Alexandre Strube

> Just buy 2x A100 LMAO My nodes have 4 and it's CLEARLY not enough for this thing.

> @surak What GPU's did you use and what size? I'm going to look at trying to run it on 8 x 32GB V100 or 4 x 64GB Xilinx VU9p....

> `jax[cuda12_pip]` installs the correct CUDA sources in your Python `site_packages`. > However, if you have other CUDA installations on your system, and your system is set up to load...

This seems to be fixed, I don't have this problem of raspistil at all.

The issue is that the SO file is 32 bit and one gets ```/usr/bin/ld: skipping incompatible /usr/lib/gcc/aarch64-linux-gnu/10/../../../../lib/libarducam_vcm.so when searching for -larducam_vcm``` when compiling it on a 64-bit system. ```bash file...

I think that you should periodically call Autofocus.py, but that's my impression only..

The default `python` packages are deprecated, as they related to python2. You should use `python3-opencv`

> @trholding and it will work with one GPU? No way. Not this model, even highly quantized. Unless it's a GH200 Data Center edition, which does have 96gb of VRAM...

It would be nice then to have a 0.8.13, which is the last known version.

I can confirm that 512gb ram and 4*A100 40gb is not enough for it.