Add more instances and node groups to cover all templates by default

examples/aws/eks-private/README.md

@@ -34,6 +34,17 @@ terraform apply
 If you are using a `tfvars` file, you will need to update the above commands accordingly.
 Note the output from Terraform which includes an example cloud registration command you will use below.
 
+#### Using Additional GPU Instance Types
+
+The default configuration includes only T4. To use additional GPU types (A10G, L4, etc.), use the provided example file which defines additional GPU configurations:
+
+```shell
+terraform plan -var-file="gpu_instances.tfvars.example"
+terraform apply -var-file="gpu_instances.tfvars.example"
+```
+
+Node groups will be created automatically for each GPU type defined in `gpu_instance_types`.
+
 ### Install the Kubernetes Requirements
 
 The Anyscale Operator requires the following components:

examples/aws/eks-private/eks.tf

@@ -27,16 +27,7 @@ locals {
   )
 
   # Map of GPU types to their product names and instance types
-  gpu_types = {
-    "T4" = {
-      product_name   = "Tesla-T4"
-      instance_types = ["g4dn.4xlarge"]
-    }
-    "A10G" = {
-      product_name   = "NVIDIA-A10G"
-      instance_types = ["g5.4xlarge"]
-    }
-  }
+  gpu_types = var.gpu_instance_types
 
   # Base configuration for GPU node groups
   gpu_node_group_base = {
@@ -76,9 +67,9 @@ locals {
     }
   ])
 
-  # Create a map of GPU node groups based on node_group_gpu_types
+  # Create a map of GPU node groups based on gpu_instance_types
   gpu_node_groups = {
-    for gpu_type in var.node_group_gpu_types : gpu_type => {
+    for gpu_type in keys(var.gpu_instance_types) : gpu_type => {
       ondemand = merge(
         local.gpu_node_group_base,
         {
@@ -222,12 +213,12 @@ module "eks" {
         iam_role_additional_policies = local.anyscale_iam
       }
     },
-    # Merge in GPU node groups based on node_group_gpu_types
+    # Merge in GPU node groups based on gpu_instance_types
     {
-      for gpu_type in var.node_group_gpu_types : "ondemand_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].ondemand
+      for gpu_type in keys(var.gpu_instance_types) : "ondemand_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].ondemand
     },
     {
-      for gpu_type in var.node_group_gpu_types : "spot_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].spot
+      for gpu_type in keys(var.gpu_instance_types) : "spot_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].spot
     }
   )
 

examples/aws/eks-private/gpu_instances.tfvars.example

@@ -0,0 +1,31 @@
+# GPU Instance Types for EKS
+#
+# This file contains additional GPU instance configurations beyond the default (T4).
+# Use this file directly or copy and modify as needed:
+#
+#   terraform plan -var-file="gpu_instances.tfvars.example"
+#   terraform apply -var-file="gpu_instances.tfvars.example"
+
+# GPU types configuration - node groups will be created for each entry
+gpu_instance_types = {
+  "T4" = {
+    product_name   = "Tesla-T4"
+    instance_types = ["g4dn.xlarge", "g4dn.2xlarge", "g4dn.4xlarge"]
+  }
+  "T4-4x" = {
+    product_name   = "Tesla-T4"
+    instance_types = ["g4dn.12xlarge"]
+  }
+  "A10G" = {
+    product_name   = "NVIDIA-A10G"
+    instance_types = ["g5.4xlarge"]
+  }
+  "L4" = {
+    product_name   = "NVIDIA-L4"
+    instance_types = ["g6.2xlarge", "g6.4xlarge"]
+  }
+  "L4-4x" = {
+    product_name   = "NVIDIA-L4"
+    instance_types = ["g6.24xlarge"]
+  }
+}

examples/aws/eks-private/variables.tf

@@ -78,13 +78,35 @@ variable "eks_cluster_version" {
   default     = "1.32"
 }
 
-variable "node_group_gpu_types" {
+variable "gpu_instance_types" {
   description = <<-EOT
-    (Optional) The GPU types of the EKS nodes.
-    Possible values: ["T4", "A10G"]
+    (Optional) GPU types configuration for the EKS cluster.
+    See gpu_instances.tfvars.example for additional GPU types.
+
+    ex:
+    ```
+    gpu_instance_types = {
+      "T4" = {
+        product_name   = "Tesla-T4"
+        instance_types = ["g4dn.xlarge", "g4dn.2xlarge", "g4dn.4xlarge"]
+      }
+      "A10G" = {
+        product_name   = "NVIDIA-A10G"
+        instance_types = ["g5.4xlarge"]
+      }
+    }
+    ```
   EOT
-  type        = list(string)
-  default     = ["T4"]
+  type = map(object({
+    product_name   = string
+    instance_types = list(string)
+  }))
+  default = {
+    "T4" = {
+      product_name   = "Tesla-T4"
+      instance_types = ["g4dn.4xlarge"]
+    }
+  }
 }
 
 variable "enable_efs" {

examples/aws/eks-public/README.md

@@ -33,6 +33,17 @@ terraform apply
 If you are using a `tfvars` file, you will need to update the above commands accordingly.
 Note the output from Terraform which includes an example cloud registration command you will use below.
 
+#### Using Additional GPU Instance Types
+
+The default configuration includes only T4. To use additional GPU types (A10G, L4, etc.), use the provided example file which defines additional GPU configurations:
+
+```shell
+terraform plan -var-file="gpu_instances.tfvars.example"
+terraform apply -var-file="gpu_instances.tfvars.example"
+```
+
+Node groups will be created automatically for each GPU type defined in `gpu_instance_types`.
+
 ### Install the Kubernetes Requirements
 
 The Anyscale Operator requires the following components:

examples/aws/eks-public/eks.tf

@@ -27,16 +27,7 @@ locals {
   )
 
   # Map of GPU types to their product names and instance types
-  gpu_types = {
-    "T4" = {
-      product_name   = "Tesla-T4"
-      instance_types = ["g4dn.4xlarge"]
-    }
-    "A10G" = {
-      product_name   = "NVIDIA-A10G"
-      instance_types = ["g5.4xlarge"]
-    }
-  }
+  gpu_types = var.gpu_instance_types
 
   # Base configuration for GPU node groups
   gpu_node_group_base = {
@@ -76,9 +67,9 @@ locals {
     }
   ])
 
-  # Create a map of GPU node groups based on node_group_gpu_types
+  # Create a map of GPU node groups based on gpu_instance_types
   gpu_node_groups = {
-    for gpu_type in var.node_group_gpu_types : gpu_type => {
+    for gpu_type in keys(var.gpu_instance_types) : gpu_type => {
       ondemand = merge(
         local.gpu_node_group_base,
         {
@@ -222,12 +213,12 @@ module "eks" {
         iam_role_additional_policies = local.anyscale_iam
       }
     },
-    # Merge in GPU node groups based on node_group_gpu_types
+    # Merge in GPU node groups based on gpu_instance_types
     {
-      for gpu_type in var.node_group_gpu_types : "ondemand_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].ondemand
+      for gpu_type in keys(var.gpu_instance_types) : "ondemand_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].ondemand
     },
     {
-      for gpu_type in var.node_group_gpu_types : "spot_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].spot
+      for gpu_type in keys(var.gpu_instance_types) : "spot_gpu_${lower(gpu_type)}" => local.gpu_node_groups[gpu_type].spot
     }
   )
 

examples/aws/eks-public/gpu_instances.tfvars.example

@@ -0,0 +1,31 @@
+# GPU Instance Types for EKS
+#
+# This file contains additional GPU instance configurations beyond the default (T4).
+# Use this file directly or copy and modify as needed:
+#
+#   terraform plan -var-file="gpu_instances.tfvars.example"
+#   terraform apply -var-file="gpu_instances.tfvars.example"
+
+# GPU types configuration - node groups will be created for each entry
+gpu_instance_types = {
+  "T4" = {
+    product_name   = "Tesla-T4"
+    instance_types = ["g4dn.xlarge", "g4dn.2xlarge", "g4dn.4xlarge"]
+  }
+  "T4-4x" = {
+    product_name   = "Tesla-T4"
+    instance_types = ["g4dn.12xlarge"]
+  }
+  "A10G" = {
+    product_name   = "NVIDIA-A10G"
+    instance_types = ["g5.4xlarge"]
+  }
+  "L4" = {
+    product_name   = "NVIDIA-L4"
+    instance_types = ["g6.2xlarge", "g6.4xlarge"]
+  }
+  "L4-4x" = {
+    product_name   = "NVIDIA-L4"
+    instance_types = ["g6.24xlarge"]
+  }
+}

examples/aws/eks-public/variables.tf

@@ -78,13 +78,35 @@ variable "eks_cluster_version" {
   default     = "1.32"
 }
 
-variable "node_group_gpu_types" {
+variable "gpu_instance_types" {
   description = <<-EOT
-    (Optional) The GPU types of the EKS nodes.
-    Possible values: ["T4", "A10G"]
+    (Optional) GPU types configuration for the EKS cluster.
+    See gpu_instances.tfvars.example for additional GPU types.
+
+    ex:
+    ```
+    gpu_instance_types = {
+      "T4" = {
+        product_name   = "Tesla-T4"
+        instance_types = ["g4dn.xlarge", "g4dn.2xlarge", "g4dn.4xlarge"]
+      }
+      "A10G" = {
+        product_name   = "NVIDIA-A10G"
+        instance_types = ["g5.4xlarge"]
+      }
+    }
+    ```
   EOT
-  type        = list(string)
-  default     = ["T4"]
+  type = map(object({
+    product_name   = string
+    instance_types = list(string)
+  }))
+  default = {
+    "T4" = {
+      product_name   = "Tesla-T4"
+      instance_types = ["g4dn.4xlarge"]
+    }
+  }
 }
 
 variable "enable_efs" {

examples/gcp/gke-new_cluster/README.md

@@ -47,6 +47,17 @@ Steps for deploying Anyscale resources via Terraform:
 If you are using a `tfvars` file, you will need to update the above commands accordingly.
 Note the output from Terraform which includes an example cloud registration command you will use below.
 
+#### Using Additional GPU Instance Types
+
+The default configuration includes only T4. To use additional GPU types (L4, A100, H100, etc.), use the provided example file which defines additional GPU configurations:
+
+```shell
+terraform plan -var-file="gpu_instances.tfvars.example"
+terraform apply -var-file="gpu_instances.tfvars.example"
+```
+
+Node pools will be created automatically for each GPU type defined in `gpu_instance_configs`.
+
 ### Install the Kubernetes Requirements
 
 The Anyscale Operator requires the following components: