Elastic Accelerated Computing Instances FAQ

更新时间:
复制 MD 格式

This topic answers common questions about Elastic Accelerated Computing Instances (EAIS).

What is Elastic Accelerated Computing Instances (EAIS)?

Elastic Accelerated Computing Instances (EAIS) is a high-performance, cost-effective, and scalable elastic computing service from Alibaba Cloud. As an Infrastructure as a Service (IaaS) offering, EAIS decouples CPU resources from GPU resources. You can attach GPU resources to ECS instances to create the GPU instance types that you need. You can use these instances for scalable AI inference, which improves resource utilization and reduces costs.

For more information, see What is EAIS?.

How is EAIS different from regular GPU-accelerated instances?

In a regular GPU-accelerated instance, the CPU, memory, and GPU are on the same physical server. With EAIS, the CPU and memory can be on a different physical machine from the GPU. This decouples the resources.

Does EAIS have high latency due to network issues?

We ran multiple tests using models such as Resnet50, Inception4, and Bidirectional Encoder Representations from Transformers (BERT). The results show negligible latency compared to regular GPU-accelerated instances.

Why choose EAIS?

Compared to regular heterogeneous instances, EAIS provides benefits such as resource decoupling, low costs, high scalability, and broad compatibility. For more information, see Benefits.

What are the main application scenarios for EAIS?

EAIS supports all inference scenarios. For more information, see Scenarios.

How is EAIS billed?

EAIS is billed on a pay-as-you-go basis. For more information, see Pay-as-you-go.

In which regions is EAIS available?

EAIS is currently available only in China. Supported regions include Beijing, Shanghai, Hangzhou, Shenzhen, and Chengdu. For more information, see Regions and zones.

How do I use EAIS?

An EAIS instance functions as an accelerator that you must attach to an ECS instance. To use the EAIS instance for AI inference, remotely log on to the attached ECS instance. For more information, see Use an EAIS instance.

Are there any prerequisites for using EAIS?

EAIS provides only GPU resources. You must have an ECS instance for the corresponding CPU and memory.