NVCaffe Release 17.09
NVCaffe Release 17.09 (PDF)
The NVIDIA container image of Caffe, release 17.09, is available.
NVCaffe container image version 17.09 is based on NVCaffe 0.16.4.
Contents of NVCaffe
This container image contains the complete source of the version of NVCaffe in
/opt/caffe. It is pre-built and installed into the
/usr/local/[bin,share,lib] directories in the container image.
The container also includes the following:
- Ubuntu 16.04
- NVIDIA CUDA® 9.0
- NVIDIA CUDA® Deep Neural Network library™ (cuDNN) 7.0.2
- NCCL 2.0.5 (optimized for NVLink™ )
Release 17.09 is based on CUDA 9, which requires NVIDIA Driver release 384.xx.
Key Features and Enhancements
This NVCaffe release includes the following key features and enhancements.
- Added the img_rand_resize_lower and img_rand_resize_upper parameters which specify that the variable-sized input image should be randomly resized.
- Added the global_grad_scale parameter which defines the constant C used to improve precision of back-propagation for float16 data storage.
- Added the default_cudnn_math_override parameter which sets the default
cudnn_math_overridevalue for every layer, if applicable.
- Added the store_blobs_in_old_format parameter which stores blobs in an old, less efficient BVLC-compatible format.
- Added the cudnn_math_override parameter which sets the default
cudnnMathType_tvalue for all cuDNN-based computations in the current layer, if applicable, otherwise, it is ignored.
- Added the ELUParameter parameter stores parameters used by ELULayer.
- Added the lambda parameter which is used for Scaled Exponential Linear Unit (SELU™).
- Latest version of CUDA
- Latest version of cuDNN
- Latest version of NCCL
- Ubuntu 16.04 with August 2017 updates
There are no known issues in this release.