NVIDIA publicizes a spate of recent improvements at Computex 2022

Article by KL Lim.

Picture, clockwise, from high left: NVIDIA VP for accelerated computing Ian Buck, senior VP for {hardware} engineering Brian Kelleher, director of product administration for accelerated computing Ying Yin Shih, CTO Michael Kagan, senior VP for GeForce Jeff Fisher, VP of embedded and edge computing Deepu Talla.

NVIDIA has introduced its newest improvements in knowledge middle, robotics, content material creation, and gaming in a digital keynote deal with on the opening day of Computex 2022 in Taipei.

Six NVIDIA leaders teamed as much as ship the keynote deal with, which lined advances from robotics to AI and silicon to software program and highlighted the work of its associate ecosystem. 

Reworking knowledge facilities to AI factories

NVIDIA VP for hyperscale and HPC Ian Buck kicked off the keynote by sharing how knowledge facilities are reworking into AI factories.

“This transformation requires us to re-imagine the info middle at each stage, from {hardware} to software program, from chips to infrastructure to programs,” he mentioned. 

This can drive large enterprise alternatives for NVIDIA’s companions in knowledge facilities, HPC, digital twins and cloud-based gaming, referencing a “half-trillion market alternative”.

Powering these trendy AI factories requires end-to-end innovation at each stage. With knowledge facilities turning into AI factories, knowledge processing is important. 

These embrace NVIDIA Hopper GPUs, NVIDIA Grace CPUs and NVIDIA BlueField DPUs because the constructing blocks networked collectively by NVIDIA Quantum and Spectrum switches. 

“The Bluefield DPU together with the Quantum and Spectrum networking switches comprise the infrastructure platform for the AI manufacturing unit of the long run,” mentioned NVIDIA CTO Michael Kagan.

NVIDIA applied sciences will likely be featured in a variety of server designs, together with NVIDIA CGX for cloud gaming, OVX for digital twins, and HGX Grace and HGX Grace Hopper for simulation, knowledge analytics and AI. 

NVIDIA introduced the primary wave of programs powered by the NVIDIA Grace CPU Superchip and Grace Hopper Superchip are anticipated within the first half of 2023. 

“Grace will likely be superb at AI, knowledge analytics, scientific computing, and hyperscale computing. And, in fact, the total suite of NVIDIA software program platforms will run on Grace,” mentioned NVIDIA senior VP of {hardware} engineering Brian Kelleher.

Grace-powered programs from ASUS, Foxconn Industrial Web, GIGABYTE, QCT, Supermicro, and Wiwynn will be a part of x86 and different Arm-based servers to supply clients a broad vary of decisions. 

“All of those servers are optimized for NVIDIA accelerated computing software program stacks, and could be certified as a part of our NVIDIA-Licensed Techniques lineup,” mentioned Ying Yin Shih, Director of Product Administration for Accelerated Computing at NVIDIA.

Liquid cooling

To supply enterprises with choices to deploy inexperienced knowledge facilities, NVIDIA additionally introduced its first knowledge middle PCIe GPU with direct chip liquid cooling. 

The liquid-cooled A100 PCIe GPUs will likely be supported in mainstream servers by no less than a dozen system builders, with the primary transport within the third quarter of this yr. 

“All of those mix to ship the infrastructure of the info middle of the long run that handles these large workloads,” Buck mentioned.

“Lastly, getting all of those to run seamlessly requires NVIDIA AI Enterprise software program, which delivers sturdy 24/7 AI deployment. Relating to re-imagining the info middle, NVIDIA has the whole, open platform of {hardware} and software program to construct the AI factories of the long run,” he added. 

Revolutionizing robotics with AI

NVIDIA VP of embedded and edge computing Deepu Talla spoke about how the worldwide drive to automation makes robotics a significant new software for AI. 

Greater than 30 main NVIDIA companions worldwide will likely be amongst these providing the primary wave of NVIDIA Jetson AGX Orin-powered manufacturing programs at Computex. New merchandise are coming from a dozen Taiwan-based digicam, sensor and {hardware} suppliers to be used in edge AI, AIoT, robotics, and embedded functions. 

“We’re getting into the age of robotics — autonomous machines which are keenly conscious of their atmosphere and that may make sensible choices about their actions,” Talla mentioned. 

Accessible since March, the NVIDIA Jetson AGX Orin developer equipment delivers 275 trillion operations per second, packing over 8x the processing energy of its predecessor, NVIDIA AGX Xavier, in the identical pin-compatible kind issue. 

Jetson Orin options the NVIDIA Ampere structure GPU, Arm Cortex-A78AE CPUs, next-generation deep studying and imaginative and prescient accelerators, high-speed interfaces, quicker reminiscence bandwidth, and multi-modal sensor help able to feeding a number of, concurrent AI software pipelines.

Providing server-class efficiency for edge AI, new Jetson AGX Orin manufacturing modules will likely be accessible in July, whereas Orin NX modules are coming in September. 

“Such modules are key to embedding smarter units on the planet round us,” mentioned Talla. 

The NVIDIA Isaac robotics platform has 4 pillars. 

The primary pillar is about creating the AI, “a really time-consuming and troublesome course of that we’re making quick and simple,” Talla mentioned, highlighting how instruments such because the Isaac Replicator for artificial knowledge era, NVIDIA pre-trained fashions accessible on NGC, and the NVIDIA TAO toolkit are addressing this problem. 

The second pillar is simulating the operation of the robotic within the digital world earlier than it’s deployed in the actual world with Isaac Sim. 

The third is constructing the bodily robots. 

And the fourth is about managing the fleet of robots over their lifetimes, usually a few years if no more than a decade. 

A part of that’s Isaac Nova Orin, a reference design for state-of-the-art compute and sensors for autonomous cellular robots (AMR) — filled with applied sciences corresponding to DeepMap, CuOpt and Metropolis. 

“That is the business’s most complete end-to-end robotics platform and we proceed to spend money on it,” mentioned Talla. 

Extra gaming and content material creation improvements

NVIDIA senior VP for GeForce Jeff Fisher, detailed how NVIDIA is working to ship innovation to avid gamers and content material creators. 

“Over the previous 20 years, NVIDIA and its companions have devoted themselves to constructing the most effective platform for gaming and creating. Lots of of hundreds of thousands now rely on it to play, work and be taught,” he mentioned. 

Launched in 2018, NVIDIA RTX has reinvented graphics, because of superior options corresponding to real-time ray tracing, and the momentum round it continues to develop. There are actually greater than 250 RTX-enabled video games and functions, doubling because the final Computex. 

NVIDIA DLSS continues to set the usual for super-resolution with best-in-class efficiency and picture high quality and is now built-in into greater than 180 video games and functions. 

DLSS is within the video games that avid gamers wish to play, with 12 new video games added to the ever-growing library.

Builders of the critically acclaimed Hitman 3 will likely be including NVIDIA DLSS together with ray-traced reflections and ray-traced shadows on Might 24.

As well as, NVIDIA Reflex is now supported in 38 video games, 22 shows and 45 mice. With greater than 20 million avid gamers enjoying with Reflex ON each month, Reflex has change into one in all NVIDIA’s most profitable applied sciences.

The Reflex ecosystem is continuous to develop: ASUS debuted the world’s first 500Hz G-SYNC show, the ASUS ROG Swift 500Hz gaming monitor. Acer additionally launched the Predator X28 G-SYNC show. In the meantime, Cooler Grasp launched the MM310 and MM730 gaming mice with Reflex.

“NVIDIA Studio, the RTX-powered platform that features dozens of SDKs and accelerates the highest inventive apps and instruments, and NVIDIA Omniverse, the corporate’s platform for constructing interconnected 3D digital worlds, are designed to allow collaboration and building of those digital worlds,” mentioned Fisher. 

Gaming laptops proceed to be the fastest-growing PC class, and 4th era Max-Q Applied sciences — the newest iteration of NVIDIA’s design for skinny and light-weight laptops — delivers a brand new energy effectivity stage. GeForce RTX laptop computer fashions now whole greater than 180.

These highly effective programs are getting used to assist construct large, interconnected 3D locations. 

“These are our most transportable, highest efficiency laptops ever,” Fisher mentioned.