Cuda project architecture examples
Interactive architecture diagrams of 43 open-source Cuda repositories with a GitDiagram architecture diagram, most stars first. Each diagram shows the project's main components, how they connect, and links every part to its file on GitHub.
Repositories 1–43 of 43
karpathy/llm.c★ 31.4k
llm.c is a compact LLM pretraining project, centered on GPT-2/GPT-3-family training in C/CUDA, with CPU and PyTorch reference implementations.
nvlabs/instant-ngp★ 17.3k
Neural Network Engine (based on tiny-cuda-nn)
deepseek-ai/deepep★ 10.4k
DeepEP is a GPU communication library for distributed ML, centered on MoE expert-parallel dispatch and combine.
deepseek-ai/deepgemm★ 6.3k
API & Interface Layer: Although minimal, the public API exposed via Python calls (e.g., gemm functions and utility functions in deep_gemm/__init__.py) forms the entry point.
flashinfer-ai/flashinfer★ 5.2k
This project – FlashInfer – is a high-performance kernel library aimed at LLM serving and inference.
infatoshi/cuda-course★ 3.4k
Course Modules: Each clearly numbered folder (e.g., “01_Deep_Learning_Ecosystem”, “02_Setup”, “03_C_and_C++_Review”, “04_Gentle_Intro_to_GPUs”, “05_Writing_your_First_Kernels”…
hazyresearch/thunderkittens★ 3.3k
The ThunderKittens project is a high-performance, CUDA-based framework focused on writing fast deep learning kernels by leveraging tile primitives and fine control of GPU…
computerhistory/alexnet-source-code★ 3k
This repository is the original 2012 AlexNet source code: a configurable GPU convolutional-neural-network system with ImageNet parameters.
mirage-project/mirage★ 2.5k
Mirage Persistent Kernel (MPK) compiles multi-GPU LLM inference graphs into a persistent fused CUDA “megakernel,” including device computation and communication.
nvidia/cudalibrarysamples★ 2.4k
This project is an extensive repository of CUDA library samples that demonstrates a wide array of GPU-accelerated libraries for different computational domains.
siboehm/sgemm_cuda★ 1.3k
k2-fsa/k2★ 1.3k
woct0rdho/sageattention★ 1.1k
nvidia/nvbench★ 839
nvidia/cuopt★ 788
brucefan1983/gpumd★ 742
clu0/unet.cu★ 666
pranjalssh/fast.cu★ 561
xuezhemax/megalodon★ 528
alicevision/popsift★ 491
mw00/project-maya★ 484
moorethreads/litegs★ 417
mit-han-lab/quest★ 410
acdslab/mppi-generic★ 263
dots-infra/ultraep★ 251
flamegpu/flamegpu2★ 145
microsoft/tilefusion★ 106
rocm/rccl-tests★ 88
rox906/tcfft★ 43
yigongqin/cupf★ 33
openhero/im2col★ 30
ntnu-hpc-lab/bat★ 19
Counted over GitDiagram's stored diagrams of public repositories, from the names their components and groups carry and the files they link to; the rules are the ones in the 2026 open-source architecture report. Stars are GitHub's when the diagram was last made. Rebuilt daily: hubs grow as more repositories' languages are known.