My FW16 seems to be turning off the dGPU with runtime power management, but the amdgpu driver then seems to be crashing, and the device is removed from my kernel, no longer showing up in nvtop or accessible to games, (but does still show up in lspci). If I use this to reload the card, it comes back, and after a while crashes again:

echo 1 | sudo tee /sys/bus/pci/devices/0000\:03\:00.0/remove
echo 1 | sudo tee /sys/bus/pci/devices/0000\:03\:00.1/remove
echo 1 | sudo tee /sys/bus/pci/rescan

Anyone else having this issue or know how to fix it?

Relevant part of my configuration.nix:

{
  config,
  inputs,
  lib,
  pkgs,
  ...
}:
let
  amdgpu-kernel-module = pkgs.callPackage ../../nixos_modules/patches/amdgpu-kernel-module.nix {
    # Make sure the module targets the same kernel as your system is using.
    kernel = config.boot.kernelPackages.kernel;
  };
  # linuxPackages_latest 6.13 (or linuxPackages_zen 6.13)
  # amdgpu-stability-patch = pkgs.fetchpatch {
  #   name = "amdgpu-stability-patch";
  #   url = "https://github.com/torvalds/linux/compare/ffd294d346d185b70e28b1a28abe367bbfe53c04...SeryogaBrigada:linux:4c55a12d64d769f925ef049dd6a92166f7841453.diff";
  #   hash = "sha256-q/gWUPmKHFBHp7V15BW4ixfUn1kaeJhgDs0okeOGG9c=";
  # };
in
{
  boot = {
    extraModulePackages = [
      (amdgpu-kernel-module.overrideAttrs (_: {
        patches = [
          ./0001-drm-amdgpu-mes11-optimize-MES-pipe-FW-version-fetchi.patch
          # amdgpu-stability-patch
        ];
      }))
    ];
  };
}

patch file itself, which can be found on the first AMD link:

From 2771f96336e3c469622f6e0e132f24ad1ba6a5c6 Mon Sep 17 00:00:00 2001
From: Alex Deucher <alexander.deucher@amd.com>
Date: Thu, 27 Mar 2025 17:33:49 -0400
Subject: [PATCH] drm/amdgpu/mes11: optimize MES pipe FW version fetching

Don't fetch it again if we already have it.  It seems the
don't reliably have the proper value at resume in some
cases.

Closes: https://gitlab.freedesktop.org/drm/amd/-/issues/4083
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
---
 drivers/gpu/drm/amd/amdgpu/mes_v11_0.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
index 16f8bc36afa07..06b51867c9aac 100644
--- a/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
+++ b/drivers/gpu/drm/amd/amdgpu/mes_v11_0.c
@@ -895,6 +895,10 @@ static void mes_v11_0_get_fw_version(struct amdgpu_device *adev)
 {
 	int pipe;
 
+	/* return early if we have already fetched these */
+	if (adev->mes.sched_version && adev->mes.kiq_version)
+		return;
+
 	/* get MES scheduler/KIQ versions */
 	mutex_lock(&adev->srbm_mutex);
 
-- 
2.49.0

forgot one other file needed. This one is amdgpu-kernel-module.nix, referenced in the configuration.nix as well. This is the part that packages the amdgpu module separately from the kernel:

{
  pkgs,
  lib,
  kernel ? pkgs.linuxPackages_latest.kernel,
}:

pkgs.stdenv.mkDerivation {
  pname = "amdgpu-kernel-module";
  inherit (kernel)
    src
    version
    postPatch
    nativeBuildInputs
    ;

  kernel_dev = kernel.dev;
  kernelVersion = kernel.modDirVersion;

  modulePath = "drivers/gpu/drm/amd/amdgpu";

  buildPhase = ''
    BUILT_KERNEL=$kernel_dev/lib/modules/$kernelVersion/build

    cp $BUILT_KERNEL/Module.symvers .
    cp $BUILT_KERNEL/.config        .
    cp $kernel_dev/vmlinux          .

    make "-j$NIX_BUILD_CORES" modules_prepare
    make "-j$NIX_BUILD_CORES" M=$modulePath modules
  '';

  installPhase = ''
    make \
      INSTALL_MOD_PATH="$out" \
      XZ="xz -T$NIX_BUILD_CORES" \
      M="$modulePath" \
      modules_install
  '';

  meta = {
    description = "AMD GPU kernel module";
    license = lib.licenses.gpl3;
  };
}