I’m trying to profile my game but I can’t profile the GPU, I updated my graphics card drivers but I still see this message on the profiler, what’s happening here guys?
I also saw this:
… by design?!
Thanks in advance!
1 Like
If anyone else is seeing this behaviour I recommend turning off “Graphics Jobs (Experimental)” in the Player settings and trying again. That seems to consistently fix the problem in at least a couple of my projects.
Edit: This has been reported and is in Unity’s issue tracker: https://issuetracker.unity3d.com/issues/gpu-profiler-doesnt-work-when-graphics-jobs-and-vr-mode-is-on
11 Likes
That did it! Thanks man!
2 Likes
Ah cool, glad to help someone!
Just remember to turn it back on again if you’re counting on the Graphics Jobs for CPU performance. 
2 Likes
Thanks again! I will remember!!
I needed to turn the “Graphics Jobs (Experimental)” on for the GPU profiler to work (Unity 2018.3.5f1).
Same. Had to turn “Graphics Jobs (Experimental)” ON for the GPU Profiler to work. (Unity 2018.3.6f1, newest drivers installed)
I had to turn “Graphics Jobs (Experimental)” off, then on.
Enabling and disabling doesn’t work for me using Unity 2019.1.2f1 and macOS Mojave 10.14.3.
Anyone have a fix?
This frustrating error because how are you supposed to debug without this info?
GTX1060 mobile users still have the issue with the GPU not being used for rendering. Turning this setting on at least improves performance by spreading rendering load across cores, but most likely not the GPU. The “GPU profiling is not supported by the graphics card driver” message is shown in the profiler! I saw posts from several years back reporting on this issue already, with older GTX mobile GPUs. Whats going on Unity?!
Edit: The GPU needs to be added to the profiler manually, but actually works after enabling “Graphics Jobs (Experimental).”
Old thread but still valid on macOS and it is still a limitation in Unity
pls see attached screenshot
so, (at least on mac)
- Graphics Jobs should be turned OFF
- must not use Metal (only OpenGL supported for profiling that is not really supported anymore, Metal is the preferred)
Seems we still should wait for Metal profiling support
Yeap, a long-standing issue and one complicated by several factors such as fragmentation across devices and platforms, different Graphics APIs, the current GPU Usage display in the Profiler window making weird and now slightly outdated assumptions about how GPU time would make sense to track, report and display …
Anyways, with 2022.1 we now got to the point where we can (at the least) get high-level GPU timings from all platforms and Graphics APIs (except for WebGL and Linux OpenGL) via Frame Timing Manager and ProfilerRecorder .
Note: that won’t fix the issue with the GPU Usage Profiler Module, that one still has the same assumption issues as before and the UI needs a mayor overhaul to be able to digest and display these measurements in a way that makes sense. (There are also still workflow issues like progressive disclosure to be taken into account for ideal profiling overhead as every GPU timing query added can have quite the performance impact, depending on the platform/Graphics API, so just outright peppering the GPU work with all possible Queries for the full details may be a bit easier to implement but not necessarily the best approach…)
As for the topical message of this thread, i.e. "GPU profiling not supported by graphics card driver"
That was a bit of a catch-all for when the profiler tried to get GPU timer queries installed in OpenGL but got nothing back. It’s very likely it was a correct statement when introduced way back when but new Graphics APIs and features where added which didn’t mash well with said bad assumptions in the UI so they just didn’t report any GPU timings, and then the same message would appear.
Starting from 2018.3 iirc the message should be corrected to give the exact reason why it isn’t supported. You may still be unlucky and get an Android phone where the OEM hasn’t implemented GL Timer Query capabilities for the device, so you might still get that message, but that was never an issue we could fix on our side anyways.