Open source cuda projects

Every project in the registry tagged cuda, ranked by real GitHub adoption.

projects 8 combined stars ★ 106K refresh nightly
01 vllm ★ 92K

A high-throughput and memory-efficient inference and serving engine for LLMs

last push3 hours ago languagePython licenseApache-2.0
02 gpustack ★ 5.7K

A GPU cluster manager for high-performance AI model serving (vLLM, SGLang) and on-demand SSH-accessible GPU instances.

last push16 hours ago languagePython licenseApache-2.0
03 Beam ★ 1.8K

Serverless GPU compute with sub-second cold starts

last push27 hours ago languageGo licenseAGPL-3.0
04 InferenceX ★ 1.7K

Open Source Continuous Inference Benchmark Research Platform — Kimi K3 2.8T, MiniMax M3, DeepSeekv4, GLM5 - GB200 NVL72 vs MI355X vs B200 vs GB300 NVL72 & soon™

last push5 hours ago languagePython licenseApache-2.0
05 gpu-hot ★ 1.6K

🔥 Real-time NVIDIA GPU dashboard

last push24 days ago languageJavaScript licenseMIT
06 sglang-omni ★ 1.2K

SGLang-Omni is a high-performance serving framework for audio models (TTS, ASR) and unified multimodal models.

last push3 hours ago languagePython licenseApache-2.0
07 ServerlessLLM ★ 715

Serverless LLM Serving for Everyone.

last push14 days ago languagePython licenseApache-2.0
08 pegainfer ★ 704

Pure Rust + CUDA LLM inference engine — no PyTorch, OpenAI-compatible, serves Qwen3 to Kimi-K2

last push3 hours ago languageRust licenseApache-2.0

Related tags

← all tags

Frequently asked questions

How many open source cuda projects are there?

This registry tracks 8 projects tagged cuda, with 105,504 GitHub stars between them. The most-adopted is vllm at 92,028 stars.

Are these cuda projects free to use?

Yes — 8 of the 8 carry an explicit open-source licence across 3 distinct licences, so there is no licence fee. Where a project also sells a hosted or enterprise version, the self-hosted path remains free.

Which cuda project should I choose?

The list above is ranked by GitHub stars, but stars measure attention rather than fit. Check three things on each card: the licence (permissive versus copyleft), the language it is written in, and the last-push date — a high-star project that has not been pushed in a year is a liability.

Are these cuda projects still maintained?

8 of the 8 were pushed in the last 90 days, and every card shows its exact last-push date so you can see the rest. Sort your shortlist by that date before committing to a migration.