Skip to content

VMEX 0.11.9

Choose a tag to compare

@rogeriojorge rogeriojorge released this 08 Oct 01:24
· 257 commits to main since this release
6f876ad
  • On a GPU, every fused run of kernels is captured into a CUDA graph
    (--xla_gpu_graph_min_graph_size=1 unless XLA_FLAGS is set): 8-11 %
    faster per solver iteration on an RTX A4000, results unchanged (#563).
  • vmex --neoclassical: DKX neoclassical transport (#548).
  • Optimization examples: dimensionless Mercier and resistive hinges for the
    bootstrap examples, a DESC-style ballooning hinge with a COBRAVMEC 3-D
    gate, and larger mode-2 budgets for the scalar and SciPy lanes
    (#555, #556, #557, #560).