Intel NPU Linux Driver Gets Power Management Controls for Core Ultra SoCs

Intel has introduced a new driver patch for the Linux kernel that enables frequency limiting on the neural processing unit (NPU) found in Core Ultra SoCs. The patch, submitted to the dri-devel mailing list, allows users to read and set minimum and maximum NPU clock speeds on Intel NPU 50XX+ hardware, including Panther Lake and newer SoCs.

This development comes as a surprise to some, given that NPUs have been marketed as power-efficient components for AI workloads. The ability to limit NPU frequency suggests that thermal and power management considerations are more significant than initially anticipated for these accelerators.

Technical Details

The patch implements several key features for the IVPU driver:

• Reading minimum and maximum NPU clock speeds
• Obtaining the optimal operating frequency
• Reading the current NPU clock frequency
• Setting minimum and maximum NPU frequencies (50XX+ hardware only)

The frequency control capabilities are limited to Intel NPU 50XX+ hardware, which includes Panther Lake and newer generations. While older Intel NPUs can still report their current clock frequencies, they lack the ability to have their frequencies adjusted by users.

Hardware Support

Intel's NPU architecture has evolved through several generations:

• NPU1-NPU4: Found in earlier Core Ultra processors
• NPU5: Introduced with Panther Lake (50XX+)
• NPU6: Coming with Nova Lake

This driver update specifically targets NPU5 and later, meaning it will be particularly relevant for users with newer hardware looking to optimize their system's power consumption and thermal performance.

Firmware Integration

The NPU frequency controls work in conjunction with updated Intel NPU firmware, ensuring proper handling of frequency changes. This integration is crucial for maintaining system stability while allowing users to fine-tune performance based on their specific needs.

Performance Implications

The ability to limit NPU frequency could have several practical applications:

1. Power Savings: Reducing NPU frequency during light workloads can lower overall system power consumption
2. Thermal Management: Limiting frequency can help keep temperatures in check during extended AI workloads
3. Battery Life: On laptops, frequency limiting could extend battery life when AI acceleration isn't critical
4. Noise Reduction: Lower temperatures may reduce the need for fan activity, leading to quieter operation

Linux Kernel Integration

The patch is currently under review for inclusion in the mainline Linux kernel. Once accepted, it will provide Linux users with granular control over their NPU's performance characteristics, similar to existing CPU frequency scaling capabilities.

This development represents Intel's continued commitment to improving Linux support for their AI acceleration hardware. As NPUs become more prevalent in consumer and professional computing, having robust power management controls will be increasingly important for optimizing system performance and efficiency.

The patch submission can be found on the dri-devel mailing list, where it's undergoing community review and discussion before potential integration into the mainline kernel.