common.title

Docs
Quantum Circuit
TYTAN CLOUD

QUANTUM GAMING


Overview
Contact
Event
Project
Research

Terms of service (Web service)

Terms of service (Quantum and ML Cloud service)

Privacy policy


Sign in
Sign up
common.title

Accelerating QAOA with CUDA-Q and H100 GPU: Solving the MaxCut Problem via Quantum Simulation

Yuichiro Minato

2025/04/01 07:35

Accelerating QAOA with CUDA-Q and H100 GPU: Solving the MaxCut Problem via Quantum Simulation

We conducted a simulation of the MaxCut problem using one of the most prominent quantum algorithms—QAOA (Quantum Approximate Optimization Algorithm). In this study, we used CUDA-Q to compare the execution speed and optimization accuracy between CPU and GPU environments.

All experiments were performed with the same random seed and identical problem setup (graph structure) for fair comparison.

🎯 What is the MaxCut Problem?

The MaxCut problem involves dividing the nodes of a graph into two sets such that the number (or total weight) of edges between the sets is maximized. QAOA is a quantum algorithm designed to approximately solve such combinatorial optimization problems using quantum circuits.

🧪 Experimental Setup

  • CUDA-Q framework used
  • QAOA depth: 2 layers
  • Optimizer: Nelder-Mead
  • Fixed random seed (for reproducibility)
  • Same Erdős–Rényi random graph used across all runs
  • CPU: Google Colab (standard CPU)
  • GPU: NVIDIA H100 80GB

⚙️ Execution Results

Example of the problem graph:
image

Example of optimization steps:
image

Comparison of qubit count vs. execution time shows that as the number of qubits increases, the resulting MaxCut value tends to increase as well.

However, since QAOA is a combinatorial optimization method, the execution time heavily depends on how quickly the optimizer converges, which can vary significantly. This makes it difficult to obtain a smooth or consistent execution time graph.

image

🔍 Discussion

  • Even for the same problem, the GPU completed optimization much faster than the CPU.
  • The GPU was able to handle 15 to 20 qubits very quickly, but for 30 qubits, execution time exceeded 200 seconds due to increased circuit complexity.
  • Optimization performed directly on the GPU greatly accelerated the overall process.

🚀 Conclusion

GPU-based quantum simulation has proven to be highly effective for mid-scale algorithms like QAOA. Going forward, we plan to further evaluate GPU performance with larger-scale circuits and other quantum algorithms.

© 2025, blueqat Inc. All rights reserved