Tag
Discussion of the shift in GPU kernel engineering from C++ CuTe/CUTLASS to NVIDIA's Python-based CuTeDSL, questioning whether new engineers should learn legacy C++ templates or prioritize the emerging stack for LLM inference work.