Nvidia tips new water-cooled GPUs, Jetson AI devices at Computex

At the start of Computex 2022, Nvidia showed off a handful of upcoming products and industry blueprints that aim to bring more computing power to different corners of the market. These include a design for a new “superchip” system based on the Grace and Hopper architecture (for large-scale modeling and training tasks) and a new “Jetson Orin” system for small GPU and AI-intensive workloads targeting data centers. There is.

Reference design and water-cooled GPU

High-performance computing hardware is not a panacea. Some types of high-end hardware perform better than others when it comes to other specialized tasks. Nvidia has made great strides in this department in recent years. The company now aims to make it easy for data center customers to use the right system. To achieve this, we have released four new reference designs for data center systems.

Four vary in performance and hardware, and are built for easy customization. All based on iteration. Grace and Hopper Architecture(Opens in a new window) And a huge computing processor based on it. These designs are built for large-scale AI and scientific computing tasks. Among the new reference designs, various Grace chip-based blueprints target cloud graphics and Nvidia Omniverse applications (the latter for 3D design simulation and collaboration), while Grace Hopper-based blueprints target AI, learning and inference.

Nvidia also announced that it was their first data center graphics card. The Nvidia A100 is a PCI Express GPU designed to fit inside a single PCI Express slot, unlike the two- and three-slot designs of most cards targeting the consumer mainstream and high-end markets. The difference: The “Ampere” based A100 uses liquid cooling to provide thermally efficient operation in a smaller package than traditional air-cooled graphics cards. In server environments where space is limited, this can allow for higher compute densities. The water-cooled H100 will be available in 2023.

Nvidia A100 PCI-E

Nvidia’s card partners will source cards based on Nvidia’s A100 reference design later this year for these GPUs. Asus, Asrock, Gigabyte, and others will be included as proposing parties.

The A100 card is intended for customers who already have upgradeable servers or cannot justify purchasing a new data center system. Details about the A100 are currently sparse, but are expected to launch later this year. The big news is simply that Nvidia entered this market with a water cooling solution.

Meet more Jetsons

Nvidia is also expanding its Jetson product line with four new devices. Prices range from $399 to $1,599. (See original Jetson and Jetson Xavier NX labs) Jetson products are designed to power AI-intensive workloads, and these new products will help you extend your line to more markets while delivering performance improvements. Better than previous-generation Jetson products

Editor’s Pick

Nvidia Jetson Orin device

The cheapest of these devices is the Jetson Orin NX 8GB. Equipped with 6 ARM Cortex-A78 CPUs, it can process up to 70 trillion operations per second (TOPS). The more powerful Jetson AGX Orin 64GB offers even more features with 12 ARM Cortex-A78 CPU cores and a maximum bandwidth of 275 TOPS.

All of these new high-performance components are expected to launch at various points in the coming months. Stay tuned for more details as Nvidia releases.

Get the best story ever!

join what’s new We deliver top news to your inbox every morning.

This newsletter may contain advertisements, deals, or affiliate links. Subscribe to our newsletter Terms of service And Privacy Policy. You may unsubscribe from the newsletter at any time.

var facebookPixelLoaded = false;
window.addEventListener(‘load’, function(){
document.addEventListener(‘scroll’, facebookPixelScript);
document.addEventListener(‘mousemove’, facebookPixelScript);

function facebookPixelScript() {
if (!facebookPixelLoaded) {
facebookPixelLoaded = true;
document.removeEventListener(‘scroll’, facebookPixelScript);
document.removeEventListener(‘mousemove’, facebookPixelScript);


fbq(‘init’, ‘454758778052139’);
fbq(‘track’, “PageView”);

Leave a Comment