MathWorks announced integration of MATLAB and NVIDIA TensorRT through GPU Coder.

This collaboration will enable engineers and scientists to develop new artificial intelligence (AI) and deep learning models in MATLAB with higher performance and efficiency to meet the growing demands of data center, embedded, and automotive applications.

MATLAB provides a complete workflow for rapidly training, validating, and deploying deep learning models. Engineers can utilize GPU resources without additional programming, allowing them to focus on application development rather than performance tuning. The integration of NVIDIA TensorRT and GPU Coder allows deep learning models developed in MATLAB to run on NVIDIA GPUs with high throughput and low latency.

Compared to TensorFlow, CUDA code generated in MATLAB, combined with TensorRT, can deploy Alexnet with 5x better performance in deep learning prediction and VGG-16 with 1.25x better performance.

“As imaging, speech, sensor, and Internet of Things (IoT) technologies rapidly advance, development teams are seeking AI solutions that deliver improved performance and efficiency,” said David Rich, director of MathWorks. “Furthermore, the increasing complexity of deep learning models is putting tremendous pressure on engineers,” he said. “This technology integration between MathWorks and NVIDIA is expected to enable development teams to train deep learning models using MATLAB and NVIDIA GPUs to implement real-time predictions across any environment, from the cloud to data centers and embedded devices.”
?