Home » Technology » Nvidia Computex Keynote: Grace will replace AMD and Intel CPUs from early 2023

Nvidia Computex Keynote: Grace will replace AMD and Intel CPUs from early 2023

Big new products are not just a matter for the boss at Nvidia. So it’s not surprising that without a spokesperson, CEO Jensen Huang didn’t make the big announcement about future products in Nvidia’s keynote at Computex. The update to the Grace (Hopper) Superchip still has it all. And there was also water cooling for server GPUs.

Two announcements for players

There were only two minor announcements for players:

  • In partnership with Nvidia, Asus will launch a 500 Hertz gaming monitor – more are likely to follow. No appointment has been made yet.
  • Hitman 3 (test) gets hardware ray tracing and Nvidia DLSS (as well as AMD FSR) via patch – developer IO Interactive had already announced this a week ago with patch 3.110 for this week.

There was no mention of a GeForce GTX 1630, and not secretly either Nvidia.com published.

Servers with Grace (Hopper) Superchip from early 2023

At first glance, there was a small, but at second glance far-reaching update from Nvidia for the first products with its own ARM CPU “Grace”. The products Grace Superchip (CPU + CPU) and Grace Hopper Superchip (CPU + GPU) presented so far will be offered by many Taiwanese partner companies in servers in early 2023. They will be based on reference systems developed by Nvidia, which will be available in various form factors and expansion phases.

All platforms already exist but are based on AMD or Intel x86 processors. For example, for cloud gaming systems, Nvidia will offer the CGX platform with Grace Superchip. In addition to a Grace Superchip with two Grace CPUs, this server features Ampere-based ones A16 Graphics Cards, as currently used by GeForce Now with a Ryzen Threadripper x86 CPU. From early 2023, the OVX and HGX platforms will be available with Nvidia’s own ARM-based CPU in addition to the well-known Intel and AMD variants. Nvidia has not yet promised a Grace model for DGX.

Nvidia’s vision to no longer have to buy processors from the market with its own ARM CPUs is already taking shape at the beginning of 2023.

Water-cooled A100 and H100 graphics card

In addition, Nvidia announced that it, along with Equinix, a company that operates 240 data centers worldwide, is currently validating a factory water-cooled graphics card with A100 GPU, the Ampere-generation large HPC chip. The data center is responsible for a large part of the power consumption and therefore also the efficiency.

Compared to a comparable data center with air-cooled A100 graphics cards, Nvidia promises an energy saving of 30 percent with the water-cooled variants. Another advantage: the graphics cards, which are only one slot wide, take up considerably less space. Delivery of the new A100 variant with water cooling is planned for the third quarter of 2022. The successor H100 with hopper architecture, which has already been presented, should also come on the market in early 2023 as a version with a full-cover water cooler. So far, a “passive” SXM5 module and an air-cooled PCIe graphics card have been announced.

Jetson AGX Orin gets even faster

Finally, Nvidia also announced the first series of products based on Jetson AGX Orin. To date, developers have only been able to purchase a development kit. The heart of the Jetson AGX Orin is the Orin SoC of the same name, which is manufactured by Samsung in 8LPP and has 17 billion transistors. The SoC combines up to 12 ARM Cortex-A78AE (Hercules) CPU cores with an integrated Ampere GPU with 2,048 CUDA cores.

Nvidia also promised 32 GB of LPDDR5 RAM, which is connected with a memory bandwidth of 204 GB/s, and an AI computing power of 200 TOPS (INT8) – six times the performance of the previous flagship Jetson AGX Xavier. With the AGX Orin 64 GB, Nvidia now wants to supply 275 TOPS and 64 GB of RAM from October 2022. However, the first variant to be presented in July will start. The Jetson developer platform is designed for robotics, autonomous machines, medical devices, and other types of embedded computing at the edge. ComputerBase has received information about this item from Nvidia under NDA. The only requirement under the NDA was the earliest possible release date.