Skip to content

v0.6.3rc1 (pre-release)

Pre-release
Pre-release

Choose a tag to compare

@waltsims waltsims released this 21 Jun 15:29
· 2 commits to master since this release
e303fde

Pre-release for hardware validation. First k-wave-python release on the unified binary pipeline (kspacefirstorder-unified v1.4.2). Install with:

pip install --pre k-wave-python
# or pin exact:
pip install k-wave-python==0.6.3rc1

What's new

  • CUDA arch coverage: 16 SMs. Now ships Turing through every shipping Blackwell variant — B200/GB200, B300/GB300, Jetson Thor, RTX 50xx, RTX PRO 6000, GB10/DGX Spark. Maxwell/Pascal/Volta dropped (CUDA Toolkit 13.0 removed them).
  • Windows CUDA binary regression fixed. v1.4.1 silently shipped only sm_75 SASS (3.4 MB). v0.6.3rc1 ships the full multi-arch fat binary (14.8 MB + cuFFT runtime) via the new unified release.
  • Runtime warning when backend="cpp" + device="gpu" is requested on a card with compute capability < 7.5 — clear message pointing to backend="python" or CUDA 12.x source build.
  • Single unified binary URL. No more per-mirror version drift; one release tag, one set of platform-suffixed assets, all built from one source-tree SHA.

Why pre-release

The binary install path was substantially rewritten: URL flip + rename-on-download + new CUDA arch list. Validate on real hardware (cloud + local Blackwell/Hopper/Ada/Ampere) before promoting to stable 0.6.3.


What's Changed

New Contributors

Full Changelog: v0.6.2...v0.6.3rc1