view article
Figure 5
Performance scaling across CPUs and GPUs for different disorder levels σ. (a) Throughput versus atom count on 72 CPU cores; the dashed lines indicate the single-core roofline and a comparison with the existing earlier implementation in Rose-X. (b) Strong scaling on a single CPU node. (c) Asymptotic performance on an NVIDIA A40 (48 GB GDDR6, 6 MB L2 cache). (d) Asymptotic performance on an NVIDIA A100 (40 GB HBM2e, 25 MB L2 cache).

Journal logoJOURNAL OF
APPLIED
CRYSTALLOGRAPHY
ISSN: 1600-5767
Volume 59| Part 5| October 2026|
https://doi.org/10.1107/S1600576726007429