RAVE requires a CUDA GPU with 5–32 GB VRAM depending on config, and hours of audio

RAVE training is GPU-bound; training without a hardware accelerator will take ages and is not recommended. GPU memory requirements by config: raspberry: 5 GB; onnx: 6 GB; v1/v2_small: 8 GB; v2/discrete/spectral_discriminator: 16–18 GB; v3: 32 GB. A minimum of a few hours of audio is required (the training-setup doc advises at least 3h, more for complex sounds). Training durations are long, running for days on a single GPU. Google Colab provides a free path for those without a local GPU (a community notebook is linked from the README).

Examples

For local training: conda create -n rave python=3.9 && conda activate rave && pip install acids-rave. For no local GPU, use the linked Colab notebook.

Assessment

A student with only a laptop CPU wants to train RAVE on 30 minutes of field recordings. Name two practical barriers they will encounter and two alternative paths forward.