How to Use Iluvatar GPU in Applications¶
This section describes how to use Iluvatar virtual GPU on DCE 5.0.
Prerequisites¶
- Deployed DCE 5.0 container management platform and it is running smoothly.
- The container management module has been integrated with a Kubernetes cluster or a Kubernetes cluster has been created, and the UI interface of the cluster can be accessed.
- The Iluvatar GPU driver has been installed on the current cluster. Refer to the Iluvatar official documentation for driver installation instructions, or contact the DaoCloud ecosystem team for enterprise-level support at peg-pem@daocloud.io.
- The GPU cards in the current cluster have not undergone any virtualization operations and not been occupied by other applications.
Procedure¶
Configuration via User Interface¶
-
Check if the GPU card in the cluster has been detected. Click Clusters -> Cluster Settings -> Addon Plugins , and check if the corresponding GPU type has been automatically enabled and detected. Currently, the cluster will automatically enable GPU and set the GPU type as Iluvatar .
-
Deploy a workload. Click Clusters -> Workloads and deploy a workload using the image. After selecting the type as (Iluvatar) , configure the GPU resources used by the application:
-
Physical Card Count (iluvatar.ai/vcuda-core): Indicates the number of physical cards that the current pod needs to mount. The input value must be an integer and less than or equal to the number of cards on the host machine.
-
Memory Usage (iluvatar.ai/vcuda-memory): Indicates the amount of GPU memory occupied by each card. The value is in MB, with a minimum value of 1 and a maximum value equal to the entire memory of the card.
If there are any issues with the configuration values, scheduling failures or resource allocation failures may occur.
-
Configuration via YAML¶
To request GPU resources for a workload, add the iluvatar.ai/vcuda-core: 1 and iluvatar.ai/vcuda-memory: 200 to the requests and limits. These parameters configure the application to use the physical card resources.
apiVersion: apps/v1
kind: Deployment
metadata:
name: full-iluvatar-gpu-demo
namespace: default
spec:
replicas: 1
selector:
matchLabels:
app: full-iluvatar-gpu-demo
template:
metadata:
labels:
app: full-iluvatar-gpu-demo
spec:
containers:
- image: nginx:perl
name: container-0
resources:
limits:
cpu: 250m
iluvatar.ai/vcuda-core: '1'
iluvatar.ai/vcuda-memory: '200'
memory: 512Mi
requests:
cpu: 250m
memory: 512Mi
imagePullSecrets:
- name: default-secret