## System details

- **CPU:** Intel Core Ultra 9 275HX (Arrow Lake)
- **GPU (integrated):** Intel Arrow Lake-S — drives the internal eDP panel
- **GPU (discrete):** NVIDIA GeForce RTX 5080 Max-Q (no display connected; HDMI/one DP are wired to it)
- **Drivers:** nvidia-open-dkms 610.57.04, Mesa 26.2.1, intel-media-driver 26.2.4, libva-nvidia-driver 0.0.17
- **Omarchy:** 4.0.1
- **Kernel:** 7.1.9-arch1-2
- **Laptop:** Lenovo Legion 7i Pro (99.9 Wh battery)

## Problem

On hybrid laptops where the **iGPU drives the panel**, `default/hypr/nvidia.lua` detects the NVIDIA GPU and unconditionally sets, during config parse:

```lua
hl.env("NVD_BACKEND", "direct")
hl.env("LIBVA_DRIVER_NAME", "nvidia")
hl.env("__GLX_VENDOR_LIBRARY_NAME", "nvidia")
```

Then `default/hypr/autostart.lua` exports the resulting environment session-wide on `hyprland.start`:

```lua
hl.exec_cmd("systemctl --user import-environment $(env | cut -d'=' -f 1)")
hl.exec_cmd("dbus-update-activation-environment --systemd --all")
```

So every app (launcher, web apps, Electron apps, browsers) inherits `LIBVA_DRIVER_NAME=nvidia` and `__GLX_VENDOR_LIBRARY_NAME=nvidia`. On an iGPU-primary hybrid this has three consequences:

### 1. Cross-GPU video corruption (the root cause behind #4901 / #3899)

Chromium/Electron apps decode video on the NVIDIA GPU (NVDEC via libva-nvidia-driver) while presenting on the Intel GPU. Importing those DMA-BUFs into the iGPU fails (`eglCreateImage → EGL_BAD_MATCH`), producing corrupted video, ghosting while scrolling, and stuttering animations. #4901 documents the Chromium symptoms and workarounds in depth; the reason `vainfo` defaults to NVDEC in the first place is this `hl.env` in `nvidia.lua`.

It's not just Chromium: any Electron app is affected (reproduced with Discord GIF rendering and a custom Electron app whose particle animations froze periodically).

### 2. The dGPU can never runtime-suspend → measurable battery drain

With `LIBVA_DRIVER_NAME=nvidia`, any app that touches video decode opens CUDA (`/dev/nvidia-uvm`). An open `nvidia-uvm` fd blocks RTD3 runtime suspend unconditionally (even with `DynamicPowerManagement` fine-grained active), pinning the dGPU at ~5 W forever.

Measured on this machine (99.9 Wh battery, ~15 W idle draw): the dGPU alone accounts for ~1/3 of idle battery drain. Verified with `/sys/bus/pci/devices/<dgpu>/power/runtime_suspended_time`: 0 ms of suspension while a browser was open; with `LIBVA_DRIVER_NAME=iHD` the dGPU stays in D3cold **even during YouTube playback** (decode runs on the Intel media engine, which is also more power-efficient).

### 3. User overrides are silently defeated

`hl.env` is last-wins during parse, and `nvidia.lua` runs inside `require("default.hypr.omarchy")`. A user who sets `hl.env("LIBVA_DRIVER_NAME", "iHD")` at the top of `hyprland.lua` (before the require, where the template invites configuration) gets silently overridden. Debugging this is extra confusing because `/proc/<hyprland-pid>/environ` still shows the exec-time environment, not the parse-time `setenv` values that children actually inherit.

## Expected behavior

On hybrids where the boot display (eDP) is connected to the iGPU, video decode and GLX should default to the iGPU. The NVIDIA-specific env should apply only when the NVIDIA GPU is primary (drives the connected display), e.g.:

```bash
# panel-on-iGPU detection sketch: check which card exposes the connected eDP
for c in /sys/class/drm/card*-eDP-*; do
  [[ $(< "$c/status") == connected ]] && ...
done
```

Alternatively (or additionally), `nvidia.lua` could skip each `hl.env` when the variable is already set by user config.

## Workaround that fixes everything (validated)

In `~/.config/hypr/hyprland.lua`, **after** `require("default.hypr.omarchy")`:

```lua
hl.env("LIBVA_DRIVER_NAME", "iHD")           -- radeonsi for AMD iGPUs
hl.env("__GLX_VENDOR_LIBRARY_NAME", "mesa")
```

Games that should use the dGPU keep working by scoping `__NV_PRIME_RENDER_OFFLOAD=1` etc. to their own launch scripts.

Results after the override + relogin: no video corruption in Chromium/Electron, dGPU suspends (D3cold) even during video playback, idle draw dropped ~5 W (~6.5 h → ~9 h estimated on this machine).

## Steps to reproduce

1. Omarchy on a hybrid laptop whose eDP panel is on the iGPU (mux in hybrid mode), with `libva-nvidia-driver` installed.
2. Boot, run `vainfo` → reports "VA-API NVDEC driver" instead of the iGPU driver.
3. `systemctl --user show-environment | grep -E 'LIBVA|GLX'` → `nvidia` values.
4. Open YouTube in Chromium; watch `/sys/bus/pci/devices/<dgpu>/power/runtime_status` stay `active` (and `runtime_suspended_time` frozen) while the Chromium GPU process holds `/dev/nvidia-uvm`.
5. Apply the override above, relogin, repeat: `iHD`, uvm closed, dGPU `suspended`.