Peking University team ports the ABACUS DFT code to Huawei's Kunpeng 920 and reports up to 10.7x speedups

The team says SVE builds ran 1.3 to 3.9 times faster than a mainstream x86 CPU at 28 cores on ten material systems and a matrix-unit build reached 10.7 times on a full node, with code pushed to GitHub; the figures are the team's own, with no independent benchmark.