Hello,
I've already posted on PixInsight forums, but CN seems to be better tailored for StarNet support.
Hello,
I've search for somebody else noticing this issue, but I couldn't find anything.
I have both PixInsight installations with both StarNet2 modules on Linux.
The "old" PI+StarNet is 3-4 times faster than the "new" one.
I've looked with "nvtop" on the GPU usage and the pattern is clear. The GPU is barely used and takes a long time to process on the "new" PI+StarNet. The GPU usage is much higher on the "old" one and it finishes much faster (blue lines on GPU0 - NVIDIA are the GPU usage).
Both installations have libtensorflow* removed, to use the one from the operating system, CUDA enabled (obviously, the GPU being used).
Anyone noticed this and/or have any idea?
LE: Forgot to mention the topic: StarNet 2.1.1 on 1.8.9-2 significantly slower than StarNet 2.1.0 on 1.8.9-1
Edited by sigxcpu, 25 October 2023 - 12:21 PM.