Listen to this story
This week is a big one for chip enthusiasts and AI lovers alike. At Computex, the world’s premier semiconductor conference, top makers have announced chips that might be the future of AI compute. While NVIDIA tried its best to steal the show with CEO Jensen Huang’s announcement-packed keynote, Intel quietly made a splash by showcasing Meteor Lake, a new line of AI-first processors.
However, the chipmaker has been plagued with multiple delays and issues in the past when taking this manufacturing approach. With the hype surrounding Meteor Lake, it seems that Intel has a lot riding on this new chip.
What makes Meteor Lake special?
Meteor Lake marks a huge milestone for Intel, as it will be the first chip to use the Intel 4 process node for manufacturing. However, this isn’t all the chip has up its sleeve, as it also incorporates various chips manufactured in Taiwan Semiconductor Manufacturing Corporation’s (TSMC) foundries, making it the first Intel chip to do so.
Intel has gone all in on both the chiplet design and SoC architecture, integrating various discrete functions on one chip in a bid to create a new kind of market. The Meteor Lake processors combine GPU, VPU, and CPU all on one chip to offer the “right balance of power and performance for AI”.
According to Intel’s announcement at Computex, the chips will first appear in laptops, seemingly in competition to Apple’s M2 and AMD’s Ryzen 7040 series of chips. Similar to its competitors, Intel has also adopted a system-on-chip architecture for this chip, mixing and matching various chiplets to achieve this effect.
The CPU itself will be manufactured by using Intel 4 manufacturing process, with the GPU and SoC Tile using TSMCs 5nm and 6nm processes, respectively. Moreover, Intel has also integrated its Foveros 3D packing technology into these chips, allowing them to fit more features in a smaller form factor.
The headline feature of this new chip is its integrated AI engine, facilitated with the help of Intel’s new vision processing unit (VPU). The VPU will not only be reserved for AI workloads, but will also accelerate the integrated GPU and the main CPU itself for these tasks. Notably, the VPU contains Intel’s Gaussian Neural Accelerator (GNA) chip, which has been, so far, its solution for on-board AI processing.
While the GNA has been Intel’s go-to for low-power AI inference at the edge, it seems that VPUs will replace them as the generations go on. Intel has stated that some of the code made for GNA runs more efficiently and with better results on VPUs, effectively obsoleting this now-legacy tech.
The chip is also scalable to various power draw and performance specifications. Each of the components can be scaled up or down depending on the requirement, resulting in a varied line of products with multiple different SKUs. The CPU, GPU, and VPU are connected with an interconnect fabric that allows them to share the same memory subsystem. This not only improves performance, but puts Intel on par with Apple and AMD for SoC design.
Harnessing the power of developers
While it seems that Intel is a few years late to the party in this specific market, this delay is to be expected. The company’s Sapphire Rapids chip, which was supposed to be its big answer to AMD’s EPYC server dominance, was hit with multiple delays and manufacturing issues. This resulted in Intel’s roadmap being pushed ahead by a few years, creating a lag that it is now eager to close.
Meteor Lake is likely to launch with full support for developers to tinker with the product and harness its full capabilities. Apart from close collaboration with Microsoft to integrate the chips’ AI capabilities into Windows, Intel has also announced support for DirectML, ONNX, and OpenVINO. These SDKs will allow AI developers to easily integrate ML features into their programs while ensuring compatibility.
Intel showcased a host of optimisations that Meteor Lake’s in-built AI could bring to applications. Firstly, the VPU can accelerate even common video and audio processing workloads running on the CPU and GPU. Secondly, the processor’s capabilities will have support in various software, like real-time motion capture in Unreal Engine.
For AI models specifically, the hybrid approach of combining CPUs and GPUs together with VPUs works wonders. For example, Stable Diffusion can run on both GPU and VPU simultaneously. While the GPU does the heavy lifting of generating the image, the VPU works side by side for super-resolution of the image.
A much-needed win?
Intel has been suffering in silence, unable to capitalise on the AI wave due to its varied problems. However, Meteor Lake could be the solution for them to get their winning streak back, biting back at Apple for ditching Intel chips for the MacBook. The only thing left for Meteor Lake is Intel getting it to market.
While it has unfortunately stumbled when getting the product to market in the past, Meteor Lake might be the exception to this. Intel announced that it would bring Meteor Lake to laptops by the end of this year. This is also backed up by the fact that Intel announced ramped up production for the chips in its Q1 2023 earnings report.
By bringing together TSMC’s manufacturing chops and their own foundries, Intel might actually go beyond the manufacturing problems it had in the past. While combining this with the complete package that Meteor Lake seems to be, it might just be the win Intel needs to get back on its feet.