CPU vs GPU design, SM vs CU, warps vs wavefronts, tensor cores vs matrix cores, memory hierarchy, and a Nvidia↔AMD terminology glossary.
Kernel launch syntax, thread hierarchy, compiling with nvcc/hipcc, streams, and writing portable kernels with HIP.
BLAS (cuBLAS/rocBLAS/hipBLASLt), deep learning primitives (cuDNN/MIOpen), collectives (NCCL/RCCL), and kernel-generation frameworks (CUTLASS/Composable Kernel/Triton).
Occupancy, warp divergence, memory coalescing, shared memory bank conflicts, kernel fusion, mixed precision, and the roofline model.
Reading compiler output, GPU ISA (SASS/RDNA/CDNA), scalar vs vector registers, and the micro-architectural effects behind "why simple code is slow."
Mixed precision/AMP, torch.compile, the PyTorch profiler, the caching allocator, and multi-GPU training basics.
More on this hub soon — debugging GPU kernels, distributed/multi-node training, inference-serving optimization, and the assembly/ISA deep dives from the low-level track.