Nvidia brings Kubernetes to GPUs for the first time

Developers first in the queue to test TensorRT and TensorFlow integration tools running on Nvidia GPUs

With Kubernetes fast becoming the de facto standard for deploying cloud applications Nvidia has been quick to jump on the bandwagon, announcing availability of Kubernetes on its GPUs for the first time.

At the moment, the technology is being made available to developers to enable them to test the software and to give their feedback.

The initiative was announced at the Computer Vision and Pattern Recognition (CVPR) conference on Tuesday, the release of the new candidate version of the AI tools let developers and DevOps engineers build and deploy GPU-accelerated deep learning training or inference applications on multi-cloud GPU clusters, at scale.

They also enable the automation of deployment, maintenance, scheduling and operation of GPU-accelerated application containers. This should help developers handle the growing number of AI-powered applications and services, Nvidia claimed.

As well as the announcement of Kubernetes, the company also said its TensorRT and TensorFlow integration tools are now also available for developers.

Orginally unveiled at the company's GTC conference in San Francisco, California in March, the TensorRT 4 software helps to accelerate deep learning inference across a broad range of applications, such as highly accurate INT8 and FP16 network execution, which can cut data center costs by up to 70 per cent, the firm says.

"TensorRT 4 can be used to rapidly optimise, validate and deploy trained neural networks in hyperscale datacenters, embedded and automotive GPU platforms," Nvidia said.

It continued: "The software delivers up to 190 times faster deep learning inference compared with CPUs for common applications such as computer vision, neural machine translation, automatic speech recognition, speech synthesis and recommendation systems."

TensorRT has been integrated by Google and Nvidia engineers into TensorFlow 1.7, making it easier to run deep learning inference applications on GPUs.

Nvidia is also using the conference to demonstrate an early release of APEX, an open-source extension that enables researchers using PyTorch to maximise deep learning training performance on Volta GPUs.

That software is available for anyone who wants it via GitHub, but is currently in beta. Nevertheless, it automatically enables mixed-precision training making it a lot easier to use.