Those who buy this fall a laptop incorporating Intel’s Ice Lake chip will have access to something very special: the magic of AI, sparingly first, but more and more afterwards. This does not mean that AI capabilities are exclusive to the Ice Lake chip, or that without this chip, office software will not benefit from significant improvements. But to access some of the spectacular features that application developers are working on, the AI and machine learning capabilities Intel has built into its 10th generation Core chip will be needed.
Some of these features are already visible today. For example, the Microsoft Photos app uses smart image analysis to assess for itself what it “sees”, such as a photo of a beach or snow. Microsoft Photos and Google Photos already identify and group the subjects of the photos, based on a recognition of the people there. But on a PC, we often equate the AI to digital assistants like Cortana. And Intel would like to change this perception of artificial intelligence.
The benefits of integrated AI
Upstream Computex Taiwan (28 May to 1 st June), Intel showed that the AI could really bring something like a stylized video in real time during playback, as easily as applying a filter in Snapchat; or remove unwanted background noise during an audio conversation or chat; or, speed up the CyberLink PhotoDirector 10 feature to enhance the sharpness of photos. Microsoft Skype and Teams applications are already able to identify the caller from a video call and to blur or replace the background. The integrated AI will accelerate these processes.
The Gaussian Neural Accelerator, Intel’s Secret Boot
Intel’s secret boot is called the Gaussian Neural Accelerator or Gaussian Neural Accelerator, a piece of logical code installed in Ice Lake. These two elements work hand in hand: the architecture of the CPU accelerates the DLBoost, which in turn accelerates the technology of inference on the Intel Ice Lake CPU. (Inference applies rules or algorithms to known facts to learn more about them.) At the same time, the Gaussian neural accelerator, which needs very little energy to perform a specialized task, could, for example, translate a conversation in real time.
On a PC, the execution of a task is often subject to some competition between the multipurpose CPU and the dedicated expansion card. For example, on the first PCs, the multimedia functions were accelerated by a native signal processing or Native Signal Processing and the MMX (Multimedia Instruction Set) of Intel, before being entrusted to dedicated audio and graphics cards. Over time, and as the cost decreased, basic graphics and audio capabilities returned to the CPU and chip. In the same way, we are only in the early stages of AI treatment. For now, Intel holds its bet by spreading the workload between CPU DLBoost instructions, the Gaussian neural accelerator and the more classic integrated Iris Plus GPU.
