The Intel IVPU accelerator driver used on Linux for the neural processing unit (NPU) on Core Ultra SoCs saw a patch posted for allowing the NPU clock frequency to be limited in the name of power and thermal management.
Somewhat surprisingly with how NPUs have been talked up as being a tiny part of the die and power efficient for AI, a patch under review for the Linux kernel’s IVPU driver will allow limiting the NPU frequency for power and thermal management reasons.
The Intel-contributed patch for the IVPU driver will allow the minimum and maximum NPU clock speeds to be read, obtaining the optimal operating frequency, reading the current clock frequency, and with Intel NPU 50XX+ hardware to support minimum and maximum NPU frequencies. This driver patch goes in conjunction with updated Intel NPU firmware for properly handling the changes.
With the ability to set the clock frequency of the NPU limited to 50XX+ hardware, this means Intel Panther Lake and newer SoCs that introduced NPU5. It’s with next-gen Nova Lake where Intel’s 6th Gen NPU is being introduced. Prior gen Intel NPUs at least support still reading the current NPU clock frequencies, just not the ability for the user to set them.
The patch implementing the NPU frequency controls for the Linux kernel driver is out for review on the dri-devel mailing list.
