Via Anandtech – SFF Network’s News Post is a quick summary, please check the linked article for a deeper dive.
Intel’s upcoming “Meteor Lake” (14th Gen Core) architecture will see the debut of true AI processing hardware on a consumer desktop CPU. With the current AI “revolution” in full swing, it was only a matter of time until we saw related technologies integrated into the systems we use every day. I say true, as Intel has had their GNA, or Gaussian & Neural Accelerator, an audio processing core included in CPUs since “Ice Lake”.
Whilst the CPUs are little too far away to actually be shown at Computex (beginning tomorrow as of posting this article), Intel will be effectively “soft launching” the tech at the show, showing off their tech and their plans for the platform, including what they expect software and OS developers to use the technology for. While integrating AI core(s) into a consumer processor may be new, Intel has been working within the AI field for some time, with a group of subsidiaries working in the market segment.
Intel clearly expects AI to integrate into everything in the coming years, hence the move to integrate the AI node into Meteor Lake. Using a Movidius (one 0f the aforementioned subsidiaries) designed core included in the SoC (Meteor Lake will be a few dies integrated into a single package, as is the norm now). The Movidius die is their third generation “Vision Processing Unit” (VPU).
While the true launch of Meteor Lake isn’t that far away, Intel hasn’t released any performance numbers for the VPU as yet – although as this is a new hardware segment for consumers, any numbers will be difficult to compare until we see further products in the market. Of note though, all Meteor Lake processors will include the VPU, although there is no mention whether or not the core will be scaled differently per product – will the i3 get the same VPU as the i9?
Intel is aiming the VPU as a third option for AI processing – alongside the CPU and GPU cores – the GPU for high intensity workloads, and the CPU for small, quick tasks. The VPU will sit in the middle – for sustained AI workloads without tapping into the comparatively power hungry GPU cores.
Being that the Movidius core is a “Vision Processing Unit”, it’s likely that the core may be more focused on visual/graphical workloads, and less so on others – so more Dall-E and less ChatGPT.
What’s missing though, is the software to really take advantage of these cores. Not many pieces of software at present use integrated AI code, but Intel wants to support devs in pivoting to including AI in their work by including the VPU core in the upcoming processors – as Anandtech puts it – “Build it and they will come”.
