GPU calculation for inhomogeneous systems (Mixed precision)

Hello Axel and everyone,

I’m simulating a nanofluid which is an inhomogeneous system, I’m looking in to go for GPU calculations and I would like to know the applicability of mixed precision gpu calculations to my inhomogeneous system. I have seen in some previous threads that for larger and inhomogeneous systems, the single precision is not advisable. What about mixed precision? Would it be OK to use it for an inhomogeneous system? It would be great if you can give me some advice on the issue.

Thank you very much.


The only person that is able to give an authoritative answer to your question is you yourself. Different properties are impacted differently: e.g. stress/pressure much more than forces. so for NVT simulations the choice of precision can have less impact than for NPT.

To do that, set up a representative (i.e. large enough to capture what you are looking for but small enough to run fast) test example and run simulations with each setting and analyse them.
Then you can for which settings the results are within acceptable error margins from the full double precision results, and you get to see the performance impact as well.