disable nouveau and use smaller instance

skypilot-org · Oct 30, 2024 · 677c001 · 677c001
1 parent c474acf
commit 677c001
Show file tree

Hide file tree

Showing 4 changed files with 7 additions and 6 deletions.
diff --git a/sky/clouds/service_catalog/images/README.md b/sky/clouds/service_catalog/images/README.md
@@ -59,9 +59,10 @@ export SECRET=xxxxxx  # Update this
 ```
 2. Build and copy images for all regions for GPU (gen 1 & 2) and CPU (gen 2 only).
 ```bash
-export TYPE=gpu  # Update this
-export VM_GENERATION=1  # Update this
-packer build --var vm_generation=${VM_GENERATION} --var client_secret=${SECRET} skypilot-azure-${TYPE}-ubuntu.pkr.hcl
+packer build --var vm_generation=2 --var client_secret=${SECRET} skypilot-azure-cpu-ubuntu.pkr.hcl
+packer build --var vm_generation=2 --var client_secret=${SECRET} skypilot-azure-gpu-ubuntu.pkr.hcl
+packer build --var vm_generation=1 --var client_secret=${SECRET} skypilot-azure-gpu-ubuntu.pkr.hcl
+packer build  --var vm_generation=2 --var client_secret=${SECRET} --var use_grid_driver=true skypilot-azure-gpu-ubuntu.pkr.hcl
 ```
 
 ## Test Images

diff --git a/sky/clouds/service_catalog/images/provisioners/cuda-azure-grid.sh b/sky/clouds/service_catalog/images/provisioners/cuda-azure-grid.sh
@@ -9,7 +9,7 @@ GRID_DRIVER_FILE="NVIDIA-Linux-x86_64-535.161.08-grid-azure.run"
 
 wget -nv $GRID_DRIVER_URL -O $GRID_DRIVER_FILE
 sudo chmod +x $GRID_DRIVER_FILE
-sudo sh $GRID_DRIVER_FILE --silent
+sudo sh $GRID_DRIVER_FILE --silent --disable-nouveau
 
 
 echo "Installing CUDA toolkit..."

diff --git a/sky/clouds/service_catalog/images/skypilot-azure-gpu-ubuntu.pkr.hcl b/sky/clouds/service_catalog/images/skypilot-azure-gpu-ubuntu.pkr.hcl
@@ -33,7 +33,7 @@ source "azure-arm" "gpu-ubuntu" {
   image_offer     = "0001-com-ubuntu-server-jammy"
   image_sku       = var.vm_generation == 1 ? "22_04-lts" : "22_04-lts-gen2"
   location        = var.use_grid_driver || var.vm_generation == 1 ? "eastus" : "centralus"
-  vm_size         = var.use_grid_driver ? "Standard_NV36ads_A10_v5" : (var.vm_generation == 1 ? "Standard_NC4as_T4_v3" : "Standard_NC24ads_A100_v4")
+  vm_size         = var.use_grid_driver ? "Standard_NV6ads_A10_v5" : (var.vm_generation == 1 ? "Standard_NC4as_T4_v3" : "Standard_NC24ads_A100_v4")
   ssh_username    = "azureuser"
   azure_tags = {
     Created_by = "packer"

diff --git a/sky/provision/azure/instance.py b/sky/provision/azure/instance.py
@@ -311,7 +311,7 @@ def _create_vm(
         vm_name=vm_name,
         parameters=vm_instance,
     )
-    # poller.result() will block on async operation until it's done.
+    vm_poller.wait()
     logger.info(f'Created VM {vm_poller.result().name}.')
     return vm_poller.result()