Release Notes#

This section documents major changes to the MathDx package and its components. Detailed changes for individual libraries are in the corresponding library’s release notes.

26.10.0#

Noteworthy changes in this release:

  • cuFFTDx: Performance retuning, new Transposition, ThreadLayout, and ThreadLinearization operators.

  • cuBLASDx: Pipelined GEMM now supports thread-block clusters.

  • cuSolverDx: Experimental cluster execution for GEQRF.

  • cuRANDDx added support for Windows / MSVC.

  • Support for Rubin (SM107) GPU architecture was added to all libraries.

  • MathDx 26.10.0 package ships with CUTLASS 4.8.0, which is the minimum supported and the recommended version to use.

  • Support for GCC 7 is deprecated in all libraries and will be removed in a future release. A warning is emitted when compiling with GCC 7. We recommend GCC 10 or newer.

Please refer to the release notes of the corresponding libraries for more details.

Components#

26.06.1#

This patch release includes bug fixes for cuBLASDx and cuFFTDx, and minor changes to the documentation of Dx libraries.

Please refer to the release notes of the corresponding libraries for more details.

Components#

26.06.0#

Major changes in this release:

  • Deprecated static LTO libraries in cuSolverDx and nvCOMPDx.

  • cuBLASDx added flexible-precision floating-point emulation GEMM along with improvements to pipelined GEMM.

  • cuSolverDx added experimental cluster execution for POTRF.

  • nvCOMPDx added support for MSVC when using fatbin library.

  • MathDx 26.06.0 package ships with CUTLASS 4.5.2, and this is the recommended version to use.

Please refer to the release notes of the corresponding libraries for more details.

Components#

26.03.0#

Major changes in this release:

  • Dropped support for CUDA Toolkit 12.X.

  • Dropped support for Volta architecture (SM70/SM72).

  • cuBLASDx 0.6.0 adds TRSM functionality and is now an LTO library.

Please refer to the release notes of the corresponding libraries for more details.

Components#

25.12.1#

25.12.1 patch release includes minor changes to the documentation and samples of cuBLASDx and cuFFTDx.

Components#

25.12.0#

In the 25.12.0 release, cuFFTDx, cuBLASDx, and cuSolverDx received significant updates. Please refer to the release notes of the corresponding libraries for more details.

Components#