cuNNQS-SCI: A Fully GPU-Accelerated Framework for High-Performance Configuration Interaction Selection withNeural Network QQantum States

2026-04-17 · Source: Artificial Intelligence · Field: Science & Research — Mathematics & Computational Sciences, Physical Sciences & Chemistry, Engineering & Applied Sciences · Depth: Expert, quick

Summary

cuNNQS-SCI is a new, fully GPU-accelerated framework designed to enhance the scalability and performance of the Neural Network Quantum State-Selected Configuration Interaction (NNQS-SCI) method. The original NNQS-SCI, while accurate, faced severe limitations in larger systems due to its hybrid CPU-GPU architecture, specifically CPU-based global de-duplication and host-resident coupled-configuration generation. cuNNQS-SCI addresses these bottlenecks by integrating a distributed, load-balanced global de-duplication algorithm, employing fine-grained CUDA kernels for exact coupled configuration generation, and incorporating a GPU memory-centric runtime with pooling, streaming mini-batches, and overlapped offloading. This design allows for significantly larger configuration spaces and shifts the computational bottleneck to on-device inference. Evaluated on an NVIDIA A100 cluster with 64 GPUs, cuNNQS-SCI achieved up to a 2.32X end-to-end speedup over the baseline NNQS-SCI while maintaining chemical accuracy and demonstrating over 90% parallel efficiency.

Key takeaway

For AI Scientists and Research Scientists working on quantum chemistry simulations, cuNNQS-SCI offers a significant advancement by enabling the application of NNQS-SCI to much larger systems. Its 2.32X speedup and high parallel efficiency on GPU clusters mean you can tackle previously intractable problems with improved computational throughput. Consider adopting fully GPU-accelerated frameworks to overcome CPU-GPU communication bottlenecks in your high-performance computing workflows.

Key insights

Fully GPU-accelerating NNQS-SCI overcomes CPU bottlenecks, enabling larger quantum system simulations with significant speedups.

Principles

• Distributed de-duplication minimizes communication overhead.
• Fine-grained CUDA kernels optimize configuration generation.
• GPU memory-centric runtime breaks single-GPU memory limits.

Method

cuNNQS-SCI integrates distributed de-duplication, uses specialized CUDA kernels for coupled configuration generation, and employs a GPU memory-centric runtime with pooling, streaming mini-batches, and overlapped offloading.

In practice

• Utilize distributed de-duplication for large-scale data.
• Implement CUDA kernels for compute-intensive tasks.
• Employ GPU memory pooling for memory-bound applications.

Topics

• cuNNQS-SCI
• Neural Network Quantum States
• Configuration Interaction Selection
• GPU Acceleration
• Distributed Computing

Best for: AI Scientist, Research Scientist

Related on AIssential

• cuNNQS-SCI: A Fully GPU-Accelerated Framework for High-Performance Configuration Interaction Selection with Neural Network Quantum States — Fully GPU-accelerated quantum chemistry framework cuNNQS-SCI significantly boosts scalability and performance by eliminating CPU bottlenecks.
• Alice & Bob expands Series B with investment from NVIDIA’s venture arm — The future of quantum computing is hybrid, integrating quantum and classical systems for real-world problem-solving.
• Running Variational Quantum Eigensolver with Qiskit Aer on AMD Instinct — GPU-accelerated quantum circuit simulation significantly extends the practical scale for Variational Quantum Eigensolver (VQE) in quantum chemistry.
• Scalable On-Hardware Training of Quantum Neural Networks and Application to Clinical Data Imputation — A novel QNN training framework reduces hardware gradient estimation costs from O(n^2) to O(log n), enabling scalable quantum computing applications.
• The future of computing is now with Jay Gambetta - IBM Think 2026 — Quantum-centric supercomputing, integrating classical and quantum hardware, is achieving practical advantage in complex scientific problems.
• How AI Will Change Quantum Computing | NVIDIA AI Podcast Ep. 294 — AI models, like NVIDIA Ising, are critical for accelerating quantum computing by addressing key challenges such as error correction and calibration.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.