Hello, I previously conducted benchmarks on various GPU machines. This time, the measurement is done on a single GPU. In addition to the previous multi-GPU configurations, we had implemented configurations of 4 units of V100 VRAM16G, 8 units of VRAM32G, and 8 units of A100 VRAM40G. This time, I conducted benchmarks for NVIDIA&#39;s cuQuantum/cuStateVec + Qiskit-Aer-GPU using T4 VRAM16GB and the consumer-grade RTX4090 VRAM24G. The circuits for the Quantum Volume (QV) had a depth of 10. The results are as follows. <img src="https://assets.blueqat.com/public/uploads/us-east-2:4805ff4b-c3cc-4344-b165-86544c34d0bf/2023/10/10/lab-redirects-1-JupyterLab.png"/> The 4090 is impressively fast. I thought the T4 was also fast, but the 4090 was quicker. As expected, it can&#39;t really compete with multiple GPUs. With 24G of VRAM, 30 qubits was the limit.

[Quantum Computing] NVIDIA cuQuantum / cuStateVec NVIDIA T4 / RTX4090 water-cooled quantum volume benchmark.

Yuichiro Minato