The processor manufacturer Intel has been drumming for some time for the next generation of processors due in 2023. Intel has now released the next bit of information into the world at Computex: Meteor Lake is bringing a function block called VPU, which is intended to relieve the processor of AI tasks.
The VPU is located in the central SoC chiplet and is therefore part of all Meteor Lake processors, regardless of the number of CPU or GPU cores. This can certainly be understood as a tip against AMD, because there is also an AI accelerator in the Ryzen 7040 series – but only at two higher expansion levels. Another difference: while AMD’s AI unit comes from the FPGA subsidiary Xilinx, Intel has not resorted to the same know-how from Altera, but from Movidius. This subsidiary was previously known mainly for automotive chips for autonomous driving.
Background noise removal and background blurring works much better than before with modern AI models, but it also requires a lot more computing power. However, the performance requirements do not increase as much as with other AI applications such as ChatGPT.
(Image: Intel)
Intel does not put its VPU in competition with the CPU or GPU, but sees one of the three as the best case, depending on the application. CPUs might be best at processing low-latency AI models that need to be singular, while the GPU can handle performance-demanding, high-throughput, parallelized models well. The VPU acts in between and is supposed to relieve the CPU cores of background tasks – for example filtering out background noise and blurring the background in video conferences. The VPU is conceptually similar to functional units such as QuickSync, which enables encoding and decoding of video streams without CPU load.
Similar to QuickSync, the VPU must be specifically addressed by the software so that it relieves the CPU. This includes the Studio Effects effects collection, which Microsoft integrates into Windows 11.
(mue)