State-Vector Simulation of 36-Qubit Ekerå–Håstad Factoring Using FP8 Storage on Commodity NVIDIA GB10 Hardware
We report an exact state-vector simulation of the semiclassically folded Ekerå–Håstad (EH) short discrete-logarithm factoring circuit up to 36 logical qubits on commodity NVIDIA GB10 unified-memory hardware (121 GiB per node). Pairing IEEE FP8 (e4m3, 2 bytes per complex amplitude) state-vector storage with FP32/FP64 arithmetic kernels and host-side double-precision global scale tracking enables factoring 35-bit balanced RSA semiprimes on two GB10 nodes (242 GiB aggregate memory), while the same single-node architecture (np = 1, without MPI) scales to 35 logical qubits on a single GB10 or proportionally on discrete-VRAM commodity GPUs. Over 100 GbE RoCE (HPCX OpenMPI + UCX rc), the verified 36-qubit factorization of N = 24,966,154,091 = 157007 x 159013 completes its successful 53-round run in 1,339.8 s (2,688.7 s across two attempts). Both the folded EH circuit and 2-byte/amplitude rescaled storage follow established prior art (Ekerå; De Raedt et al., JUQCS-A); this technical note serves as an open-access defensive publication and engineering benchmark. Both English and Chinese PDF editions are included.
Authors
- Rui Liu
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24377952
- Primary Topic
- Quantum Computing Algorithms and Architecture
- Type
- article
- Field-Weighted Citation Impact
- 0.00