Under the Hood: CUDA, Tensors, and Tensor Cores

The libraries that turn "it works" into "it works fast." cuDNN for training, TensorRT for inference, mixed precision for both. And how to enable each one in just a few lines of PyTorch.

Measure before you optimize. Read an Nsight timeline, spot the bottlenecks, and build a mini end-to-end pipeline with profiling, mixed precision, and TensorRT. The capstone that ties the whole course together.