add al2keplergpu build recipe to build gpu amis for kepler arch

Makefile

@@ -57,6 +57,10 @@ al2arm: check-region init validate release.auto.pkrvars.hcl
 al2gpu: check-region init validate release.auto.pkrvars.hcl
 	./packer build -only="amazon-ebs.al2gpu" -var "region=${REGION}" .
 
+.PHONY: al2keplergpu
+al2keplergpu: check-region init validate release.auto.pkrvars.hcl
+	./packer build -only="amazon-ebs.al2keplergpu" -var "region=${REGION}" .
+
 .PHONY: al2inf
 al2inf: check-region init validate release.auto.pkrvars.hcl
 	./packer build -only="amazon-ebs.al2inf" -var "region=${REGION}" .

README.md

@@ -6,12 +6,15 @@ It will create a private AMI in whatever account you are running it in.
 ## Instructions
 
 1. Setup AWS cli credentials.
-2. Make the recipe that you want, REGION must be specified. Options are: al1, al2, al2arm, al2gpu, al2inf, 
+2. Make the recipe that you want, REGION must be specified. Options are: al1, al2, al2arm, al2gpu, al2keplergpu, al2inf, 
 al2kernel5dot10, al2kernel5dot10arm, al2023, al2023arm, al2023neu.
 ```
 REGION=us-west-2 make al2
 ```
 
+**NOTE**: `al2keplergpu` is a build recipe that this package supports to build ECS-Optimized GPU AMIs for instances with GPUs
+with Kepler architecture (such as P2 type instances). ECS-Optimized GPU AMIs for this target are not officially built and published.
+
 ## Configuration
 
 This recipe allows for configuration of your AMI. All configuration variables are defined and documented

al2.pkr.hcl

@@ -37,6 +37,7 @@ build {
     "source.amazon-ebs.al2",
     "source.amazon-ebs.al2arm",
     "source.amazon-ebs.al2gpu",
+    "source.amazon-ebs.al2keplergpu",
     "source.amazon-ebs.al2inf",
     "source.amazon-ebs.al2kernel5dot10",
     "source.amazon-ebs.al2kernel5dot10arm"

al2keplergpu.pkr.hcl

@@ -0,0 +1,33 @@
+locals {
+  ami_name_al2keplergpu = "${var.ami_name_prefix_al2}-kepler-gpu-hvm-2.0.${var.ami_version}-x86_64-ebs"
+}
+
+source "amazon-ebs" "al2keplergpu" {
+  ami_name        = "${local.ami_name_al2keplergpu}"
+  ami_description = "Amazon Linux AMI 2.0.${var.ami_version} x86_64 ECS HVM GP2"
+  instance_type   = var.gpu_instance_types[0]
+  launch_block_device_mappings {
+    volume_size           = var.block_device_size_gb
+    delete_on_termination = true
+    volume_type           = "gp2"
+    device_name           = "/dev/xvda"
+  }
+  region = var.region
+  source_ami_filter {
+    filters = {
+      name = "${var.source_ami_al2}"
+    }
+    owners      = ["amazon"]
+    most_recent = true
+  }
+  ssh_interface = "public_ip"
+  ssh_username  = "ec2-user"
+  tags = {
+    os_version          = "Amazon Linux 2"
+    source_image_name   = "{{ .SourceAMIName }}"
+    ecs_runtime_version = "Docker version ${var.docker_version}"
+    ecs_agent_version   = "${var.ecs_agent_version}"
+    ami_type            = "al2keplergpu"
+    ami_version         = "2.0.${var.ami_version}"
+  }
+}

scripts/enable-ecs-agent-gpu-support.sh

@@ -1,7 +1,7 @@
 #!/usr/bin/env bash
 set -ex
 
-if [[ $AMI_TYPE != "al2gpu" ]]; then
+if [[ $AMI_TYPE != "al2gpu" && $AMI_TYPE != "al2keplergpu" ]]; then
     exit 0
 fi
 
@@ -28,21 +28,41 @@ sudo mv $tmpfile /etc/yum.repos.d/amzn2-nvidia-tmp.repo
 sudo yum install -y system-release-nvidia
 sudo rm /etc/yum.repos.d/amzn2-nvidia-tmp.repo
 
-sudo yum install -y kernel-devel-$(uname -r) \
-    system-release-nvidia \
-    nvidia-driver-latest-dkms \
-    nvidia-fabric-manager \
-    pciutils \
-    xorg-x11-server-Xorg \
-    docker-runtime-nvidia \
-    oci-add-hooks \
-    libnvidia-container \
-    libnvidia-container-tools \
-    nvidia-container-runtime-hook
+# for building AMIs for GPUs with Kepler architecture, fix package versions
+# also exclude nvidia and cuda packages to update. Newer Nvidia drivers do not support Kepler architecture
+# TODO: The package versions are fixed for Kepler. They have to be manually updated when there is a minor version update in AL repo.
+if [[ $AMI_TYPE == "al2keplergpu" ]]; then
+    sudo yum install -y kernel-devel-$(uname -r) \
+        system-release-nvidia \
+        nvidia-driver-latest-dkms-470.182.03 \
+        nvidia-fabric-manager-470.182.03-1 \
+        pciutils-3.5.1-2.amzn2 \
+        xorg-x11-server-Xorg \
+        docker-runtime-nvidia-1 \
+        oci-add-hooks \
+        libnvidia-container-1.4.0 \
+        libnvidia-container-tools-1.4.0 \
+        nvidia-container-runtime-hook-1.4.0
 
-sudo yum install -y cuda-drivers \
-    cuda
+    sudo yum install -y cuda-toolkit-11-4
+    echo "exclude=*nvidia* *cuda*" | sudo tee -a /etc/yum.conf
+else
+    # Default GPU AMI
+    sudo yum install -y kernel-devel-$(uname -r) \
+        system-release-nvidia \
+        nvidia-driver-latest-dkms \
+        nvidia-fabric-manager \
+        pciutils \
+        xorg-x11-server-Xorg \
+        docker-runtime-nvidia \
+        oci-add-hooks \
+        libnvidia-container \
+        libnvidia-container-tools \
+        nvidia-container-runtime-hook
 
+    sudo yum install -y cuda-drivers \
+        cuda
+fi
 # The Fabric Manager service needs to be started and enabled on EC2 P4d instances
 # in order to configure NVLinks and NVSwitches
 sudo systemctl enable nvidia-fabricmanager