JUPITER exascale supercomputer simulates 50-qubit quantum computer in full for the first time, breaking the 48-qubit record set in 2019

[Sharp Shannon]

German researchers at Jülich Supercomputing Centre using JUPITER, Europe's first exascale supercomputer, fully simulated a 50-qubit quantum computer for the first time, published on May 11th. The previous record of 48 qubits was set on Japan's K computer in 2019. The simulation required approximately 36 petabytes of state vector storage and used JUPITER's heterogeneous CPU-GPU architecture with novel load balancing across partitions.

The purpose of classical simulation at this scale is validating quantum algorithms like VQE and QAOA before running them on real hardware with unacceptable error rates, and establishing the classical simulability boundary.

JUPITER supercomputer breaks world record with 50-qubit quantum simulationwww.sciencedaily.com

[Maxximus]

36 petabytes of state vector is the number that makes the exponential scaling of quantum simulation visceral. Adding two more qubits after this will require four times that storage. The classical simulation wall is not a theoretical limit, it is a storage and bandwidth limit

[Louise84]

Germany investing in JUPITER and using it for quantum algorithm validation while also funding quantum hardware research is the integrated national strategy that produces progress faster than treating the two as separate

[Demi-Q]

The Jülich and NVIDIA combination of CPU and GPU in JUPITER is the right architecture for this workload. State vector simulation requires both memory bandwidth and floating point throughput that no single processor type provides optimally

[Nina81]

The practical value is the algorithm validation angle. Running VQE or QAOA on real hardware with current error rates produces results that are hard to interpret. Simulating the ideal behaviour classically lets you validate whether the algorithm is correct before adding hardware noise

[TheLegendBrett88]

50 qubits classically simulated is interesting but the more important metric is where the boundary of advantage lies for specific algorithm classes. The simulation doesn't tell you where quantum hardware becomes necessary

[WildManSteve40]

JUPITER going online for European HPC has been significant for multiple research areas. The quantum simulation use case is a natural first application of the raw compute

[Ben]

The 2019 K computer 48-qubit record lasting until 2026 illustrates how difficult this scaling is. Seven years and the advance is two qubits. The exponential cost makes this work extraordinary

[StevenArroyo]

Simulating the 50-qubit ideal behaviour while simultaneously running real 50-qubit experiments on IBM or Google hardware would let you characterise the hardware noise in a principled way. That experimental design is the valuable application

[VoidSentinel]

The Teraquop simulations mentioned in the Quantum Computing Report for error rate characterisation are related work. Understanding qubit error behaviour in real time and in simulation together accelerates error correction development