19:19 dcunit3d: is it possible to use an AMD gpu for compute on linux via docker if the GPU is being used for graphics?
19:20 dcunit3d: i'm running rocm/tensorflow image. my rocminfo output in the container shows two devices: 5950x (no integrated GPU....) and 6700xt
19:20 dcunit3d: but nothing shows with `tf.config.list_physical_devices('GPU')`
19:21 dcunit3d: i'm getting a warning about "This TF binary is optimized with oneAPI DNN to use CPU instructions: SSE3, AVX, etc"
19:22 dcunit3d: but so far, i've only run into issues with that if trying to compute with the wrong types (i guess on the GPU)
19:35 Ristovski: dcunit3d: afaik it also requires "--device=/dev/kfd"
19:35 Ristovski: see: https://github.com/RadeonOpenCompute/ROCm-docker
19:38 dcunit3d: k, i've got that on my docker script. i'll take a look and also try running some tf_cnn_benchmarks instead of running with `jupyter lab --allow-root`.
19:40 dcunit3d: i'm kinda new to python & tf, so i'm not sure how to find the gap between what the system commands run on docker are showing and what TF is showing
19:41 dcunit3d: clinfo is showing the device though