Skip to content
This repository was archived by the owner on May 12, 2026. It is now read-only.
This repository was archived by the owner on May 12, 2026. It is now read-only.

Intel NPU - wrong/missing values #622

Closed
Closed
Intel NPU - wrong/missing values#622
Labels
bugSomething isn't working

Description

@sand-r
sand-r
opened on Feb 2, 2026

Is there an existing issue for this?

  • I searched the existing issues and did not find anything similar.

Current Behavior

On a Lunar Lake laptop (Intel NPU) the NPU page in Resources shows NPU Frequency and Link as N/A, but the kernel exposes related sysfs nodes under /sys/class/accel/accel0/device. Total usage calculation also seems to be off.

Device / driver

$ readlink -f /sys/class/accel/accel0/device/driver
/sys/bus/pci/drivers/intel_vpu

$ cat /sys/class/accel/accel0/device/uevent
DRIVER=intel_vpu
PCI_CLASS=120000
PCI_ID=8086:643E
PCI_SUBSYS_ID=17AA:2339
PCI_SLOT_NAME=0000:00:0b.0
MODALIAS=pci:v00008086d0000643Esv000017AAsd00002339bc12sc00i00

Frequency sysfs

ACCEL=/sys/class/accel/accel0/device

$ cat $ACCEL/npu_max_frequency_mhz
1900

$ cat $ACCEL/npu_current_frequency_mhz
0

Power related

$ cat $ACCEL/power_state
D3hot
$ cat $ACCEL/power/runtime_status
suspended

Counters present

$ cat $ACCEL/npu_busy_time_us
979473
$ cat $ACCEL/npu_memory_utilization
68714496

Link-related sysfs exists but probably irrelevant

for f in current_link_speed current_link_width max_link_speed max_link_width; do
  echo "== $f =="; cat "$ACCEL/$f"
done

== current_link_speed ==
Unknown
== current_link_width ==
0
== max_link_speed ==
Unknown
== max_link_width ==
255
Image

Expected Behavior

No response

Steps to Reproduce

No response

Debug Logs

Expand logs 
REPLACE THIS SENTENCE WITH THE TERMINAL OUTPUT OF THE AFOREMENTIONED COMMAND.

Environment



Anything Else?


No response
Activity
  1. sand-r
    added 
    bugSomething isn't working
     on Feb 2, 2026
  2. gordan-ramsey commented on Feb 16, 2026 
    @gordan-ramsey
    gordan-ramsey
    on Feb 16, 2026
    The same problem exist for my AMD NPU on AMD Ryzen 7 AI 350 but it's worse.
    
At least you are getting utilization info.
    

    user@pop-os:~$ readlink -f /sys/class/accel/accel0/device/driver
/sys/bus/pci/drivers/amdxdna
user@pop-os:~$ cat /sys/class/accel/accel0/device/uevent
DRIVER=amdxdna
PCI_CLASS=118000
PCI_ID=1022:17F0
PCI_SUBSYS_ID=1043:20CF
PCI_SLOT_NAME=0000:04:00.1
MODALIAS=pci:v00001022d000017F0sv00001043sd000020CFbc11sc80i00
user@pop-os:~$

    

    Linux pop-os 6.18.7-76061807-generic 
#202601231045~1769703228~24.04~cb87b5b SMP PREEMPT_DYNAMIC Thu J x86_64 x86_64 x86_64 GNU/Linux

    
Image
Image
  3. nokyan commented on Feb 25, 2026 
    @nokyan
    nokyan
    on Feb 25, 2026
    Owner
    Hi, thanks for the issue and sorry for taking so long to respond. Did this happen on Resources 1.9.1 as well, @sand-r? Also, is the NPU utilization metric in the Processes tab also that high for the processes utilizing your NPU?
  4. nokyan commented on Feb 25, 2026 
    @nokyan
    nokyan
    on Feb 25, 2026
    Owner
    Regarding core frequency on intel_vpu, I've committed to the branch bugfix/622-implausible-and-missing-values, I'd appreciate if you could check that out
  5. sand-r commented on Feb 26, 2026 
    @sand-r
    sand-r
    on Feb 26, 2026
    Author
    

    Regarding core frequency on intel_vpu, I've committed to the branch bugfix/622-implausible-and-missing-values, I'd appreciate if you could check that out
    

    

    I think you forgot to push that branch.
    
I could not find an NPU column on the processes tab to see how utilization is displayed there.
    

    I'm pretty sure it was also the same on 1.9.1 but if you let me know what should I test specifically, I can take a look.
  6. nokyan commented on Feb 26, 2026 
    @nokyan
    nokyan
    on Feb 26, 2026
    Owner
    Oh, you're correct, sorry, branch is now actually pushed. 😅
  7. sand-r commented on Feb 26, 2026 
    @sand-r
    sand-r
    on Feb 26, 2026
    Author
    

    Oh, you're correct, sorry, branch is now actually pushed. 😅
    

    

    Tested the branch, the frequency fix wasn't working, and usage is still off (correct first two digits but wrong decimal place). I had AI help me dig into  the code and it found two issues:
    

      

    1. core_frequency() passes a bare relative path to read_parsed, so it reads from the working directory instead of the sysfs device path. It silently fails and falls back to hwmon_core_frequency() (the old behavior). Needs self.sysfs_path().join() like usage() already does for npu_busy_time_us.
      2. 

    2. usage() divides npu_busy_time_us (microseconds) by the timestamp delta (seconds) without converting units, so the result is ~1,000,000x too large instead of a 0.0–1.0 fraction.
      3. 

    

      diff --git a/src/utils/npu/intel.rs b/src/utils/npu/intel.rs
  --- a/src/utils/npu/intel.rs
  +++ b/src/utils/npu/intel.rs
  @@ -81,7 +81,7 @@ impl NpuImpl for IntelNpu {
           let delta_timestamp = new_timestamp - last_timestamp;
           let delta_busy_time = new_busy_time.saturating_sub(last_busy_time) as f64;

  -        Ok(delta_busy_time / delta_timestamp)
  +        Ok(delta_busy_time / 1_000_000.0 / delta_timestamp)
       }

  @@ -100,7 +100,7 @@ impl NpuImpl for IntelNpu {

       fn core_frequency(&self) -> Result<f64> {
  -        read_parsed::<f64>("device/npu_current_frequency_mhz")
  +        read_parsed::<f64>(self.sysfs_path().join("device/npu_current_frequency_mhz"))
               .map(|mhz| mhz * 1_000_000.0)
               .or_else(|_| self.hwmon_core_frequency())
       }

    

    With both fixes applied, frequency and usage display correctly.
    
Image
  8. nokyan commented on Feb 26, 2026 
    @nokyan
    nokyan
    on Feb 26, 2026
    Owner
    Thanks a lot (though admittedly those errors are a bit embarassing for me 😅), I'll apply the fixes shortly. :)
  9. nokyan commented on Feb 26, 2026 
    @nokyan
    nokyan
    on Feb 26, 2026
    Owner
    I've committed the fix, mind testing it?
  10. sand-r commented on Feb 26, 2026 
    @sand-r
    sand-r
    on Feb 26, 2026
    Author
    Tested and it works, thank you!
  11. nokyan commented on Feb 26, 2026 
    @nokyan
    nokyan
    on Feb 26, 2026
    Owner
    In regards to amdxdna, @h-n-a, can you use the discussion starting here to check whether the amdxdna firmware is loading correctly?
    
Also, if you're familiar with sifting through log files, can you run either RUST_LOG=trace ~/.local/share/flatpak/app/net.nokyan.Resources/current/active/files/libexec/resources/resources-processes -o -r if you've installed Resources as a Flatpak locally or RUST_LOG=trace /var/lib/flatpak/app/net.nokyan.Resources/current/active/files/libexec/resources/resources-processes -o -r if you've installed Resources as a Flatpak system-wide while you're running an NPU ᚃorkload and look for lines containing "amdxdna" in the logs of the command?
  12. nokyan commented on Feb 27, 2026 
    @nokyan
    nokyan
    on Feb 27, 2026
    Owner
    As this issue was originally intended for Intel NPUs, I'll close it for now. :)
  13. nokyan
    closed this as completedon Feb 27, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Metadata
Metadata
Assignees
No one assigned
    Labels
    bugSomething isn't working
    Projects
    No projects
      Milestone
      No milestone
      Relationships
      None yet
      Development
      No branches or pull requests
      Issue actions