Upgrading PixInsight with GPU-accelerated TensorFlow v2.6.0 for StarNet++(RTX 30 Series)

I tested the following versions of the required software packages and confirm that StarNet++ becomes GPU-accelerated when running on a NVIDIA RTX3090 GPU in PixInsight for Windows 10. For reference, I used the excellent instructions at Dark Skies, replacing the older referenced versions with the following:

TensorFlow, CUDA, and CuDNN must be matched based on compatibility. I haven’t found an updated matrix that goes up to TF v2.7 and couldn’t find an easy download link for the v2.7 dll, so I went with v2.6. Here is the reference I used to determine this specific compatibility combination.

• PixInsight Core 1.8.8-9 Ripley (x64) (2021-09-02)
• TensorFlow 2.6.0
• NVIDIA CUDA 11.2.2
• NVIDIA cuDNN v8.1.1

StarNet++ now maxes out the GPU and is many multiples faster than when using the CPU even though I’m running an 18-core Intel Core i9 7980XE overclocked to all-core 3.8Ghz (from 2.6Ghz).