Docs

Volcano Engine Function as a Service

Serverless GPU

Serverless GPU overview

Overview

The integration of cloud-native technologies with AI is accelerating the deployment of AI applications. Traditional AI computing tasks require a fixed allocation of designated GPU resources, but uneven task distribution may leave some GPUs constantly idle, incurring high costs for unused resources. To address this, veFaaS offers the Serverless GPU service, which extends the concept of serverless computing to provide efficient and cost-effective support for the deployment of AI applications.

By providing on-demand, scalable GPU instances, veFaaS effectively resolves the issues of low resource utilization and high costs associated with long-running GPU instances. You do not need to plan GPU resource needs in advance or worry about GPUs remaining idle for extended periods due to uneven resource utilization. You pay only for the GPU resources you actually use.

tip

[Beta - apply for trial] Serverless GPU is in the beta phase. If you need to apply for this feature, submit a ticket or contact your account manager.

Application scenarios

veFaaS offers two options: CPU instances and GPU instances. Traditional functions use CPU instances, which support basic computing scenarios. GPU functions use GPU instances and excel at executing highly threaded parallel processing tasks (large-scale computing tasks). They are suitable for scenarios such as AI model inference, AI model training, accelerated audio and video production, and graphics and image acceleration.

Procedure

The basic workflow for using GPU functions is shown in the following diagram:

GPU function creation

Based on a GPU instance, deploy the function via a container image.

Function release

Release the function and its configuration online.

Event trigger

Configure triggers to initiate function execution through HTTP(S) invocations or other event triggers.

Monitoring and logging

Leverage comprehensive monitoring metrics and log observability to monitor function status.

Limits

Each Volcano Engine root user has a limited GPU instance quota. For details, see Limits.

NVIDIA A10 and NVIDIA GPU types are available. The supported GPU types vary slightly by region. See the console for the actual supported types.

Each GPU instance supports only one GPU.

GPU instances and CPU instances are independent of each other, and after a function is created, you cannot switch it between CPU and GPU instance types.

The image size limit for a GPU instance is 10 GiB.

Last updated: 2026.08.19 14:06:06