These articles cover the tools used day-to-day when developing, debugging, and profiling CUDA programs. They assume familiarity with the fundamentals.
Compilers (WIP) — nvcc, PTX, SASS, JIT compilation, useful flags, and Compiler Explorer
Compilation Time (WIP) — why CUDA builds are slow and how to measure and fix it
CMake (WIP) — build system setup for mixed C++/CUDA projects
Compute Sanitizer (WIP) — finding memory errors, races, and leaks automatically
cuda-gdb (WIP) — breakpoints and stepping inside kernels
Nsight Systems (WIP) — the application-level timeline profiler (and NVTX annotations)
Nsight Compute (WIP) — the kernel-level deep-dive profiler (speed of light, roofline, source view)