NVCaffe 0.17.4

Bug fix release: cuDNN v8, CUDA 11.1, Ubuntu 20.04 etc.

NVCaffe 0.17.3

• CUDA 10.1 support
• SSD improved
• Bug fixes and performance improvements

NVCaffe 0.17.2

Bug fix release

NVCaffe 0.17.1

• CUDA 10 compatibility
• Turing GPU support
• Multinode support (experimental)
• TRT integration (experimental)
• Ubuntu 18.04 compatibility
• Adaptive scaler for mixed precision training
• Bug fixes and performance improvements

NVCaffe 0.17.0

• SSD Object Detection
• CuDNN support for Deconvolution
• Recurrent Networks
• Bug Fixes
• Performance improvements

NVCaffe 0.16.6

This release delivers:

• LARC
• ImageDataLayer revamp
• CuDNN support for dilation
• CPU_ONLY mode removed
• Snapshotting of last N epochs
• Multiple fixes and improvements

NVCaffe 0.16.5

01.01.2018 release with major performance, accuracy and memory usage improvements.

NVCaffe 0.16.4

This release brings the following new features, fixes and optimizations:

• New fast grouping supported by cuDNN v7 is now integrated with NVCaffe.
• New augmentation algorithms.
• Convolution workspace management simplified and improved to provide more memory for better algorithms.
• Stability improvements
• Performance improvements.
• Fixed issues #391 #351 (Matlab support is back) #357 (ImageDataLayer deadlock)

NVCaffe 0.16.3

July 2017 release brings cuDNN v7 and Volta Tensor Cores support plus other improvements and bug fixes.

NVCaffe 0.16.2

Enhancements and bug fixes:

• CUDA 9 support
• Data Cache & Shuffle
• Global Gradient Scaling for 16-bit float math support
• Better Python support and error handling
• Improved determinism across different number of GPUs
• Fixed #347 #353 #255