MATLAB provides a workflow for training, validating, and deploying deep learning models. Engineers can use GPU resources without additional programming, allowing them to focus on their applications instead of tuning performance. The new integration of NVIDIA TensorRT with GPU Coder enables deep learning models developed in MATLAB to run on NVIDIA GPUs with high performance and low latency. Internal benchmarking shows that CUDA code generated by MATLAB combined with TensorRT can deploy Alexnet with 5x the performance of TensorFlow, and can deploy VGG-16 with 1.25x the performance of TensorFlow for deep learning inference.*
* All benchmark tests were run in MATLAB R2018a with GPU Coder, TensorRT 3.0.1, TensorFlow 1.6.0, CUDA 9.0 and cuDNN 7 on an NVIDIA TITAN Xp GPU, on a Linux machine with a 12-core Intel® Xeon® E5-1650 v3 processor and 64 GB of RAM.
