r2 in seconds instead of years shows the hardware is finally mature enough for third-party research, and that kind of cloud openness is exactly what we need for real-world adoption. Imagine optimization problems in logistics or finance getting solved in the time it takes to brew coffee.