Upcoming Intel ‘Meteor Lake’ CPUs Will Feature ‘VPUs’ for AI Acceleration

Intel has begun teasing details of its upcoming ‘Meteor Lake’ CPU architecture, which should be released as part of the 14th Gen Intel Core family starting later this year. On the sidelines of the Computex trade show in Taipei, Intel demonstrated for the first time a prototype laptop running an unidentified Meteor Lake engineering sample CPU, showing off its AI inference processing capabilities. In addition to CPU and GPU cores, Meteor Lake introduces a new ‘VPU’ (vision processing unit) for AI acceleration. The VPU hardware is based on Intel’s Movidius computer vision tech, which it acquired in 2016 and then offered as a standalone USB device.  

Meteor Lake is a major release for Intel, continuing its heterogenous core strategy but also using a chiplet-based layout and leveraging the company’s Foveros manufacturing process to stack multiple layers. It will also mix and match chiplets fabbed by Intel itself on the new Intel 4 manufacturing process as well as ones outsourced to TSMC, for the first time. As confirmed to Gadgets 360 by former Intel graphics chief Raja Koduri a short while ago, Meteor Lake CPUs will also feature integrated Intel Arc GPUs based on the first-gen ‘Alchemist’ Intel Xe architecture.

Now, company representatives tout the increasing importance of AI processing on-device, rather than through the use of cloud services. ‘Hundreds’ of apps and experiences are ready to leverage AI, including commonly used productivity and creativity tools from Adobe, Microsoft and Google. According to Intel, users can expect to experience things such as far better background separation on video calls, automatic voice isolation and enhancement, superior language understanding and synthesis for interactions, real-time motion capture, and radically reimagined user interfaces that adapt to what the user is doing, all in the near future.

Gadgets 360 was present at a private briefing where Intel showed its prototype laptop running the Stable Diffusion generative AI model, which allows anyone to create completely original images based solely on text prompts for things such as subject matter, theme, and artistic style. Intel has developed a simple plug-in for Stable Diffusion in the open-source GIMP image editor. Targeting different workloads at the chip’s CPU, GPU and VPU components, the demonstration ran in roughly 20 seconds. The image was generated entirely on-device, with no Internet connectivity needed.

As a major selling point for the future 14th Gen Intel Core CPU family, local offline AI processing can help increase power efficiency, preserve privacy, and reduce latency. With AI models such as ChatGPT becoming more and more complex with each iteration, dedicated hardware should become relevant.

The company is not yet ready to disclose CPU specifications for 14th Gen Core models, so it is still unclear whether Meteor Lake will serve all segments, and what the launch dates, speeds, and balance of CPU, GPU and VPU hardware of actual shipping CPUs will be. However, the company has indicated that power management is a priority for this generation, and has also confirmed leaks that a major shakeup to the product naming scheme is imminent. Recent leaks have suggested that the ‘i’ in the Core i3, i5, i7 and i9 names is going away, and at least one instance of a purported “Core Ultra 7” CPU has appeared in benchmark databases online, as product development has likely begun.