flox-cuda

The flox-cuda skill provides a comprehensive framework for CUDA and GPU development using Flox, streamlining the setup and management of NVIDIA CUDA environments. It addresses the complexity of installing and configuring CUDA toolkits, compilers, libraries, and runtime components across supported Linux platforms, ensuring developers can efficiently harness GPU computing capabilities. By centralizing package discovery, version management, and installation procedures, this skill reduces errors and simplifies the deployment of GPU-accelerated applications.

The skill includes core functionalities such as searching for CUDA packages, displaying available versions, installing essential CUDA components like nvcc, cudatoolkit, cuDNN, and various GPU libraries (cuBLAS, cuFFT, cuRAND, NCCL), and verifying installations via standard NVIDIA commands. It also handles conflict resolution for LICENSE file issues by assigning package priorities and supports multiple CUDA versions, including current 12.x releases and legacy 11.x versions. Additionally, flox-cuda provides cross-platform considerations by offering CPU fallbacks for macOS environments, allowing developers to maintain portability between Linux and Darwin systems.

This skill is particularly valuable for developers working on deep learning, high-performance computing, and GPU-accelerated simulations. It targets software engineers, data scientists, and researchers who require precise control over CUDA installations and dependencies. Typical use cases include setting up GPU environments for training neural networks, performing large-scale linear algebra computations, and ensuring reproducibility of CUDA-based applications across multiple Linux platforms. By offering structured package management and clear installation guidance, flox-cuda minimizes setup complexity and enhances productivity for GPU-focused development projects.