Intel has released information about a new type processor will soon make its way into the company’s portfolio: the versatile processing unit, or VPU.
While no formal announcement has been made, written material has been published by Intel alongside a new Linux driver confirmed the existence of the processor, which is designed to accelerate AI inference workloads.
According to the documentation, the VPU will be incorporated into Intel’s 14th Gen Core CPUs (aka Meteor Lake) and improve inference performance in “computer vision and deep learning applications”. It is most likely the brainchild of the team behind Movidius, an AI acceleration company acquired by Intel in 2016.
GadgetMasti Pro asked Intel for more details, but didn’t get an immediate response.
With rivals like Nvidia fighting to establish itself as the Intel, the leading chipmaker of the AI era, will think hard about how to demonstrate its own credentials in space.
In general, there are two types of AI workload: training and inference. The former refers to using large-scale datasets to develop AI applications with specific capabilities, while the latter refers to entering new data into these systems to generate a result.
Due to the amount of data involved, training is both very computationally intensive and very time consuming. In contrast, inference must be performed almost immediately, which means that the computational requirements differ significantly.
While details are few, Intel’s new VPU appears to be designed to speed up interference workloads specifically on client devices. It is unclear whether the line of the company server CPUs will benefit from similar technology in the future.
With Meteor Lake expected to land in late 2023, the new VPU won’t see the light of day anytime soon, but the chip’s existence offers some additional insight into the broader strategy at Intel.
With some of the world’s largest companies (including Apple) discovering that custom silicon designed specifically for internal configurations can offer significant performance gains over general-purpose chips from Intel or AMD, chipmakers must find a way to compete. stay .
One option would be to expand the portfolio with specialized SKUs with additional accelerators and co-processors designed to improve the performance of the target workloads.
Via Phoronix (opens in new tab)Tom’s Hardware (opens in new tab)