Установка cuPyNumeric для GPU-ускоренных вычислений

★ 7.6 · content

cupynumeric-install is a Claude Code skill that guides installation and verification of cuPyNumeric — an NVIDIA GPU-accelerated drop-in replacement for NumPy built on Legate — via conda or pip. It covers system requirement checks (CUDA 12.2+, Compute Capability ≥ 7.0, Python 3.11–3.14, Linux x86_64/aarch64 or WSL), isolated environment creation, and both conda and pip install paths, including how to force the GPU variant with CONDA_OVERRIDE_CUDA in containerized setups. A mandatory smoke test runs through the legate launcher, and a separate GPU-usage check with nvidia-smi confirms actual hardware utilization rather than a silent CPU-variant fallback. Building from source is explicitly out of scope. Useful whenever someone needs cuPyNumeric running for accelerated NumPy-compatible workloads on a GPU machine or distributed cluster.