GPU-accelerated version OpenMX 3.9.9 is now available |
- Date: 2025/09/09 12:20
- Name: Hiroyuki Kawai
<f21j501c@mail.cc.niigata-u.ac.jp>
- Dear OpenMX users,
I have recently developed and released GPU-accelerated code based on OpenMX 3.9.9.
Currently, only certain processes within band calculations (collinear and non-collinear) and cluster calculations (collinear and non-collinear) are GPU-accelerated, specifically matrix multiplication and eigenvalue problem processing.
I have confirmed that it achieves faster speeds than the CPU version for systems with several hundred atoms. I encourage you to try this code and assist us in finding any bugs.
It is available for download at the URL below.
Note: Please do not post questions about OpenMX 3.9.9 GPU on this OpenMX Forum. Instead, please ask me questions using the method described at the URL below.
https://github.com/dc1394/openmx3.9.9_gpu
Best regards,
Hiroyuki Kawai
| |
