Skip to content

RELEASE_NOTES_v1.0.2.md

Latest commit

 

History

History
61 lines (46 loc) · 2.26 KB

RELEASE_NOTES_v1.0.2.md

File metadata and controls

61 lines (46 loc) · 2.26 KB
Raw

IQC v1.0.2 Binary User Distribution

This release provides a binary-only offline installer for Linux x86_64, CPython 3.12, CUDA 12 runtime wheels, and NVIDIA A100/A800 GPUs (sm_80).

IQC binary releases are licensed under the PolyForm Noncommercial License 1.0.0. Commercial use requires a separate commercial license from the IQC authors.

Built from IQC source commit:

  • 932ace48ef8b998176d8850b9927aea0a06d19c0

Changes Since v1.0.1

  • Made the IQC GPU chemistry runtime self-contained, including molecular, basis-set, SCF, XC, integral, symmetry, and CUDA runtime components.
  • Added and optimized multi-GPU execution for SCF, FTROKS, direct JK, XC grids, and SATDDFT response calculations.
  • Added CUDA FFI paths for direct JK and semilocal XC contractions, together with bounded-memory and sharded response workflows for larger calculations.
  • Added FT-SATDDFT transition-density and oscillator-strength analysis.
  • Added analytic GPU molecular dipole integrals and SATDDFT oscillator strengths.
  • Upgraded the packaged runtime to JAX 0.11 and refreshed public examples and native installation support.

Assets

  • iqc-linux-x64-cp312-cu12-sm80.tar.gz.part00
  • iqc-linux-x64-cp312-cu12-sm80.tar.gz.part01
  • PARTS_SHA256SUMS.txt
  • SHA256SUMS.txt

Join the parts before installation:

sha256sum -c PARTS_SHA256SUMS.txt
cat iqc-linux-x64-cp312-cu12-sm80.tar.gz.part* > iqc-linux-x64-cp312-cu12-sm80.tar.gz
sha256sum -c SHA256SUMS.txt
tar -xf iqc-linux-x64-cp312-cu12-sm80.tar.gz
cd iqc-linux-x64-cp312-cu12-sm80
bash install.sh

On macOS, replace sha256sum -c with shasum -a 256 -c.

Validation

The package was validated with a fresh user-style offline install from the generated release tarball on an NVIDIA A100-SXM4-80GB worker, with the process restricted to one GPU. The installer verification passed with:

  • IQC installed as version 1.0.2
  • 133 IQC compiled .so modules installed
  • no non-__init__.py IQC Python source files exposed
  • JAX and CuPy detecting the selected CUDA device
  • the IQC direct-JK CUDA FFI loading successfully
  • CUDA/JAX smoke testing passing
  • installed-package imports for SATDDFT transition analysis passing
  • analytic dipole-integral value and origin-shift smoke testing passing

Verification finished with IQC VERIFY OK.