QunaSys uses 1,024 GPUs to rehearse a quantum-chemistry algorithm
The reported H4 calculation tests a long circuit on classical hardware, with accuracy and scale attached to the specific workload.
.jpg){kind=link}
QunaSys reports a classical simulation of a quantum-chemistry circuit exceeding 300 million gates, using 1,024 NVIDIA H100 GPUs on ABCI-Q. The company announcement is visibly dated October 27. This is a large GPU computation of a quantum algorithm, not execution on a fault-tolerant quantum processor.[1]
The accompanying SC26 poster abstract describes an H4 calculation reduced from a naive 99-, 1.78-billion-gate circuit to 28 simulated qubits and about 320 million gates. The reported run takes approximately 1.5 hours and produces a one-million-shot distribution whose peak is 0.53 millihartree from the exact full-configuration-interaction reference.[2]
Those figures identify a useful but bounded rehearsal: can the circuit construction, reduction and execution deliver the expected answer for this model? The company’s broader H2-to-H10 examples reach 42 simulated qubits with relaxed energy-error requirements. The stated H4 accuracy must not be transferred to that larger scale.[1]
A known reference is an advantage for this kind of software check. When the intended answer is available, developers can look for errors introduced by the transformations needed to make a long calculation manageable. Passing that check would support the implementation for the tested case; it would not demonstrate that the same resource requirements suffice for every molecule.
Gate count also needs its partner, circuit width. Many operations on a modest simulated state and fewer operations on a much wider state can create different computational demands. A single large number does not establish that one experiment is harder than another. The workload description is what lets readers decide which comparison is meaningful.
The SC26 entry is an author-supplied conference abstract, not an independent replication. Its listing schedules the poster for November 17. The company release and poster therefore provide two primary records of the reported result, but they do not supply two independent experimental confirmations.[2]
The practical follow-up is reproducible accounting: the exact circuit, reductions, accuracy target and complete classical cost. That would let another group test the same rehearsal and examine where it stops scaling. A successful classical validation can improve preparation for future quantum hardware without establishing a present quantum advantage.