──[User─space]──────────────────────────────────────────────────────────────────────────────
┌──────────────────────────(optional)─┐ ┌────────────────────────────────────────┐
│ User Mode Driver tests │ │ OpenVINO │
│ │ │ │
│ intel-validation-npu │ │ openvino_toolkit │
│ (vpu-umd-test) │ │ │
│ │ │ │
└──────────────────▲──────────────────┘ └────────────────────▲───────────────────┘
╚════════════════════╦══════════════════════╝
............................................║..............................................
┌───────────────────────────────────────▼──────────────────────────────────────────┐
│ oneAPI Level-Zero level-zero │
│ (libze_loader.so) │
│ │
└───────────────────────────────────────▲──────────────────────────────────────────┘
╔════════════════════╝
┌──────────────────▼──────────────────┐ ┌────────────────────────────────────────┐
│ User Mode Driver │ │ NPU compiler │
│ │ │ │
│ intel-level-zero-npu ◀═══▶ intel-driver-compiler-npu │
│ (libze_intel_vpu.so) │ │ (libvpux_driver_compiler.so) │
│ │ │ │
└──────────────────▲──────────────────┘ └────────────────────────────────────────┘
╚════════════════════╗
──[Kernel─space]────────────────────────────║───────────────────────────────────────────────
┌───────────────────▼─────────────────────┐
│ Kernel Mode Driver │
│ intel-npu-driver │
│ (intel_vpu.ko) │
│ │
└───────────────────▲─────────────────────┘
──[Device]──────────────────────────────────║──────────────────────────────────────────────
┌───────────────────▼─────────────────────┐
│ Firmware │
│ intel-fw-npu │
│ │
└─────────────────────────────────────────┘
Model driver caching (from v1.5.0)
Starting from v1.5.0 release the driver allows to cache compiled model in
filesystem. Whenever user compiles a model (using zeGraphCreate), the driver
saves the compiled model representation in the user directory
~/.cache/ze_intel_npu_cache/ or in the directory specified in
ZE_INTEL_NPU_CACHE_DIR environment variable. The cache directory size is
controlled by the driver. Whenever the cache directory exceeds 1GB the least
used compiled models are removed to save the filesystem space.
| Environment variable | Description |
|---|---|
| ZE_INTEL_NPU_CACHE_DIR= | The cache path. Set it to empty ("") to disable driver cache |
| ZE_INTEL_NPU_CACHE_SIZE= | The cache directory size. Whenever the cached files exceeds the size, some cached files are removed using the least recently used strategy |
Kernel module functional tests - vpu-kmd-test (from v1.5.0)
Introduced vpu-kmd-test in v1.5.0 release that can be used for intel_vpu
module validation. vpu-kmd-test is built together with the driver and can be
found in <build-dir>/bin
Support for Arrow Lake and Lunar Lake CPUs (from v1.5.0)
In v1.5.0 release introduced support for Arrow Lake and Lunar Lake family processors.
Log mask for granular control over logging (from v1.5.0)
In v1.5.0 release there is new environment variable ZE_INTEL_NPU_LOGMASK that
allows the user to enable specific log groups in driver. The
ZE_INTEL_NPU_LOGLEVEL variable is used to enable logging. The available log
level has been reduced to ERROR, WARNING and INFO. Whenever INFO is
set, the ZE_INTEL_NPU_LOGMASK allows to target specific log group. The log
group are listed in
umd/vpu_driver/source/utilities/log.hpp
# Set log level to INFO
export ZE_INTEL_NPU_LOGLEVEL=INFO
# Set log mask to only print from DEVICE, DRIVER and CACHE groups
export ZE_INTEL_NPU_LOGMASK=$((1<<4|1<<3|1<<17))
The Intel® NPU driver supports following Intel® processor families
| Platform | Kernel support | NPU driver release |
|---|---|---|
| Meteor Lake | v6.5.0-35-generic - Ubuntu22.04 Stock kernel | v1.1.0 |
| Arrow Lake | v6.8.12 | v1.5.0 |
| Lunar Lake | v6.8.12 | v1.5.0 |
The kernel module driver intel_vpu can be found in drivers/accel
tree
in the kernel sources.
Enable the intel_vpu module in the kernel config for compilation
In the Kernel Menu Configuration from Device Driver select Compute Acceleration Framework
and set "modularize" for Intel NPU (Neural Processing Unit).
Finding the intel_vpu kernel module in the system
# check if the intel_vpu exists is in the system
modinfo intel_vpu
# check if the intel_vpu is loaded in the kernel
lsmod | grep intel_vpu
# if the previous command nothing produced, load the intel_vpu
sudo modprobe intel_vpu
# verify that the intel_vpu has been loaded successfully
sudo dmesg | tail -n 20
# check if accel device is available
ls /dev/accel/accel0
Install the required dependencies in Ubuntu:
sudo apt update
sudo apt install -y build-essential git git-lfs cmake libudev-dev libboost-all-dev libssl-dev libudev-dev
Commands to build the driver:
cd linux-npu-driver
git submodule update --init --recursive
git lfs pull
cmake -B build -S .
cmake --build build --parallel $(nproc)
# set the LD_LIBRARY_PATH to lib to make driver visible for loader
export LD_LIBRARY_PATH=$PWD/build/lib:$LD_LIBRARY_PATH
# or install the driver in the system
cmake --install build --prefix /usr
# if possible reload the intel_vpu module to load new firmware
rmmod intel_vpu; modprobe intel_vpu
The driver has a target in compiler/vpux_compiler.cmake to build
Compiler-in-Driver component from NPU plugin
repository.
OpenVINO runtime is required by compiler. About the dependencies for building OpenVINO, please check the OpenVINO build document.
To build a compiler from the driver repository the ENABLE_VPUX_COMPILER_BUILD flag has to be set:
cd linux-npu-driver
cmake -B build -S . -DENABLE_VPUX_COMPILER_BUILD=ON
cmake --build build --parallel $(nproc)
# set the LD_LIBRARY_PATH to lib to make driver visible for loader
export LD_LIBRARY_PATH=$PWD/build/lib:$LD_LIBRARY_PATH
# or install the driver in the system
cmake --install build --prefix /usr
# if possible reload the intel_vpu module to load new firmware
rmmod intel_vpu; modprobe intel_vpu
The compiler binary libvpux_driver_compiler.so can be found in build/lib/.
The validation/umd-test directory contains vpu-umd-test application with functional tests.
This application allows to configure test content using YAML configuration file.
More information can be found in validation/umd-test/configs.
The binary vpu-umd-test is located in the build folder, ex. build/bin/
Command line:
./vpu-umd-test --config=basic.yaml
Non-root access to the NPU device
To access the NPU device, the user must be in the "render" or "video" group. A group depends on system configuration:
# check user groups
groups
# add user to the "render" group
sudo usermod -a -G render <user-name>
# log out and log in to apply the new group
The patch for systemd to set the "render" group for accel subsystem has been merged, but might not be available in your Linux distribution. See systemd change.
If setting the "render" group does not resolve the non-root access issue, this must be done by an administrator manually:
# check device permissions
ls -l /dev/accel/
# change group for accel device
sudo chown root:render /dev/accel/accel0
# allow to read/write from device for group
sudo chmod g+rw /dev/accel/accel0
# check final permissions
$ ls -lah /dev/accel/accel0
crw-rw---- 1 root render 261, 0 Jan 31 15:58 /dev/accel/accel0
Compilation problem due to lack of memory
The compilation may fail due to memory shortage. The recommendation is to use the Ninja generator instead of Unix Makefiles. If it does not help, please file a new issue.
# install Ninja
sudo apt update
sudo apt install -y ninja-build
# remove the old build and create a new one
rm build -rf
cmake -B build -S . -G Ninja
Enable driver log using an environment variable
Valid logging levels are ERROR, WARNING, INFO (and VERBOSE for driver
older than v1.5.0 release)
Seting the logging level using the ZE_INTEL_NPU_LOGLEVEL environment variable:
export ZE_INTEL_NPU_LOGLEVEL=<logging_level>
Command to clear an exported value:
unset ZE_INTEL_NPU_LOGLEVEL
Setting ZE_INTEL_NPU_LOGMASK allows to print specific log groups in driver.
The log group are listed in
umd/vpu_driver/source/utilities/log.hpp
# Set log level to INFO to enable LOGMASK
export ZE_INTEL_NPU_LOGLEVEL=INFO
# Set log mask to only print from DEVICE, DRIVER and CACHE groups
export ZE_INTEL_NPU_LOGMASK=$((1<<4|1<<3|1<<17))
Firmware version mismatch
The user can use different kernel and firmware combination for NPU device. The user might receive the following error message:
ERROR! MAPPED_INFERENCE_VERSION is NOT compatible with the ELF Expected: 6.1.0 vs received: 7.0.0
It means that NPU compiler mismatches the NPU firmware. To fix this issue the user needs to upgrade the firmware. Firmware update should be done from driver repository using release tag that matches the NPU compiler:
cmake -B build -S .
cmake --install build/ --component fw-npu --prefix /