Run and train AI on GPUs: top picks for 2026

An end-to-end open source platform for building and deploying machine learning models, with Python and C++ APIs.

A node-graph GUI, API and backend for running diffusion models locally to generate images, video, audio and 3D content.

LLM inference engine in plain C/C++ for running large language and vision-language models locally or in the cloud across many hardware backends.

A fast, memory-efficient library and server for running and serving large language models with an OpenAI-compatible API.

A cross-platform, GPU-accelerated OpenGL terminal emulator that relies on other tools like tmux for tabs and splits.

A deep learning optimization library for making distributed training and inference of large models easier and more efficient.

A C++ library with Python bindings for efficient similarity search and clustering of dense vectors, with optional GPU acceleration.

Python library that runs large language models layer by layer so 70B-class models can infer on a single 4GB GPU, without quantization.

SGLang is an open-source inference framework for serving large language, vision-language, and diffusion models on many GPU, TPU, and CPU platforms.

A minimal deep learning framework with a PyTorch-like tensor API, autograd, a compiler, and support for many accelerator backends.

A simulation platform for robotics and embodied AI combining a multi-physics engine, a photo-realistic renderer, and a cross-platform compiler behind a Python interface.

Local inference engine that runs the large Qwen3.8-Flash-Next model on consumer NVIDIA or AMD GPUs under Windows or Linux, with OpenAI and Anthropic compatible APIs.

Official implementation of FlashAttention, a memory-efficient exact attention kernel for training and inference of Transformer models on GPUs.

GPU-optimized library and reference training scripts for training large transformer models with multi-dimensional parallelism.

GPU.js transpiles simple JavaScript functions into GPU shader code so they run on the GPU in browsers and Node.js, falling back to plain JavaScript.

An edge-native MoE serving engine that runs large open-weight models like DeepSeek-V4 and Qwen3.6 on consumer GPUs and CPUs.

A KV cache management layer that stores, shares and reuses LLM attention caches across serving engines to cut time-to-first-token.

An open-source book and code collection of modern CUDA learning notes, with 200+ kernels covering HGEMM, FP8/FP4 GEMM, and Flash-Attention.

An open-source system to run, manage, and scale AI workloads across Kubernetes, Slurm, and 20+ clouds or GPU providers with one interface.

CUDA C++ template library and Python DSL for writing high-performance GEMM and related linear algebra kernels on NVIDIA GPUs.

A GPU communication library for expert-parallel MoE dispatch and combine, with FP8 dispatch and NVLink/RDMA support for training and inference.

Desktop app that stabilizes video using gyroscope data recorded by cameras and external sensors.

An open-source offline inference framework from DeepSeek for running DeepSeek-V4.1-Flash on NVIDIA SM100 and Huawei Ascend 950 hardware.

Windows tool for viewing and editing NVIDIA driver profile settings, including many hidden options not exposed in the NVIDIA Control Panel.

An interactive terminal monitor for NVIDIA GPUs that shows device and per-process status, with a CLI device selector and Python APIs.

A CLI that detects your GPU, CPU and RAM and ranks the HuggingFace local LLMs that actually run on your hardware, by benchmarks rather than parameter count.

PyTorch's on-device inference stack for running exported models on phones, laptops, browsers, and microcontrollers.

A Docker image that runs a headless Steam client with a desktop, letting you stream games to a browser, Moonlight, or Steam Link.

A modular web UI for AI image, video and audio generation, built on Stable Diffusion, Flux, Wan and other models, with a ComfyUI workflow tab.

A cross-platform C++ image and video filter library that runs on OpenGL/ES, with beauty and face-based effects.

A GPU-over-IP bridge that lets CPU-only machines use CUDA GPUs installed on remote servers over the network.

An open-source control plane that provisions and orchestrates GPU compute for training and inference across GPU clouds, Kubernetes, VMs, and on-prem clusters.

Open-source Go and Python runtime for serverless AI workloads, covering GPU inference endpoints, sandboxes, and background tasks.

A GPU-rendered terminal emulator in Rust whose background server keeps shells and panes alive after the window closes, with agent-aware sessions and SSH remote work.

Run ffmpeg on a remote GPU machine from any client over one TCP port, without GPU passthrough or shared filesystems.

A week-by-week course for building an AI voice call center with FastRTC, Superlinked, Twilio, Opik and Runpod.

A training framework for LLMs, VLMs, diffusion and embodied models on NVIDIA GPUs and Kunlun XPUs, built on Megatron-LM and torch-native backends.

Serving setup for Mirai S 27B (a Qwen3.8-27B GGUF) on a 12 GB RTX 4070, with a full 262k-token window and q8_0 K/V cache.

Type a text prompt and get a rigged 3D character performing that motion, using fal video models, GVHMR pose recovery and Mixamo retargeting.