hpc-gpu-stack

Automate CUDA toolchain setup, GPU-aware MPI, and scheduler integration for HPC workflows.

71|9|Updated Mar 16, 2026
One-click install
npx skills add https://github.com/SciMate-AI/HPC-Skills --skill hpc-gpu-stack
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: hpc-gpu-stack
Source: https://github.com/SciMate-AI/HPC-Skills/tree/main/skills/hpc-gpu-stack
Command: npx skills add https://github.com/SciMate-AI/HPC-Skills --skill hpc-gpu-stack

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) and assets (resource) components.

What problem does it solve?

This Skill addresses the complexities of building, reviewing, debugging, and launching CUDA- and GPU-accelerated HPC workflows, ensuring compatibility and performance optimization.

Core Features & Use Cases

  • CUDA Toolchain Management: Configure and manage CUDA toolchain and host compiler compatibility.
  • GPU-Aware MPI: Support for CUDA-aware MPI and rank-to-GPU mapping.
  • Scheduler Integration: Integrate with Slurm and other schedulers for GPU allocation and visibility.
  • Memory and Stream Management: Handle GPU memory and stream behavior for performance optimization.
  • Build and Launch Workflow: Provide reproducible build and launch sequences for CUDA applications.
  • Runtime Debugging and Profiling: Debug and profile CUDA applications for performance issues.
  • Error Recovery: Provide guidance for common errors and recovery strategies.
  • Use Case: A researcher is working on a GPU-accelerated simulation and encounters performance issues. The Skill can help configure the CUDA toolchain, optimize memory usage, and debug the simulation.

Quick Start

Build and launch a CUDA application with GPU acceleration using the hpc-gpu-stack skill.

Frequently Asked Questions about hpc-gpu-stack

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I configure the CUDA toolchain and host compiler for HPC workflows?

To configure the CUDA toolchain for HPC workflows, you need to manage compiler compatibility and set up the build environment. This ensures reproducible build and launch sequences for CUDA applications.

What is GPU-aware MPI and how does rank-to-GPU mapping work?

GPU-aware MPI enables direct memory transfers between GPUs across nodes, while rank-to-GPU mapping assigns specific MPI ranks to designated GPUs. This optimizes communication and memory management in HPC environments.

Can I integrate GPU allocation with Slurm schedulers?

Yes, you can integrate GPU allocation with Slurm schedulers. This integration manages GPU visibility and resource distribution, ensuring that CUDA-accelerated HPC workflows receive the required GPU resources.

Why does my GPU-accelerated simulation have performance issues and memory errors?

GPU-accelerated simulations experience performance issues and memory errors due to suboptimal stream behavior or memory management. Runtime debugging and profiling can identify these bottlenecks and guide error recovery strategies.

What's the best way to debug and profile CUDA applications?

The best way to debug and profile CUDA applications is by using runtime profiling to analyze GPU memory and stream behavior. This identifies performance bottlenecks and provides error recovery guidance for HPC workflows.