gpu-utils-lite (v0.1)

Note: CUDA build/run requires a Linux/Windows machine with an NVIDIA GPU + CUDA Toolkit. macOS is not supported as a CUDA target environment.

Minimal CUDA micro-bench utilities:

• cuda_kernel_benchmark: simple kernel timing -> CSV
• memcpy_benchmark: HtoD/DtoH bandwidth sweep -> CSV
• latency_profiler: tiny latency (kernel launch + tiny memcpy)

Build

mkdir -p build
cmake -S . -B build -DCMAKE_BUILD_TYPE=Release
cmake --build build -j

Binaries: build/bin/

Run (from repo root)

./build/bin/cuda_kernel_benchmark --out results/kernel_benchmark.csv
./build/bin/memcpy_benchmark --out results/memcpy_benchmark.csv
./build/bin/latency_profiler

Notes

If build is slow, set your CUDA arch: cmake -S . -B build -DCMAKE_BUILD_TYPE=Release -DCMAKE_CUDA_ARCHITECTURES=86