Use runfile to install NVIDIA driver and update driver versions (#79)

ci/compute-image-name.sh

@@ -8,17 +8,19 @@ if [ -z "${DRIVER_VERSION}" ]; then
   VARIANT=cpu
 fi
 
+DRIVER_VERSION_SHORT="${DRIVER_VERSION%%.*}"
+
 IMAGE_NAME=$(
   jq -nr \
     --arg OS "${OS}" \
     --arg VARIANT "${VARIANT}" \
-    --arg DRIVER_VERSION "${DRIVER_VERSION}" \
+    --arg DRIVER_VERSION_SHORT "${DRIVER_VERSION_SHORT}" \
     --arg ARCH "${ARCH}" \
     --arg RUNNER_VERSION "${RUNNER_VERSION}" \
   '[
     $OS,
     $VARIANT,
-    $DRIVER_VERSION,
+    $DRIVER_VERSION_SHORT,
     $ARCH,
     $RUNNER_VERSION
   ] | map(select(length > 0)) | join("-")'

linux/installers/nvidia-driver.sh

@@ -7,17 +7,16 @@ if [ "${NV_VARIANT}" != "gpu" ]; then
   exit 0
 fi
 
-KEYRING=cuda-keyring_1.1-1_all.deb
 ARCH=x86_64
-
 if [ "${NV_ARCH}" == "arm64" ]; then
-  ARCH=sbsa
+  ARCH=aarch64
 fi
 
-wget -q "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/${ARCH}/${KEYRING}"
-sudo dpkg --install "${KEYRING}"
-sudo apt-get update
-
-sudo apt-get -y install "nvidia-driver-${NV_DRIVER_VERSION}"
+TMP_DIR=$(mktemp -d)
+RUNFILE_NAME="NVIDIA-Linux-${ARCH}-${NV_DRIVER_VERSION}.run"
+RUNFILE_URL="https://download.nvidia.com/XFree86/Linux-${ARCH}/${NV_DRIVER_VERSION}/${RUNFILE_NAME}"
+RUNFILE_PATH="${TMP_DIR}/${RUNFILE_NAME}"
 
-sudo dpkg --purge "$(dpkg -f "${KEYRING}" Package)"
+wget --no-verbose -O "${RUNFILE_PATH}" "${RUNFILE_URL}"
+sudo sh "${RUNFILE_PATH}" --no-questions --ui=none
+rm -rf "${TMP_DIR}"

matrix.yaml

@@ -12,9 +12,9 @@ OS:
 DRIVER_VERSION:
   # keep this blank entry. an empty driver version corresponds to CPU machines
   - ""
-  - "470"
-  - "550"
-  - "560"
+  - "535.183.01"
+  - "550.107.02"
+  - "560.35.03"
 
 RUNNER_VERSION:
   # renovate: repo=actions/runner
@@ -29,18 +29,16 @@ ENV:
   - qemu
 
 exclude:
-  - ARCH: arm64
-    DRIVER_VERSION: "470"
   # only use amd64 for windows
   - OS: windows
     ARCH: arm64
   # only make CPU images for windows
   - OS: windows
-    DRIVER_VERSION: "470"
+    DRIVER_VERSION: "535.183.01"
   - OS: windows
-    DRIVER_VERSION: "550"
+    DRIVER_VERSION: "550.107.02"
   - OS: windows
-    DRIVER_VERSION: "560"
+    DRIVER_VERSION: "560.35.03"
   # only make AMI images for windows
   - OS: windows
     ENV: qemu

variables.pkr.hcl

@@ -25,8 +25,8 @@ variable "driver_version" {
   description = "The NVIDIA driver version to install on the EC2 instance. If empty, no driver will be installed."
 
   validation {
-    condition     = can(regex("(^\\d{3}$|^$)", var.driver_version))
-    error_message = "The driver_version value must be an empty string or 3 digits."
+    condition     = can(regex("(^\\d{3}\\.\\d{2,3}\\.\\d{2,3}$|^$)", var.driver_version))
+    error_message = "The driver_version value must be an empty string or 3 groups of digits splitted by dots."
   }
 }