cuda support and immich cuda support

5 files changed, 102 insertions, 20 deletions

diff --git a/archetypes/profiles/cuda/default.nix b/archetypes/profiles/cuda/default.nix
new file mode 100644
index 0000000..14de560
--- /dev/null
+++ b/archetypes/profiles/cuda/default.nix
@@ -0,0 +1,28 @@
+{ lib, pkgs, ... }: let
+	mkCuda = lib.mkOverride 800;
+
+	nixosConfig = {
+		# Binary cache packages built with cudaSupport
+		nix.settings = {
+			substituters = mkCuda [ "https://cache.nixos-cuda.org" ];
+			trusted-public-keys = mkCuda [ "cache.nixos-cuda.org:74DUi4Ye579gUqzH4ziL9IyiJBlDpMRn9MBN8oNan9M=" ];
+		};
+
+		# Enable cudaSupport
+		nixpkgs.config.cudaSupport = mkCuda true;
+
+		# Nvidia packages
+		environment.systemPackages = with pkgs; mkCuda [
+			config.hardware.nvidia.package  # nvidia-smi
+			nvtopPackages.nvidia
+		];
+
+		# Enable nvidia graphics
+		services.xserver.videoDrivers = mkCuda [ "nvidia" ];  # xserver.videoDrivers does not imply X
+		hardware.graphics.enable = mkCuda true;
+	};
+
+	homeConfig = {};
+in {
+	imports = [ (lib._mkProfileArchetype "cuda" nixosConfig homeConfig) ];
+}
diff --git a/archetypes/profiles/default.nix b/archetypes/profiles/default.nix
index 23b5c05..ef5d87b 100644
--- a/archetypes/profiles/default.nix
+++ b/archetypes/profiles/default.nix
@@ -1,6 +1,7 @@
 {
 	imports = [
 		./btrfs
+		./cuda
 		./desktop
 		./headless
 		./pi
diff --git a/hosts/optiplex/configuration.nix b/hosts/optiplex/configuration.nix
index 74d09b5..5108e72 100644
--- a/hosts/optiplex/configuration.nix
+++ b/hosts/optiplex/configuration.nix
@@ -80,6 +80,7 @@
 	hardware.nvidia = {
 		modesetting.enable = true;  # Required
 		powerManagement.enable = false;  # Can cause bugs
+		powerManagement.finegrained = false;  # Only works on turing and newer
 		nvidiaSettings = true;
 		open = false;  # Not compatible w/ GTX-1050
 		package = config.boot.kernelPackages.nvidiaPackages.legacy_580;
diff --git a/hosts/poweredge/configuration.nix b/hosts/poweredge/configuration.nix
index 3935daf..40803bf 100644
--- a/hosts/poweredge/configuration.nix
+++ b/hosts/poweredge/configuration.nix
@@ -11,6 +11,9 @@
 		#./notification-mailer.nix  # TODO move some of this stuff to archetype
 	];
 
+	# Allow unfree packages for nvidia/cuda
+	nixpkgs.config.allowUnfree = true;
+
 	# Setup bootloader
 	boot._loader.enable = true;
 
@@ -22,6 +25,7 @@
 				home.users.timmy.enable = true;
 			};
 			zfs.enable = true;
+			cuda.enable = true;
 		};
 	};
 
@@ -30,8 +34,12 @@
 
 	# Enable nvidia
 	hardware.nvidia = {
-		open = false;  # Need for NVENC
-		package = config.boot.kernelPackages.nvidiaPackages.stable;
+		modesetting.enable = true;  # Required
+		nvidiaSettings = false;  # Don't need gui
+		open = false;
+		package = config.boot.kernelPackages.nvidiaPackages.legacy_580;  # Support for P600
+		powerManagement.enable = false;  # Can cause bugs
+		powerManagement.finegrained = false;  # Only works on turing and newer
 	};
 
 	# Enable user timmy
diff --git a/hosts/poweredge/immich.nix b/hosts/poweredge/immich.nix
index 51d7885..24b32c2 100644
--- a/hosts/poweredge/immich.nix
+++ b/hosts/poweredge/immich.nix
@@ -1,4 +1,6 @@
-{
+{ pkgs, ... }: let
+	onnxruntimeCuda = pkgs.onnxruntime.override { cudaSupport = true; };
+in {
 	containers.immich = {
 		autoStart = true;
 		privateNetwork = true;
@@ -14,30 +16,55 @@
 		};
 
 		# GPU
-		#allowedDevices = [
-		#	{ node = "/dev/dri/card1"; modifier = "rw"; }
-		#	{ node = "/dev/dri/renderD128"; modifier = "rw"; }
-		#];
-
-		#bindMounts = {
-		#	"/dev/dri/card1" = {
-		#		hostPath = "/dev/dri/card1";
-		#		isReadOnly = false;
-		#	};
-		#	"/dev/dri/renderD128" = {
-		#		hostPath = "/dev/dri/renderD128";
-		#		isReadOnly = false;
-		#	};
-		#};
+		allowedDevices = [
+			{ node = "/dev/nvidia0";           modifier = "rw"; }
+			{ node = "/dev/nvidiactl";         modifier = "rw"; }
+			{ node = "/dev/nvidia-uvm";        modifier = "rw"; }
+			{ node = "/dev/nvidia-uvm-tools";  modifier = "rw"; }
+			{ node = "/dev/nvidia-modeset";    modifier = "rw"; }
+		];
+
+		bindMounts = {
+			# NVENC/NVDEC - video transcoding
+			"/dev/nvidia0"           = { hostPath = "/dev/nvidia0";           isReadOnly = false; };
+			"/dev/nvidiactl"         = { hostPath = "/dev/nvidiactl";         isReadOnly = false; };
+			# CUDA - required for ML inference
+			"/dev/nvidia-uvm"        = { hostPath = "/dev/nvidia-uvm";        isReadOnly = false; };
+			"/dev/nvidia-uvm-tools"  = { hostPath = "/dev/nvidia-uvm-tools";  isReadOnly = false; };
+			# Modeset - needed by some driver paths
+			"/dev/nvidia-modeset"    = { hostPath = "/dev/nvidia-modeset";    isReadOnly = false; };
+		};
 
 		config = { lib, pkgs, config, ... }: {
 			# Network
 			networking.interfaces.eth0.useDHCP = true;
 			networking.firewall.allowedTCPPorts = [ 80 ];  # Caddy
 
+			# Configure nixpkgs for nvidia/cuda
+			nixpkgs.config.allowUnfree = true;
+			nixpkgs.config.cudaSupport = true;
+
 			# Immich
-			services.immich.enable = true;
-			environment.systemPackages = with pkgs; [ immich immich-cli ];
+			services.immich = {
+				enable = true;
+				accelerationDevices = [
+					"/dev/nvidia0"
+					"/dev/nvidiactl"
+					"/dev/nvidia-uvm"
+					"/dev/nvidia-uvm-tools"
+					"/dev/nvidia-modeset"
+				];
+
+				# Tell immich-server to use NVENC for transcoding
+				environment = {
+					NVIDIA_VISIBLE_DEVICES = "all";
+					NVIDIA_DRIVER_CAPABILITIES = "compute,video,utility";
+				};
+
+				# Enable the ML microservice with CUDA
+				machine-learning.enable = true;
+			};
+			#environment.systemPackages = with pkgs; [ immich immich-cli ];
 
 			# Reverse proxy
 			services.caddy = {
@@ -47,6 +74,23 @@
 				'';
 			};
 
+			# NVIDIA
+			services.xserver.videoDrivers = [ "nvidia" ];  # xserver.videoDrivers does not imply X
+			hardware.graphics.enable = true;
+			hardware.nvidia = {
+				modesetting.enable = true;  # Required
+				nvidiaSettings = false;  # Don't need gui
+				open = false;
+				package = config.boot.kernelPackages.nvidiaPackages.legacy_580;  # Must match host
+			};
+
+			# Packages
+			environment.systemPackages = with pkgs; [
+				immich
+				immich-cli
+				config.hardware.nvidia.package  # nvidia-smi
+			];
+
 			system.stateVersion = "25.11";
 		};
 	};