Скачать или смотреть How to Change the MKL Version for NumPy in Conda

Learn how to easily change the `MKL` version used by NumPy in Conda environments, especially for better performance on AMD processors.
---
This video is based on the question https://stackoverflow.com/q/68965495/ asked by the user 'AstroTeen' ( https://stackoverflow.com/u/14392308/ ) and on the answer https://stackoverflow.com/a/68991641/ provided by the user 'merv' ( https://stackoverflow.com/u/570918/ ) at 'Stack Overflow' website. Thanks to these great users and Stackexchange community for their contributions.

Visit these links for original content and any more details, such as alternate solutions, latest updates/developments on topic, comments, revision history etc. For example, the original title of the Question was: How to change NumPy and conda MKL version

Also, Content (except music) licensed under CC BY-SA https://meta.stackexchange.com/help/l...
The original Question post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license, and the original Answer post is licensed under the 'CC BY-SA 4.0' ( https://creativecommons.org/licenses/... ) license.

If anything seems off to you, please feel free to write me at vlogize [AT] gmail [DOT] com.
---
How to Change the MKL Version for NumPy in Conda

Intel's Math Kernel Library (MKL) is widely used for its optimized performance in scientific computing. However, users with AMD processors have reported issues where MKL does not perform to its full potential, resulting in significantly slower computations. If you're using NumPy and Miniconda, you might have the latest MKL installed, but unfortunately, it doesn't yield the desired performance on non-Intel CPUs. In this guide, we’ll walk through how to downgrade your MKL version from 2021 to 2019, significantly improving your numerical calculations on AMD processors.

The Problem with the Latest MKL

The latest versions of MKL have been designed to work optimally with Intel processors. This can lead to performance degradation on AMD CPUs because:

The MKL is optimized specifically for Intel architecture.

Recent changes removed certain environment variables (like MKL_DEBUG_CPY_TYPE=5) that previously offered a workaround for performance issues on AMD processors.

As a result, users searching for better numerical performance may find themselves stuck with an inefficient setup.

Given these challenges, it's essential to find a way to revert to an older version of MKL that is friendlier to non-Intel CPUs, specifically version 2019.

Solution: Downgrading MKL Version

To achieve this, the recommended approach is to create a new conda environment and specify the version of MKL alongside NumPy. Here's how you can do it step by step:

Step 1: Create a New Conda Environment

Open your terminal or command prompt and run the following command to create a new environment named foo and install NumPy as well as the desired MKL version.

[[See Video to Reveal this Text or Code Snippet]]

Step 2: Activate the Environment

Once the environment is created, activate it using:

[[See Video to Reveal this Text or Code Snippet]]

Step 3: Verify the Installation

After you've activated your new environment, start a Python session by simply typing python. Now you can check the version of MKL that NumPy is using:

[[See Video to Reveal this Text or Code Snippet]]

Step 4: Configuration Check

To ensure everything is configured correctly, use the following command to view detailed configuration information about the installed libraries:

[[See Video to Reveal this Text or Code Snippet]]

This output will help you verify that the libraries are pointing to the correct MKL version and paths.

Important Notes

Environment Management: Create new environments for experiments to avoid conflicts with existing setups.

Differences Between Channels: There can be variations in how Anaconda and Conda Forge manage NumPy and MKL installations. Make sure to use the right channels to get the best results.

Python Compatibility: As of now, the Anaconda channel might only support older Python versions (e.g., Python 3.8) with MKL 2019. Ensure that you are using a version of Python that fits your project requirements.

By following these steps, you should be able to successfully downgrade your MKL to a version that better meets the needs of your AMD processor, enhancing your scientific computational performance.

Conclusion

In conclusion, dealing with the performance limitations of MKL on AMD systems can be frustrating but manageable. By reverting to an older version of MKL, you can regain much-needed performance in your scientific calculations. Happy coding!