your123 -

Intel, Nvidia aim latest systems-on-a-chip at AI workstations

Nvidia's RTX 500 and RTX 1000 GPUs are for the lightest mobile workstations. Intel's Core Ultra with a built-in GPU can handle many AI-powered tasks on the workhorse computers.

Intel and Nvidia have introduced systems on a chip aimed at corporate workstations where AI will become the norm.

At this week's Mobile World Congress (MWC) 2024 conference in Barcelona, Spain, Nvidia launched two lightweight GPUs for thin mobile workstations. Intel also introduced a Core Ultra processor with a built-in GPU.

Chipmakers and PC manufacturers are upgrading product lines to address the needs of companies preparing to perform AI-driven tasks within business software. Analyst firm IDC expects PCs to perform those tasks locally with a specially designed system-on-a-chip (SoC).

"With many new workloads moving to the workstation that includes AI -- modeling systems, graphics creation, video editing, oil and gas exploration, etc. -- it makes sense to have a GPU subsystem that can be used for AI acceleration," said Jack Gold, principal analyst at J. Gold Associates.

Shipments of so-called AI PCs will account for 60% of all PC shipments globally by 2027, or more than 167 million units, according to IDC. The analyst firm expects AI PC shipments to reach nearly 50 million units this year.

Nvidia lightweight RTX GPUs

Nvidia built its new RTX 500 and RTX 1000 GPUs for AI acceleration within the lightest mobile workstations. The company also sells more powerful RTX GPUs for faster and more expensive midrange and high-end workstations.

It makes sense to have a GPU subsystem that can be used for AI acceleration.
Jack GoldPrincipal analyst, J. Gold Associates

Nvidia typically launches GPUs for higher-end workstations before addressing the less powerful models, IDC analyst Shane Rau said. "The introduction of these low-end products for laptop workstations will likely see their features move into laptop PC discrete GPUs next."

Nvidia builds its RTX GPUs using the Ada Lovelace architecture. An Ada SoC includes a neural processing unit (NPU) as a CPU component and an Nvidia RTX GPU with Tensor Cores for AI processing. The NPU handles lightweight AI, while the GPU handles tasks requiring more processing power.

The RTX 500 and 1000 GPUs add up to 154 trillion operations per second and 193 TOPS, respectively, for AI.

The GPUs are available in Dell Technologies' Precision 3280 workstation, introduced this week and scheduled for availability March 26. Other computer makers with upcoming workstations using the Nvidia GPUs include HP, Lenovo and Micro-Star International.

Latest Intel Core Ultra

With the company's Arc GPU built-in, the new Intel Core Ultra provides graphics performance and features for AI processing and ray tracing, a rendering technique used in 3D computer graphics.

The SoC is available with an Arc Pro workstation graphics driver that computer makers can use to boost performance across creative, design and engineering software.

The features of Intel Core Ultra.
Intel's latest Core Ultra system-on-a-chip includes a built-in Arc GPU for AI.

IT departments can manage and secure workstations with Intel's Core Ultra processors and Arc GPUs using the company's vPro technology. The platform, introduced in 2007, lets IT remotely manage business PCs and workstations.

Dell plans to offer the Intel Core Ultra with the built-in Arc GPU in its Precision 3000 and 5000 Series mobile workstations.

Also at MWC, Intel introduced an AI Development Kit for telecom operators that want to build, train, optimize and deploy AI models within their virtual radio access networks. Virtual random access networks (vRANs) let carriers separate network functions from their underlying hardware.

The kit, available to only select partners today, includes libraries, frameworks and software tools to run models on Intel's fourth-generation Xeon processors with built-in AI accelerators.

Companies working with Intel on developing the kit include AT&T, Deutsche Telekom, SK Telecom and Vodafone.

In 2025, Intel plans to release a Xeon processor codenamed Granite Rapids-D that the company said would boost the performance of operators' vRANs.

Intel is working with Dell, Ericsson, HPE, Lenovo and Samsung to get the chip ready for release.

Antone Gonsalves is an editor at large for TechTarget Editorial, reporting on industry trends critical to enterprise tech buyers. He has worked in tech journalism for 25 years and is based in San Francisco.

Next Steps

Windows XP EOL anniversary: 10 years of endpoint evolution

AMD, Intel, Nvidia and Qualcomm update chips for AI at Computex

Dig Deeper on AI infrastructure

Business Analytics
Data Management