VAAPI on N100 Stops Working After 2 Hours, Requires Reboot #16438
Replies: 10 comments 16 replies
-
|
This is likely a bug in the intel_gpu_top tool, I don't think there is any issue with the GPU itself, qsv should be used on this hardware |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. now happens something very strange, when i use VAAPI both camera works fine. when i use qsv, only one camera working. my back camera works good. my front camera doesn't work at all. i added the logs when i try to use only the front camera. 2025-02-14 09:56:59.502341557 [2025-02-14 11:56:59] ffmpeg.entrancezoom.detect ERROR : [vist#0:0/h264 @ 0x564f90aff8c0] [dec:h264_qsv @ 0x564f90ab3e80] Decoding error: Invalid argument |
Beta Was this translation helpful? Give feedback.
-
|
Something has changed Now the VAAPI works with 0.4%. This is a small change but it's still something. Just for the demonstration I did another reboot To see how it's should be. I was thinking about running Frigate directly through Ubuntu in this case i need a new pc for that use case. |
Beta Was this translation helpful? Give feedback.
-
|
I'm having issues with VAAPI on N100 aswell (see: #16609). Mine is totally not working (0% GPU), just wondering how you got yours to work. |
Beta Was this translation helpful? Give feedback.
-
|
Any solutions? Works few hours and after this time GPU 0% |
Beta Was this translation helpful? Give feedback.
-
|
Adding my own experience here. Did a fresh Debian on N100 bare metal install with docker, frigate, coral USB. See ~5% GPU for the first few hours. Then later it is ~0.37%. CPU stays roughly the same FWIW so that may support this simply being a reporting bug but would be good to know for sure. Restarting the container does not change the GPU reporting but rebooting the whole system does. I didn't install any Coral or Intel GPU drivers on Debian assuming the docker container takes care of both. Not sure if that is the correct approach but it seems to work. |
Beta Was this translation helpful? Give feedback.
-
|
have the same issue on my N100 running home assistant with Frigate as an addon. (have tested both Beta and main release). Everything works for a few hours, then suddenly Frigate stops using GPU, and CPU skyrockets. Rebooting the Home Assisstant machine resolves the issue until it repeats itself. I'm using a identical hw on a secondary site, however with Proxmox instead, and Frigate as a docker and a seperate VM. Same config otherwise, but at this location no issue. If I we're to guess there seems the 'res some sort of bug in the HAOS Docker environment causing frigate addon with GPU passthrough not working properly/or stop working |
Beta Was this translation helpful? Give feedback.
-
|
I have the same issue, N100 on Debian, Frigate 0.15 (vaapi) hardware acceleration works for a few hours, then it ssuddenly stops working. needs reboot. always the same. |
Beta Was this translation helpful? Give feedback.
-
|
I don't know what is the fix exactly, but it looks like vaapi is finally working properly in my N100 with Debian 12. I tried to make QSV work, but I could not get it working, althought vaapi got fixed in the process. It was a hell, I cannot say how many things I tried, but resuming: I installed/updated too many things, drivers, intel-media-sdk, backports of the kernel, and anything related to graphics that could cause a problem. That looks like it fixed the problem of vaapi for me. (by now, after 5 days it keeps running) |
Beta Was this translation helpful? Give feedback.
-
|
Hello @Wasted-Engineer, I would like to understand, did you install the custom ffmpeg and solve it? Here I also tried everything, updated several colors such as Linux Debian 12 with Kernel 6.12.32+bpo-amd64 via bookworm-backports repository, updated iHD driver intel-media-va-driver-non-free 23.3.4 and Google Coral updated with the official components of Github. Other things I also did was set memory and resource limits for the GPU, through the GRUB, which also did not solve. Thank you if you can share in detail what you did to reproduce here on my N100. |
Beta Was this translation helpful? Give feedback.
-
|
I'm sorry, I'm Brazilian and I don't think the translation was very good. |
Beta Was this translation helpful? Give feedback.
-
|
@Wasted-Engineer can you tell me the kernel version you are running currently |
Beta Was this translation helpful? Give feedback.
-
|
I'm currently running kernel version: Let me know if you'd like me to share any additional system details. |
Beta Was this translation helpful? Give feedback.
-
|
I never was able to find an officialish source that this is a bug in intel_gpu_top but do we think this is the same thing/the right place to maybe have some more activity? https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/issues/175 I tried |
Beta Was this translation helpful? Give feedback.
-
|
After updating everything possible — kernel (via backports), Intel iHD video drivers, and even Google Coral drivers — unfortunately, none of these steps permanently resolved the issue where However, I recently decided to follow the Frigate documentation and replaced the built-in I mapped this custom
I’ll keep monitoring and will share more findings as I go. |
Beta Was this translation helpful? Give feedback.
-
|
I am running my compiled kernel with the patch applied and will report back if it fixes the problem or is unrelated. @fernandolcardoso the patch I was looking at was not in the kernel (via backports) that you were running. Please report back if your |
Beta Was this translation helpful? Give feedback.
-
|
Despite several adjustments, the runtime was longer this time — around 8 hours — but unfortunately, the issue wasn't resolved. So I’ve decided to run a series of targeted tests now, aiming to pinpoint something more specific. |
Beta Was this translation helpful? Give feedback.
-
|
I would run the mainline kernel or at least 6.14 as a test if you are eager to try something. I am at 24 hrs with my patched kernel so far. I would change no other variables. |
Beta Was this translation helpful? Give feedback.
-
|
My testing is still inconclusive. The behavior seems changed but still not quite right, I might revisit it. |
Beta Was this translation helpful? Give feedback.
-
|
"Here are my current specs for VA-API on Debian 12 with Intel N100:* Key fixes I applied: Added user to render and video groups. And installed the things I say in my first post, I cannot remember if I updated something else... But I confirm is working like a charm, no stops anymore. |
Beta Was this translation helpful? Give feedback.
-
|
So, I continue monitoring, now with all the kernel, gpu + firmware updates, just like what was reported here, but I still suspect that something is wrong.
|
Beta Was this translation helpful? Give feedback.
-
System Configuration and FFmpeg Testing Plan (Frigate 0.16.0-beta3)I'm participating in this discussion because I’m facing exactly the same issue: after about 1 to 2 hours of continuous usage, the GPU (Intel N100 / Alder Lake-N) stops reporting any activity. Both It appears that VAAPI acceleration or the iHD driver itself silently stops working. Several other users have reported the same behavior, which suggests the issue may lie somewhere between the video driver, FFmpeg integration, or kernel-level handling. To ensure a controlled and clean test environment, I fully reformatted the system and reinstalled Debian 12 (Bookworm) from scratch, leaving it in its most basic form, without graphical interface and without third-party or nonfree repositories. My current setup:
Testing PlanI'm starting a new round of diagnostics with the following goals:
If anyone is conducting similar testing or has ideas for additional diagnostic commands or kernel tweaks, I’m happy to collaborate. Once I identify a reproducible behavior or narrow down the root cause, I’ll update this thread with findings. |
Beta Was this translation helpful? Give feedback.
-
|
My test is still going strong...I would wait another day to hear back from me before putting in more effort to be honest What was your result from yesterday #16438 (reply in thread) |
Beta Was this translation helpful? Give feedback.
-
Update on my testingAlthough I initially started with a clean, minimal environment using the most basic versions of drivers and firmware — including a custom static FFmpeg build — the issue persisted. After confirming that the problem was not resolved by downgrading components, I proceeded to update everything: kernel, Intel GPU drivers, firmware, and other dependencies to their latest stable or backports versions. Unfortunately, the behavior remains unchanged. After approximately 1 to 2 hours of continuous usage, the GPU performance drops dramatically, especially when using advanced features like:
In this degraded state:
However, when I disable the features mentioned above and run Frigate with only basic object detection, the GPU remains stable, maintaining an average of 3% usage for multiple days with no signs of failure. This strongly suggests a potential issue related to:
I'll continue investigating, and I’m open to collaboration. If anyone has suggestions (e.g., kernel/sysctl tweaks, FFmpeg parameters, i915 module options), feel free to share.
|
Beta Was this translation helpful? Give feedback.
-
|
My setup
My config is as follows |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Describe the problem you are having
I am using Frigate as an add-on in Home Assistant. Additionally, I have integrated Double Take, Deepstack, CompreFace, and Plate Recognizer. The system is set up to detect humans as well as vehicles and license plates. Currently, I have only two Reolink cameras connected.
I have been trying for a while to get the GPU to work. I have also tried VAAPI and QSV, but neither worked. It constantly shows 0% usage. The only way I can get it to work is by performing a full reboot of the computer, after which it functions for about two hours only.
I have repeated this reboot process three times, and each time the behavior is the same—it works for about two to three hours and then stays at 0% usage without recovering.
I saw that I can try using the i965 environment variable, but I got an error stating that i965 couldn't initialize. When I disable it, the system works for the two hours I mentioned earlier.
Version
0.15.0-cea210d
Frigate config file
docker-compose file or Docker CLI command
Relevant Frigate log output
Relevant go2rtc log output
FFprobe output from your camera
Install method
HassOS Addon
Object Detector
Coral
Network connection
Wired
Camera make and model
Reolink Tracemix wifi
Screenshots of the Frigate UI's System metrics pages
Any other information that may be helpful
{"return_code":0,"stderr":"","stdout":"vainfo: VA-API version: 1.22 (libva 2.10.0)
vainfo: Driver version: Intel iHD driver for Intel(R) Gen Graphics - 24.3.3 ()
vainfo: Supported profile and entrypoints
VAProfileNone : VAEntrypointVideoProc
VAProfileNone : VAEntrypointStats
VAProfileMPEG2Simple : VAEntrypointVLD
VAProfileMPEG2Simple : VAEntrypointEncSlice
VAProfileMPEG2Main : VAEntrypointVLD
VAProfileMPEG2Main : VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointVLD
VAProfileH264Main : VAEntrypointEncSlice
VAProfileH264Main : VAEntrypointFEI
VAProfileH264Main : VAEntrypointEncSliceLP
VAProfileH264High : VAEntrypointVLD
VAProfileH264High : VAEntrypointEncSlice
VAProfileH264High : VAEntrypointFEI
VAProfileH264High : VAEntrypointEncSliceLP
VAProfileVC1Simple : VAEntrypointVLD
VAProfileVC1Main : VAEntrypointVLD
VAProfileVC1Advanced : VAEntrypointVLD
VAProfileJPEGBaseline : VAEntrypointVLD
VAProfileJPEGBaseline : VAEntrypointEncPicture
VAProfileH264ConstrainedBaseline: VAEntrypointVLD
VAProfileH264ConstrainedBaseline: VAEntrypointEncSlice
VAProfileH264ConstrainedBaseline: VAEntrypointFEI
VAProfileH264ConstrainedBaseline: VAEntrypointEncSliceLP
VAProfileVP8Version0_3 : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointVLD
VAProfileHEVCMain : VAEntrypointEncSlice
VAProfileHEVCMain : VAEntrypointFEI
VAProfileHEVCMain : VAEntrypointEncSliceLP
VAProfileHEVCMain10 : VAEntrypointVLD
VAProfileHEVCMain10 : VAEntrypointEncSlice
VAProfileHEVCMain10 : VAEntrypointEncSliceLP
VAProfileVP9Profile0 : VAEntrypointVLD
VAProfileVP9Profile0 : VAEntrypointEncSliceLP
VAProfileVP9Profile1 : VAEntrypointVLD
VAProfileVP9Profile1 : VAEntrypointEncSliceLP
VAProfileVP9Profile2 : VAEntrypointVLD
VAProfileVP9Profile2 : VAEntrypointEncSliceLP
VAProfileVP9Profile3 : VAEntrypointVLD
VAProfileVP9Profile3 : VAEntrypointEncSliceLP
VAProfileHEVCMain12 : VAEntrypointVLD
VAProfileHEVCMain12 : VAEntrypointEncSlice
VAProfileHEVCMain422_10 : VAEntrypointVLD
VAProfileHEVCMain422_10 : VAEntrypointEncSlice
VAProfileHEVCMain422_12 : VAEntrypointVLD
VAProfileHEVCMain422_12 : VAEntrypointEncSlice
VAProfileHEVCMain444 : VAEntrypointVLD
VAProfileHEVCMain444 : VAEntrypointEncSliceLP
VAProfileHEVCMain444_10 : VAEntrypointVLD
VAProfileHEVCMain444_10 : VAEntrypointEncSliceLP
VAProfileHEVCMain444_12 : VAEntrypointVLD
VAProfileHEVCSccMain : VAEntrypointVLD
VAProfileHEVCSccMain : VAEntrypointEncSliceLP
VAProfileHEVCSccMain10 : VAEntrypointVLD
VAProfileHEVCSccMain10 : VAEntrypointEncSliceLP
VAProfileHEVCSccMain444 : VAEntrypointVLD
VAProfileHEVCSccMain444 : VAEntrypointEncSliceLP
VAProfileAV1Profile0 : VAEntrypointVLD
VAProfileHEVCSccMain444_10 : VAEntrypointVLD
VAProfileHEVCSccMain444_10 : VAEntrypointEncSliceLP"}
No response
Beta Was this translation helpful? Give feedback.
All reactions