Apache
MXNet version 0.12 now adds two important features which is Sparse Tensors and
NVIDIA Volta GPU’s. The support for NVIDIA Volta GPU will allow the customers
to train convolutional neural network up to 3.5 times more swiftly than on the
Pascal GPUs. The Volta GPU architecture introduced the Tensor Cores that
enables mixed precision training. AWS Deep Learning AMIs is now pre-installed
with MXNet v0.12 and also with various deep learning frameworks. The support
for Sparse Tensor will help in efficiently store and compute tensors that will
enable the developers to carry out sparse matrix operations in storage and
train deep learning models faster and compute efficient manner.
No comments:
Post a Comment