This topic describes how to select an instance type and provides suggestions on how to select instance types in common scenarios.

You can select instance families recommended in the following figure based on the scenarios to which your service is applicable. For more instance families, see Instance families.
Note The available instance families and types vary with the region. You can go to the ECS Instance Types Available for Each Region page to view the instance types available in each region.
Instance type selection flowchart
Note For more information about scenarios, see Selection of enterprise-grade instance types.
Select instance types based on the requirements of your scenario. You can refer to the following suggestions:
  • Balanced performance

    A balanced CPU-to-memory ratio is required to meet the requirements for application resources in most scenarios.

  • Applications with high packet forwarding rates

    High packet forwarding rates are required. You can select a reasonable computing-to-memory ratio based on the scenario.

  • High-performance computing

    Massive computing resources are required. Typical requirements include GPU parallel computing and high clock speed.

  • High-performance client games

    A high clock speed processor is required to support more users.

  • Mobile or web games

    Massive computing resources are required. We recommend that you select a CPU-to-memory ratio of 1:2 to achieve the optimal cost-effectiveness of computing resources.

  • Video forwarding

    Massive computing resources are required. We recommend that you select a CPU-to-memory ratio of 1:2 to achieve the optimal cost-effectiveness of computing resources.

  • Live bullet screens

    High packet forwarding rates are required. You can select a reasonable computing-to-memory ratio based on the scenario.

  • Relational databases

    Standard SSDs or higher-performance local NVMe SSDs are required to achieve high storage IOPS and low read/write latency. We recommend that you select a balanced (1:4) or lower (1:8) CPU-to-memory ratio.

  • Distributed caches

    Stable computing performance is required. We recommend that you select a balanced (1:4) or lower (1:8) CPU-to-memory ratio.

  • NoSQL databases

    Standard SSDs or higher-performance local NVMe SSDs are required to achieve high storage IOPS and low read/write latency. We recommend that you select a balanced (1:4) or lower (1:8) CPU-to-memory ratio.

  • Elasticsearch

    Standard SSDs or higher-performance local NVMe SSDs are required to achieve high storage IOPS and low read/write latency. We recommend that you select a balanced (1:4) or lower (1:8) CPU-to-memory ratio.

  • Hadoop

    Data nodes require a high disk throughput, high network throughput, and balanced CPU-to-memory ratio. Computing nodes rely more on the computing performance, network bandwidth, and CPU-to-memory ratio.

  • Image transcoding

    Hardware parallel acceleration is required. You can select a reasonable computing-to-memory ratio based on the scenario.

  • AI
    • Deep learning training

      High-performance NVIDIA GPU accelerators are required. Select a GPU to CPU ratio of 1:8 to 1:12.

    • General-purpose deep learning

      High-performance NVIDIA GPU accelerators are required. Select a GPU to CPU ratio of 1:4 to 1:48.

    • Image recognition and reasoning

      High-performance NVIDIA GPU accelerators are required. Select a GPU to CPU ratio of 1:4 to 1:12.

    • Speech recognition and speech synthesis

      High-performance NVIDIA GPU accelerators are required. Select a GPU to CPU ratio of 1:16 to 1:48.

  • Super computing

    Powerful and stable computing capabilities and an excellent network with high bandwidth and low latency are required.