This is the release note of v5.4.0. This is the final release of v5.x series. See here for the complete list of solved issues and merged PRs.

Highlights

• CUDA 10.1 and cuDNN 7.5 are now supported. CuPy also starts to compile for compute capability 7.5 for Turing GPUs.

Enhancements

• Avoid using pytest attributes during import (#2057)
• Fix fp16 issue in batch normalization (#2094, thanks @anaruse!)
• Keep backward compatibility on cupy.cudnn.batch_normalization_forward_training (#2094)
• Support CUDA 10.1 + cuDNN 7.5 + Turing (#2123)

Bug Fixes

• Ensure that sparse matrix shapes are always a tuple of int (#2082, thanks @grlee77!)
• Fix assigning from complex to float (#2092)
• Check array contiguity in copy (#2093)
• Fix assertion error in _Chunk.split (#2112, thanks @liwt31!)
• Support dltensor with strides of NULL (#2119, thanks @crcrpar!)
• Avoid sharing handles between threads (#2122)

Code Fixes

• Remove unused variable (#2108, thanks @crcrpar!)

Documentation

• Add NCCL v2.4 support to docs (#2089)
• Add strides to docstring of ndarray (#2109, thanks @crcrpar!)

Tests

• Do not fail by warnings when building docs (#2086)