Help Center/ Elastic Cloud Server/ User Guide/ Instances/ Managing GPU Drivers of GPU-accelerated ECSs/ GPU Driver
Updated on 2026-08-10 GMT+08:00
View PDF
Share

GPU Driver

What Is a GPU Driver?

GPU-accelerated ECSs provide powerful floating-point compute performance to handle massive workloads with high real-time and high concurrency requirements. GPU-accelerated ECSs do not come with drivers. You need to install the required drivers before using them to enable graphics acceleration or compute acceleration.

For details about GPU-accelerated ECSs, see GPU-accelerated ECSs.

GPU-accelerated ECSs support GRID and Tesla drivers.

Table 1 GPU driver description

Driver

Scenario

Acceleration Capability

Required Installation

Description

GRID

Graphics acceleration:

3D rendering, graphics workstations, and game acceleration

CUDA

OpenGL

DirectX

Vulkan

GRID driver + GRID license
  • The GRID driver provides basic CUDA compute capabilities. You do not need to install the CUDA Toolkit.
  • To use graphics acceleration, such as OpenGL, DirectX, or Vulkan, install a GRID driver and separately purchase and configure a GRID license.
  • If both graphics acceleration and compute acceleration are required, you can install the GRID driver and use the vDWS license to provide full CUDA capabilities.

Tesla

Compute acceleration:

Scientific computing, deep learning training, and inference

CUDA

Tesla driver + CUDA
  • Generally, the CUDA Toolkit needs to be installed to enable compute acceleration.
  • The CUDA Toolkit can be downloaded for free.

Constraints

  • An ECS can install either the GRID driver or the Tesla driver, but not both. Different drivers cause kernel conflicts and mutual service exclusion. Installing both may cause issues such as black screens or GPU identification failures.
  • GPU-accelerated ECSs support GPU drivers. Table 2 lists the relationships among drivers, GPUs, and ECS series.
    Table 2 ECS series supported by GPU drivers

    Driver

    Applicable GPU Type

    Corresponding GPU-accelerated ECS Series

    GRID

    T4

    G6v, G6, and Pi2

    V100

    G5, P2s, P2vs, and P2v

    M60

    G3 and G1

    Tesla

    T4

    G6 and Pi2

    V100

    G5, P2s, P2vs, and P2v

    P4

    Pi1

    P100

    P1

GPU Driver Installation Methods

GPU driver installation depends on the instance type, image type, and operating node.

  • When creating a GPU-accelerated ECS from a public image, install the GPU driver as described below.
    Select a public image that contains a specific GPU driver.
    • If the GRID driver is contained, you need to purchase and configure the corresponding GRID license after the installation.
    • If the Tesla driver is contained, the CUDA Toolkit corresponding to the Tesla driver is also included. No further operation is required.
    Figure 1 Selecting a public image that contains a specific GPU driver
  • When creating a GPU-accelerated ECS from a private image, install the GPU driver after the ECS is created. For details, see Install After ECS Creation (Linux and Windows).
Table 3 Installation methods

Driver

Installation Method

Instruction

Scenario

GRID

Manual

Manually Installing a GRID Driver on a GPU-accelerated ECS
  • Linux and Windows GPU-accelerated ECSs
  • GPU-accelerated ECSs that do not support automatic installation

Tesla

Manual

Manually Installing a Tesla Driver on a GPU-accelerated ECS
  • Linux and Windows GPU-accelerated ECSs
  • GPU-accelerated ECSs that do not support automatic installation
Parent Topic: Managing GPU Drivers of GPU-accelerated ECSs