Although NVIDIA did not release product information related to GeForce RTX 40 series graphics cards and personal computers at Computex 2022, it brought A100 PCIe computing cards using liquid cooling solutions, as well as a variety of Grace Hopper, AGX Orin Design reference platform and corresponding products.
Introducing liquid cooling for professional computing cards
Before the official launch of the NVIDIA H100 GPU, the A100 was still the main professional computing GPU. In order to meet the data center’s requirements for high performance, stability, and carbon neutrality, the A100 PCIe computing card with a liquid cooling solution was launched.
Equinix, an American telecommunications service provider with multiple data centers around the world, set up a department dedicated to improving energy efficiency in January. One of its tasks is to study liquid cooling (Liquid Cooling, Direct-Chip Cooling, which refers to the use of liquid substances or Water cooling system for chip cooling), this technology is increasingly mature and widely used in the world’s fastest supercomputers. NVIDIA said that the traditional air-cooled (Air-Cooled, using a fan for heat dissipation) GPU has brought 20 times the AI inference of the processor, high-performance computing power efficiency, and the introduction of liquid cooling is NVIDIA’s next focus for accelerated computing.
In independent test experiments, both Equinix and NVIDIA found that under the same workload, using liquid cooling can save about 30% of energy consumption compared to air cooling. Power saving, ideal PUE value of 1.0) reduced from 1.6 to 1.15. On the other hand, the air-cooled A100 GPU needs to occupy 2 sets of PCIe slots, while the water-cooled version only needs 1 set, which can accommodate twice the computing card in the same rack space, greatly improving the computing density.
A variety of design reference platforms debut
Recalling that NVIDIA has previously released Grace architecture processors and Hopper architecture GPUs, this time NVIDIA will launch a Grace CPU Superchip (consisting of 2 sets of Grace processors) or Grace Hopper Superchip (composed of Grace processors and Hopper GPUs each). 4 kinds of 2U form server design reference platform.
NVIDIA CGX for cloud gaming and graphics applications and NVIDIA OVX for digital twin applications are equipped with Grace CPU Superchip, NVIDIA graphics card and BlueField-3 data processor, while NVIDIA HGX for AI computing and high-performance computing is equipped with Grace CPU Superchip Or the Grace Hopper Superchip version with BlueField-3 data processor, custom I/O interface and 4th generation NVLink.
On the other hand, more than 30 global manufacturers launched productivity systems equipped with Jetson AGX Orin at Computex, with a variety of edge AI, AIoT, robotics and embedded applications, covering manufacturing, retail, transportation Transportation, smart cities, healthcare and other industries.
Unlike NVIDIA’s announcement of the GeForce RTX 3080Ti graphics card at COMPUTEX last year, NVIDIA has not brought any news about the home market product this year, and we will have to wait a while to hear about the GeForce RTX 40 series graphics card.