Nvidia says it’s doubling down on the data center. At this week’s GPU Technology Conference (GTC) the chip company made this very clear through a series of announcements — including a new partnership with Amazon Web Services (AWS) and technologies that push artificial intelligence (AI) to the data center.
Nvidia has been heavily bolstering its data center business this year. In February, following a disappointing quarter, Nvidia founder and CEO Jensen Huang wrote in a shareholder letter that its new data center GPU applications and partnerships will drive growth this year. So it seems the company is following through on this promise.
Last week Nvidia purchased Mellanox Technologies for $6.9 billion in a direct attempt to boost its data center business and help it better compete against Intel in the space.
Of the purchase Huang said that Mellanox’s interconnect technology complements its strength in data centers and high-performance computing. Prior to the purchase, the two companies partnered in the data center space for many years. This includes building servers that combine Nvidia graphics processing units (GPUs) and Mellanox Interconnects.
Even prior to the acquisition the chipmaker launched a new program in February with a number of data center operators to give them access to its DGX systems. These systems are purpose-built for AI, machine learning, and deep learning.
And at GTC, Nvidia took this one step further.
Pushing AI to the Data Center
At the event, the company highlighted DGX POD, which it says provides a blueprint for data center architects seeking to optimize their compute, storage, and network infrastructure to meet the needs of AI applications.
Nvidia used this DGX system to build a GPU-powered AI infrastructure, which it calls DGX SATURNV and says will improve the performance of AI services and applications.
According to Nvidia, this infrastructure has been adopted across the data center industry by Arista Networks, Cisco, DDN, Dell EMC, IBM Storage, Mellanox, NetApp, and PureStorage. Each of these companies, it says, offer their own turnkey infrastructure systems using DGX.
Nvidia also made several announcements about its DGX partnerships.
Dell EMC released general availability of a new reference architecture that combines its Isilon All-Flash NAS storage with Nvidia’s DGX-1 servers for AI and deep learning workloads. NetApp build its own AI architecture based on DGX-1 and Nvidia’s GPU networking fabric. And Pure Storage launched a hyperscale machine learning platform that is powered by DGX to build and produce AI initiatives.
Additionally, Nvidia DGX-2 systems are now certified for Red Hat Enterprise Linux.
In terms of AI, Nvidia released new data science acceleration libraries, named CUDA-X AI, that accelerate the use of machine learning and analytics in a number of instances, including in enterprise data centers. These integrate with its Tensor Core GPUs to serve the end-to-end AI pipeline.
CUDA-X AI has been adopted by AWS, Google Cloud, and Microsoft Azure to accelerate the use of data science across their services.
Nvidia Lands AWS
AWS will now use Nvidia’s T4 data center chips. According to the chipmaker, the T4 Tensor Core GPUs will be deployed through AWS’ Elastic Compute Cloud (EC2) G4 instances in the coming weeks. G4 is the newest generation of GPU equipped EC2 instances.
In a blog post, AWS Chief Evangelist Jeff Barr wrote that the addition of Nvidia’s T4 data center chips will bring support for machine learning interfacing, video processing, and increased graphics performance.
AWS is just the most recent cloud provider to use Nvidia data center chips. Google Cloud Platform said it would use the T4 chips its data centers. Nvidia has said that Baidu and Tencent also use the T4 chips and that Alibaba will adopt them in the future as well.
In addition to the T4 announcement, AWS and Nvidia are integrating AWS IoT Greengrass and Nvidia Jetson, its AI set of software tools and software development kits, to deploy AI and deep learning across connected devices.