My system specs are:
Ubuntu 16.04 x86_64 OS
16 GB RAM
Compute capability = 2.1 Fermi.
Gracphics card = GF106 GeForce 450
CUDA 8.0 (8.0.61 GA2) (successfully installed)
Using NVIDIA Graphics Driver 384.130 that was installed by the CUDA 8.0 (8.0.61 GA2) Toolkit
Version of kernel system = 4.15.0-48-generic
My understanding is that the cuDNN v6.0 [April 27, 2017] is compatible with CUDA 8.0 (8.0.61 GA2).
After reading the “cuDNN Install Guide”, it appears that at least a 3.0 compute capability is required.
If this is correct, are there no options available for using CUDA 8.0 (8.0.61 GA2) with a cuDNN for a Fermi 2.1 compute capability?