Updated on 2024-09-30 GMT+08:00

Heterogeneous Resources

GPU Settings

GPU virtualization

CCE uses the proprietary xGPU virtualization technologies to dynamically divide the GPU memory and compute. A single GPU can be virtualized into up to 20 virtual GPUs. Virtualization is more flexible than static allocation. You can specify the number of GPUs on the basis of stable services to improve GPU utilization. For details, see Overview.

GPU Driver

  • Default Cluster Driver: specifies the default GPU driver version used by the GPU nodes in a cluster. To use a custom driver, enter the download link of the NVIDIA driver. For details, see Obtaining the Driver Link from Public Network.
  • Custom Node Pool Driver: If you do not want all GPU nodes in a cluster to use the same driver, CCE allows you to install a different GPU driver for each node pool. After you customize a GPU driver for a node pool, nodes in the node pool will preferentially use the custom driver. Nodes for which no driver is specified will use the cluster's default driver.
    • The system installs the driver of the version specified for the node pool. The driver applies only to new nodes in the node pool.
    • After the driver version is updated, it takes effect on the nodes newly added to the node pool. Existing nodes must restart to apply the changes.
    • When installing the CCE AI Suite (NVIDIA GPU) add-on of v2.7.2 or later, you can configure xGPU virtualization for node pools.

NPU Settings

If driver selection is disabled, you cannot specify driver versions or maintain drivers using the CCE AI Suite (Ascend NPU) add-on. When you add an NPU node on the console, the system adds the command to install an NPU driver (version and type decided by the system) and automatically restarts the node after the driver installation is complete. Adding an NPU node in another way, such as using an API, requires you to add the driver installation command to the post-installation command.

Once driver selection is enabled, CCE AI Suite (Ascend NPU) will automatically install the driver during its startup based on the driver configuration of the corresponding model. This makes driver maintenance more flexible. You are advised to use the default driver version. You can also enter the complete driver address to use custom drivers.

The following table lists what NPUs and OS specifications are supported.

NPU Type

Supported OS

D310

EulerOS 2.5 x86, CentOS 7.6 x86, EulerOS 2.9 x86, and EulerOS 2.8 arm