Dear all,
I have been trying (and then succeeding) to compile Amber with CUDA support using cuda-12.8 and gcc-14. Initially, the compilation failed, so I tried to figure out what the problems were. Ultimately,
I found more than I thought.
First, in many cases, the Ada architecture was "skipped" in the configuration files. Thus, if one has an RTX4090, like in my case, the code would be generated with SM8.6, not SM8.9.
Additionally, in the cpptraj config files some CUDA options (for SM7.0 and SM7.5 had typos), and in many other cases support for some of the later versions of CUDA was missing.
Also, cuda-12.8 requires gcc-14, which needed the addition of one more option in the cmake config files where appropriate.
Finally, I also fixed the compatibility with the current version of numpy for parmed.
Please find attached the patch files that fix these problems.
Cheers,
Charo
--
Prof. Charo I. del Genio
Institute of Smart Agriculture for Safe and Functional Foods and Supplements, Trakia University, Stara Zagora, Bulgaria
Institute of Interdisciplinary Intelligent Science, Ningbo University of Technology, Ningbo, China
School of Mathematics, North University of China, Taiyuan, China
Editor in Chief of Chaos, Solitons & Fractals
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Oct 30 2025 - 03:30:04 PDT
