Trying to get the GPU to cooperate

2 files changed, 15 insertions, 0 deletions

diff --git a/hosts/desktop/configuration.nix b/hosts/desktop/configuration.nix
index 96cc210..810fa8b 100644
--- a/hosts/desktop/configuration.nix
+++ b/hosts/desktop/configuration.nix
@@ -39,6 +39,16 @@ in
     package = unstable.ollama;
     acceleration = "rocm";
     rocmOverrideGfx = "10.3.0";
+    environmentVariables = {
+      # This forces ROCm to see ONLY the RX 6800.
+      # On most Ryzen 7000 systems, '0' is the dGPU and '1' is the iGPU.
+      HIP_VISIBLE_DEVICES = "0";
+      ROCR_VISIBLE_DEVICES = "0";
+
+      # Keep these for RDNA 2 stability with 26B models
+      HSA_OVERRIDE_GFX_VERSION = "10.3.0";
+      HSA_ENABLE_SDMA = "0";
+    };
   };
 
   environment.variables.HSA_OVERRIDE_GFX_VERSION = "10.3.0";
diff --git a/hosts/desktop/hardware-configuration.nix b/hosts/desktop/hardware-configuration.nix
index ba7f356..cc0c460 100644
--- a/hosts/desktop/hardware-configuration.nix
+++ b/hosts/desktop/hardware-configuration.nix
@@ -12,6 +12,11 @@
   boot.initrd.kernelModules = [ "amdgpu" ];
   boot.kernelModules = [ "kvm-amd" ];
   boot.extraModulePackages = [ ];
+  boot.kernelParams = [
+    "amdgpu.noretry=0"   # Allows the GPU to retry memory access instead of crashing
+    "amdgpu.gpureset=1"  # Enables better recovery if a fault occurs
+    "amdgpu.pcie_gen_cap=0x00040000" # Forces PCIe 4.0 if your board is being weird
+  ];
 
   hardware.graphics = {
     enable = true;