Intel's NPU Driver Officially Supports Ubuntu 26.04 LTS, Five Months After It Shipped

Intel's NPU Driver Officially Supports Ubuntu 26.04 LTS, Five Months After It Shipped

Intel Linux NPU Driver 1.38 has exactly one change in its release notes: official support for Ubuntu 26.04 LTS.

Ubuntu 26.04 shipped in April. This is September.

What the driver is

The NPU driver package is the userspace half of Intel’s Neural Processing Unit support, the accelerator built into Core Ultra parts for running inference workloads without engaging the CPU or GPU.

The kernel half is ivpu, already upstream, which is why the stack has worked on 26.04 for months regardless of what Intel formally claimed. The split matters: the kernel driver arrives with your distribution, while the userspace components come from Intel’s own packaging on a separate schedule.

What “official support” bought

Validation rather than code. Intel now states it has tested the userspace driver on Ubuntu 26.04 LTS with its default Linux 7.0 kernel across:

  • Meteor Lake
  • Arrow Lake
  • Lunar Lake
  • Panther Lake
  • Wildcat Lake

Ubuntu 24.04 LTS remains supported as well.

For a hobbyist that distinction is academic, because it already worked. For anyone deploying NPU workloads in a business, the difference between “works” and “vendor-validated” is the difference between a supportable configuration and an argument with a support desk.

The timing is the story

Five months is a long gap between an LTS shipping and a major vendor’s accelerator stack formally supporting it, and it arrives weeks before Ubuntu 26.10.

There is a defensible reading. Validating five CPU generations against a new distribution is real work, LTS releases are what enterprises actually deploy, and being thorough beats being early.

There is a less comfortable one, given that Intel has spent two years trimming open-source commitments and has just suspended its bug bounty programme. Validation effort is headcount, and headcount is what has been under pressure.

Both can be true. The observable fact is that Intel’s accelerator support trails the distributions it targets by months, which is a planning input if you are buying Core Ultra hardware for Linux inference work.

Checking your own

# is the kernel driver present
lsmod | grep intel_vpu
ls /dev/accel/

# what the kernel sees
dmesg | grep -i ivpu

If /dev/accel/accel0 exists, the kernel side is working. The userspace package from Intel’s GitHub releases is what sits above it and connects OpenVINO to the hardware.

For self-hosted inference more generally, our AI and machine learning app coverage leans heavily on GPU acceleration, which remains the better-supported path on Linux today. NPUs are lower power and considerably narrower in what they accelerate well.

Background reading

Explainers for the concepts behind this story.