Skip to content

CHANGELOG.md

Latest commit

 

History

History
79 lines (60 loc) · 2.64 KB

CHANGELOG.md

File metadata and controls

79 lines (60 loc) · 2.64 KB
Raw

Changelog

All notable changes to gpuci will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

0.1.0 - 2025-01-25

License

  • PolyForm Noncommercial 1.0.0 - Free for non-commercial use
  • Commercial use requires separate license from RightNow AI

Added

Core Features

  • Multi-GPU Testing: Run CUDA kernels across multiple GPU targets simultaneously
  • CUDA Event Timing: Microsecond-precision kernel timing using cudaEventRecord/cudaEventElapsedTime
  • Parallel Execution: ThreadPoolExecutor-based parallel dispatch to all targets
  • Rich Output: Beautiful terminal output with Rich tables

GPU Providers (6 total)

  • SSH: Direct connection to your own GPU machines
  • RunPod: On-demand cloud GPUs via RunPod SDK
  • Lambda Labs: High-performance cloud GPUs via REST API
  • Vast.ai: GPU marketplace via vastai-sdk
  • FluidStack: Enterprise GPUs via REST API
  • Brev: NVIDIA Brev cloud GPUs via CLI

CLI Commands

  • gpuci init: Interactive configuration wizard
  • gpuci test <kernel.cu>: Run kernel tests across targets
  • gpuci targets: List configured GPU targets
  • gpuci check: Verify connectivity to all targets

Configuration

  • YAML-based configuration (gpuci.yml)
  • Automatic config file discovery (searches parent directories)
  • Customizable warmup runs, benchmark runs, and timeout
  • Per-target nvcc flags support

Kernel Support

  • Full kernels: Kernels with main() using GPUCI macros
  • Kernel-only: Auto-generated test harness for __global__ functions
  • Automatic GPU architecture detection via nvidia-smi

GitHub Actions Integration

  • Reusable GitHub Action (action.yml)
  • PR comment support with benchmark results
  • Multiple workflow examples included

Documentation

  • Comprehensive README with examples
  • Detailed provider setup guide (docs/providers.md)
  • GitHub Actions integration guide (docs/github-actions.md)

Verified

  • Tested on NVIDIA GeForce RTX 5070 Ti (Vast.ai)
  • All example kernels pass: vector_add.cu, matmul.cu, simple_kernel.cu

Future Releases

Planned for v0.2.0

  • JSON output format for programmatic use
  • Retry logic for transient cloud provider failures
  • Cost estimation before running tests
  • Historical benchmark tracking

Planned for v0.3.0

  • Web dashboard for results visualization
  • Slack/Discord notifications
  • Custom metrics beyond timing
  • Multi-kernel test files