Installing AMGX on Your Computer

2 minute read

Published:

This guide provides a step-by-step approach to installing AMGX on your system. The installation has been tested on:

  • Ubuntu 20.04.5 LTS
  • CentOS Linux 7 (Core)

Ensure you have an NVIDIA driver installed and can run nvidia-smi to check details about the driver and GPU. For example:

nvidia-smi

Output:

Sun Jun 19 16:52:32 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 515.76       Driver Version: 515.76       CUDA Version: 11.7     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA RTX A5000    Off  | 00000000:C1:00.0 Off |                  Off |
| 30%   56C    P0    72W / 230W |      0MiB / 24564MiB |      2%      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+

Choosing the Correct CUDA Toolkit Version

Check the CUDA version supported by your driver (shown in the nvidia-smi output). Install a CUDA Toolkit version equal to or lower than the one mentioned—in this case, 11.7. Although NVIDIA supports forward compatibility for minor versions (reference), issues may arise with unsupported versions. For instance, I encountered problems using CUDA 11.8 despite forward compatibility claims.

To avoid issues, select a CUDA Toolkit version lower than or equal to the supported version.

Installing the CUDA Toolkit

Visit the CUDA Toolkit download page for the latest version. For older versions, search for “CUDA Toolkit " (e.g., "CUDA Toolkit 11.7") in your browser. You can then select the appropriate version for your operating system, architecture, distribution, and version—for example:

CUDA 11.0 Archive for CentOS 7

Installation with the Run File

I used the run file for installation. If you have sudo access, the installation is straightforward. However, in most workstation environments where sudo is unavailable (as in my case), you can:

  1. Run the file without sudo.
  2. Select only the toolkit for installation.
  3. Use advanced options to change the installation path—I set mine to a directory in my home folder.

Make sure that you add installation path in your .bashrc:

export CUDA_HOME=/path/to/user/cuda
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH

Verifying the Installation

Once the installation is complete, run one of the examples provided with the CUDA Toolkit to ensure everything works correctly.

Installing AMGX

To install AMGX, I compiled and built version 2.1.x from the 2.1.x branch of the AMGX GitHub repository.

Ensure you have GCC version 9.4 installed, as other versions may cause errors during the installation process.

After completing the installation, verify it by running the examples mentioned in the README file of the repository. This ensures the installation was successful and that the library is functioning as expected.

Leave a Comment