> If you can process "offline" for an hour and then cache the results, CPU inference is fine. > GPUs are expensive. Depends on the GPU. I've found T4 GPUs to

> If you can process "offline" for an hour and then cache the results, CPU inference is fine. > GPUs are expensive. Depends on the GPU. I've found T4 GPUs to be cheaper than CPU compute on AWS when testing throughput per $ of spend.