Support cuTENSOR 1.4.0

Tasks

Read Release Notes (cuTENSOR / cuSPARSELt / cuDNN / NCCL).
Update FlexCI configuration to test the version.
- Fix .pfnci/schema.yaml and .pfnci/matirx.yaml #6187
- Backport. #6330
Update the library installer.
- Fix cupyx/tools/install_library.py. #6187
- Backport. #6330
Documentation
- Update Requirements section in Installation Guide #6187
- Backport. #6330
Implement new features using the version, if any. #6187

Issue Analytics

State:
Created 2 years ago
Reactions:2
Comments:5 (5 by maintainers)

Top GitHub Comments

2reactions

takagicommented, Nov 30, 2021

cuTENSOR 1.4 has preview support for distributed, multi-GPU tensor operations, whose API is prefixed as cutensorMgXxx.

1reaction

leofangcommented, Dec 1, 2021

Sorry, ignore my comment above, that was my misunderstanding.

Top Results From Across the Web

cuTENSOR Data Types

cudaDataType_t is an enumeration of the types supported by CUDA libraries. cuTENSOR supports real FP16, BF16, FP32 and FP64 as well as complex...

NVIDIA cuTENSOR 1.4 now available

Supports up to 64-dimensional tensors. Supports distributed, multi-GPU tensor operations. Improved tensor contraction performance model (i.e., ...

Cutensor :: Anaconda.org

cuTENSOR is used to accelerate applications in the areas of deep learning training and inference, computer vision, quantum chemistry and computational physics.

Help setting up CUDA.jl and cuTENSORS - GPU

Hi all, I am having trouble using CUDA.jl and cuTENSOR and would ... CUDNN: 8.30.2 (for CUDA 11.5.0) │ - CUTENSOR: 1.4.0 (for...

CUDA.jl 3.5-3.8

When your hardware supports it, CUDA.jl will automatically enable ... Thanks to @kshyatt for help with this change! ... CUTENSOR 1.4.0.