Chinese Academy of Sciences self-developed HIP-DFPT software: Domestic GPU quantum perturbation calculation achieved breakthrough

Kuai Technology reported on July 26 that according to media reports, the team of 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 common in quantum perturbation computing, this software uses technical means such as micro-task reorganization, GPU memory optimization, and multi-stream asynchronous pipelining to reorganize a large number of fine-grained computing tasks into micro-task units that are more suitable for GPU execution, effectively improving thread utilization and data access efficiency, while reducing communication and data transmission overhead. Quantum perturbation theory is an important theoretical tool for studying the response properties of materials and molecular systems. It can reveal microphysical properties such as phonon spectrum, polarizability, and Raman spectrum. It has wide application value in the fields of materials science, condensed matter physics, and molecular simulation. In terms of load balancing, the scientific research team proposed an AI performance modeling method for complex branch codes, which can quickly predict task execution time with only a small amount of sampled data. Based on this, they designed a hybrid multi-level load balancing algorithm to achieve collaborative optimization of dynamic scheduling between nodes and within the GPU. Experimental results show that HIP-DFPT can be successfully extended to 8192 domestic GPUs, supporting full electronic precision simulation of systems up to 200,000 atoms in size, and increasing atomic throughput by 70%. This achievement fully verifies the software's parallel expansion capabilities and engineering application potential in large-scale quantum perturbation calculations, and also provides a new technical path for the intelligent optimization of complex scientific computing programs on heterogeneous GPU clusters.