> Reopened with a confirmed single-variable reproduction. An earlier revision of
> this issue was withdrawn because I had asserted a cause I had not yet
> demonstrated; that evidence now exists and is below.

## Problem

`default/hypr/nvidia.lua` sets `LIBVA_DRIVER_NAME=nvidia` whenever an NVIDIA GPU
with GSP is detected:

```lua
if o.shell_succeeds(o.shell_quote(nvidia_gsp)) then
  hl.env("NVD_BACKEND", "direct")
  hl.env("LIBVA_DRIVER_NAME", "nvidia")
  hl.env("__GLX_VENDOR_LIBRARY_NAME", "nvidia")
```

The condition tests only whether such a GPU is *present*, never whether it
drives any output. On a hybrid laptop whose panel hangs off the iGPU, this
forces browsers onto the NVIDIA VA-API path while compositing and scanout happen
on the other GPU.

On this Razer Blade the NVIDIA dGPU drives **nothing**:

```
card1-DP-1     [NVIDIA] status=disconnected enabled=disabled
card1-DP-2     [NVIDIA] status=disconnected enabled=disabled
card1-HDMI-A-1 [NVIDIA] status=disconnected enabled=disabled
card2-eDP-1    [AMD]    status=connected    enabled=enabled
```

## Symptom

Severe scroll stalling and jitter in Chrome on video-heavy pages — reddit.com
main feed, espn.com. The page hitches hard enough to feel stuck. Static pages
are perfectly smooth, which makes this look like a browser or mesa regression
and sends you chasing the wrong thing.

## Reproduction

Changing only `LIBVA_DRIVER_NAME` and restarting Chrome, with nothing else
touched:

| | `LIBVA_DRIVER_NAME=nvidia` (Omarchy default) | `LIBVA_DRIVER_NAME=radeonsi` |
|---|---|---|
| `nvidia_drv_video.so` in GPU process | loaded | not loaded |
| `/dev/nvidia0` fds held by GPU process | 23 | 0 |
| scrolling reddit/espn | stalls badly | smooth |

Measured with:

```bash
GP=$(pgrep -f 'type=gpu-process')
grep -o '/usr/lib/dri/[a-z0-9_]*_drv_video\.so' /proc/$GP/maps | sort -u
ls -l /proc/$GP/fd | grep -oE 'renderD[0-9]+|nvidia[a-z0-9]*' | sort | uniq -c
```

Chrome self-selects the AMD node either way
(`--render-node-override=/dev/dri/renderD129`), so with the default it is
compositing on AMD while decoding on NVIDIA.

Worth noting: switching to `radeonsi` does **not** cost hardware decode — it
gains it. With `radeonsi` set, Chrome constructs a `VaapiVideoDecoder` for h264
and vp9 and decodes on the iGPU:

```
media/gpu/vaapi/vaapi_video_decoder.cc:136] VaapiVideoDecoder():
media/gpu/chromeos/video_decoder_pipeline.cc:1225] PickDecoderOutputFormat():
  Initializing frame pool with up to 25 VideoFrames. No ImageProcessor needed.
```

Measured over 10s of 720p playback: 660 ms CPU with hardware decode vs 1560 ms
with `--disable-accelerated-video-decode`. (AV1 is unsupported by this VCN 2.2
engine — `Video configuration is not supported` — and falls back to software
under either setting, so it is unaffected either way.)

So the current default is strictly worse: it engages a decode path on a GPU that
drives no display, and gives up working iGPU decode to do it.

## Suggested fix

Gate the `LIBVA_DRIVER_NAME` assignment on the NVIDIA GPU actually driving a
connected output, rather than merely being present — e.g. check for any
`connected` connector under the NVIDIA card in `/sys/class/drm/`. That reads
cached sysfs only, so it respects the constraint already documented in the
comment above this block about `lspci` waking a runtime-suspended GPU.

`NVD_BACKEND` and `__GLX_VENDOR_LIBRARY_NAME` appear harmless here; only the
VA-API variable caused trouble.

## Note for anyone overriding this locally

`hl.env` in `~/.config/hypr/hyprland.lua` looks like it works but does nothing
until you log out. Hyprland pushes `hl.env` into the systemd user environment at
session start, and `hyprctl reload` does not refresh it, so a reloaded config
leaves the old value in place for uwsm-launched apps
(`app-com.google.Chrome-<pid>.scope`, parent `systemd --user`). Either re-login,
or set it in `~/.config/environment.d/`.

## System details

- Omarchy 4.0.2-1
- Kernel 7.1.9-arch1-2 (Arch), Hyprland on Wayland
- GPU 0: NVIDIA GA104M [RTX 3080 Mobile] — `nvidia-open-dkms` 610.57.04, `card1` / `renderD128`
- GPU 1: AMD Cezanne [Radeon Vega] — `amdgpu`, `card2` / `renderD129`, drives `eDP-1`
- mesa 1:26.2.1-1, vulkan-radeon 1:26.2.1-1
- Google Chrome 152.0.7977.82