MPI and cuFFT

Accelerated Computing GPU-Accelerated Libraries
1

As far as I can read from the cuFFT manual,

Distributed-memory FFTW with MPI is not supported yet. Are there any plans to support it in the future
with full parallelization in all three directions?

2

this thread may be of interest:

[url]https://devtalk.nvidia.com/default/topic/1023097/gpu-accelerated-libraries/trouble-with-cufft-on-multiple-gpus/[/url]

3

Thanks for the info but that thread does not help much because developers there are referring to computational fluid dynamics packages that we do not want use such dependencies. I was wondering if nvidia will provide such a thing in cuFFT at some point? I found this online GitHub - amirgholami/accfft: A Massively Parallel FFT Library for CPU/GPU but supports up to partitioning in 2D not three which is what we are interested for.

4

Generally speaking, NVIDIA doesn’t communicate future plans in these forums, nor respond to questions about future plans. If you’d like to see a specific feature in future CUDA libraries, the recommendation is to file a bug at developer.nvidia.com

[url]https://devtalk.nvidia.com/default/topic/1044668/cuda-programming-and-performance/-how-to-report-a-bug/[/url]