Nvidia’s 6 Biggest GTC 2021 Product Announcements

The chipmaker is making major expansions beyond its traditional GPU business with the reveal of a new data center CPU, a next-generation data processing unit and new DGX systems for AI computing at the company’s GTC 2021 event.

Far More Than Just A GPU Maker

Nvidia wants to cement itself as the premier company for accelerated computing, and it’s doing that by expanding beyond its original GPU vision to sell other kinds of chips as well as the systems they power.

The Santa Clara, Calif.-based company revealed its plans at the virtual GTC 2021 event to become a three-chip company with a new Arm-based data center CPU that is due out in early 2023.

[Related: 10 Hottest Semiconductor Companies To Watch In 2021]

It’s a big move for a company that has largely sold GPUs for most of its 28-year history, but Nvidia made its intentions of becoming a heterogenous compute technology provider known last year with its $7 billion acquisition of Mellanox Technologies, which resulted in Nvidia’s line of BlueField data processing units.

And yet, Nvidia clearly isn’t content to just be a components provider. The company has continued to expand its line of DGX appliances and clusters for AI computing, and it’s making new software investments in the enterprise world to become a next-generation data center vendor.

What follows are the six biggest product announcements Nvidia made at GTC 2021, from its new data center GPU and BlueField-3 DPU to new DGX products and workstation GPUs.

Nvidia Challenges Intel, AMD With Grace CPU For AI Data Centers

Nvidia revealed an Arm-based data center CPU for AI and high-performance computing it says will provide 10 times faster AI performance than AMD’s 64-core EPYC Rome processors. The new data center CPU is named Grace after computer programming pioneer Grace Hopper, and it’s set for launch in early 2023. The CPU can deliver 900 GB per second of bi-directional bandwidth for CPU-to-GPU communication thanks to a next-generation Nvidia NVLink interconnect that connects the Arm-based CPU and an Ampere GPU on Grace’s system-on-chip. Grace will use next-generation Neoverse CPU cores from Arm as well as an LPDDR5x memory subsystem, which the company said will provide double the bandwidth and 10 times better energy efficiency. Nvidia CEO Jensen Huang said the company will now have road maps for GPUs, DPUs and CPUs, with architecture updates happening every two years, with one year putting emphasis on x86 platforms and the other on Arm platforms.

Nvidia’s BlueField-3 DPU Will Offload Many Data Center Workloads

Nvidia revealed more details for its next-generation data processing unit, the BlueField-3, which the company said will be capable of offloading up to 300 CPU cores’ worth of networking, storage and security workloads in data centers. That’s 10 times greater offload capability than the company’s latest DPU in the market, BlueField-2. Nvidia said BlueField-3 will transform data center infrastructure into “zero-trust” environments by providing new real-time network visibility, detection and response capabilities as well as a monitoring agent, called Nvidia Morpheus. The company called BlueField-3 the industry’s first DPU with a 400 GbE capability and said it will support PCIe 5.0 connectivity when it launches in the first quarter of 2022. Dell Technologies, Inspur, Lenovo and Supermicro are now integrating BlueField DPUs into their systems, and the component also has the support of security vendors like Fortinet, storage vendors like NetApp and edge platform vendors like Cloudflare.

New Nvidia Workstation GPUs For Desktops, Laptops, Servers

Nvidia announced new workstation GPUs for desktops, laptops and servers. On the desktop side, the company announced the Nvidia RTX A5000 and A4000, both of which feature new RT Cores, Tensor Cores and CUDA cores to accelerate AI, graphics and real-time rendering performance by up to two times compared to previous generations. The new laptop GPUs consist of the Nvidia RTX A2000, A3000, A4000 and A5000, which include the latest versions of Max-Q and RTX technologies to deliver high performance in thin-and-light devices. For data centers, Nvidia is releasing the Nvidia A10 (pictured) and A16 GPUs, with the former offering up to 2.5 times the performance of previous generations for virtual workstations and the latter providing up to double the user density and lower total cost of ownership for virtual desktop infrastructure. The new desktop and data center GPUs will be available through distributors and OEMs in late April while the new laptop GPUs will be available in mobile workstations from global OEMs in the second quarter of this year.

Nvidia Makes Big Enterprise Push With Certified Servers, Software

Nvidia is making a big enterprise push this year with servers from OEMs and ODMs that are certified to run the company’s new Nvidia AI Enterprise software suite, which has been exclusively certified to run on VMware’s vSphere 7 virtualization platform. Among the server vendors with Nvidia-certified systems are Atos, Dell Technologies, Gigabyte, Lenovo and Supermicro - all launching new servers supporting Nvidia’s EGX platform, which allows enterprises “to run AI workloads on the same infrastructure used for traditional business applications,” according to Nvidia. More than 20 Nvidia-certified systems are available now, and systems including Nvidia’s new A30 and A10 GPUs will be available later this year from server vendors.

Nvidia Launches New Nvidia DGX SuperPOD, DGX Station Subscription Service

Nvidia announced a new version of its DGX SuperPOD reference architecture for AI supercomputers that introduces BlueField-2 DPUs to improve security and optimize workload performance with new CPU offload capabilities. The company is also expanding DGX SuperPOD capabilities with Nvidia Base Command, a new solution for allowing multiple users to securely access, share and operate DGX servers, making it easier to provision and schedule workloads. Nvidia also announced that it is now providing its Nvidia DGX Station A100 through a subscription program, with prices starting at $9,000 per month. The new DGX SuperPODs and Base Command will be available in the second quarter of this year while subscription-based DGX Station A100 systems are available now.

Nvidia Partners With Google Cloud, Others For AI-On-5G Platform

Nvidia said it is partnering with Google Cloud, Fujitsu, Wind River and other companies to build solutions on top of its AI-on-5G platform, which is designed for cities, factories, hospitals and retail stores, among other kinds of environments that can benefit from high-performance 5G RAN and AI applications. The platform brings together 5G and edge AI capabilities through its use of the Nvidia BlueField-2 A100, which integrates GPUs and DPUs, and Nvidia’s Aerial software development kit. Use cases that can benefit from the platform include automated guided vehicles, drones, wireless cameras, manufacturing robots and retail self-checkout systems. Google Cloud will support Nvidia’s AI-on-5G platform through its Anthos hybrid cloud platform, which supports GPU-accelerated servers.