Chinese Academy of Sciences' Self-developed HIP-DFPT Software: Domestic GPU Achieves Breakthrough in Quantum Perturbation Calculations

according to media outlets, a team from the Computer Network Information Center of the Chinese Academy of Sciences recently launched a large-scale quantum perturbation computing software for heterogeneous GPU clusters—HIP-DFPT.

Aiming at the irregular load characteristics commonly found in quantum perturbation computing, this software employs micro-task reorganization, GPU memory optimization, and multi-stream asynchronous pipelining techniques to reorganize a large number of fine-grained computing tasks into micro-task units more suitable for GPU execution, effectively improving thread utilization and data access efficiency while reducing communication and data transfer overhead.

Quantum perturbation theory is an important theoretical tool for studying the response properties of materials and molecular systems. It can reveal microscopic physical characteristics such as phonon spectra, polarizability, and Raman spectra, holding broad application value in fields like materials science, condensed matter physics, and molecular simulation.

In terms of load balancing, the research team proposed an AI-based performance modeling method for complex branching code. This method quickly predicts task execution time with only a small amount of sampling data, and accordingly designs a hybrid multi-level load balancing algorithm to achieve collaborative optimization of dynamic scheduling between nodes and within GPUs.

Experimental results show that HIP-DFPT can be successfully scaled to 8,192 domestic GPUs, supporting full electronic precision simulations of systems up to 200,000 atoms in size, with atomic throughput increased by 70%.

This achievement fully validates the software's parallel scalability and engineering application potential in large-scale quantum perturbation computing, and also provides a new technical path for intelligent optimization of complex scientific computing programs on heterogeneous GPU clusters.