# NVIDIA and Jülich Simulate a 50-Qubit Quantum Computer on JUPITER

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/35183
Published: 2026-10-20T01:28:32.000Z
Updated: 2026-10-20T01:28:32.000Z
Section: Technology

> The 2025 run used 4,096 nodes and byte encoding to keep the simulated state within 2,048 TiB of memory.

NVIDIA and the Jülich Supercomputing Centre used JUPITER to simulate a universal 50-qubit quantum computer in a 2025 run, giving researchers a full-state reference for examining circuits without quantum hardware. The result was announced Nov. 11, 2025, and surpassed the team’s previous 48-qubit simulation on Japan’s K computer in 2019.

The software, JUQCS-50, ran across 4,096 JUPITER nodes equipped with 16,384 NVIDIA GH200 superchips. A full-state simulation tracks the system across combinations of qubit values. Each additional qubit doubles the amount of state information the computer must store.

At 50 qubits, the encoded state required 2,048 TiB of memory. The same state would have taken 16,384 TiB in uncompressed double precision. Byte encoding reduced the requirement eightfold, though it can lower numerical precision in some cases. When GPU memory alone was insufficient, the software also used memory attached to the CPUs.

The 16.6-fold speed difference came from a separate, matched benchmark at 36 qubits. JUPITER took 2.83 seconds per gate in that test, compared with 47.03 seconds on the K computer. Those timings do not compare the 50-qubit run with the earlier 48-qubit simulation.

Kristel Michielsen, director of the Jülich Supercomputing Centre, said, “This use case illustrates how closely progress in high-performance computing and quantum research are intertwined today.” The full simulation offers an idealized reference for testing circuits, while physical quantum processors contend with noise and gate errors.

NVIDIA announced a separate JUPITER project on Oct. 19, 2026, simulating a 104-qubit fault-tolerance protocol that models hardware errors. That project tracks a smaller encoded state as it changes over time, so its qubit count describes a different task from JUQCS-50’s full-state simulation.