Cuda Trending

llm.c

LLM training in simple, raw C/CUDA

30,1333,625+15

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,187204+8

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3391,023+8

instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more

17,4232,063+6

3d-reconstructioncomputer-graphicscomputer-visioncudafunction-approximation

cuopt

GPU accelerated decision optimization

934190+5

BBuf /

how-to-optim-algorithm-in-cuda

#15

how to optimize some algorithm in cuda.

nvbench

#14

CUDA Kernel Benchmarking Library

870109+2

benchmarkcudacuda-kernelsgpukernel-benchmark

cuvs

cuVS - a library for vector search and clustering on the GPU

777194+2

annsclusteringcudadistancegpu

ThunderKittens

Tile primitives for speedy kernels

3,411290+2

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,411427+1

attentioncudaefficient-attentioninference-accelerationllm

CUDALibrarySamples

CUDA Library Samples

2,425459+1

causal-conv1d

Causal depthwise conv1d in CUDA, with a PyTorch interface

896190+1

nccl-tests

NCCL Tests

1,542375+1

DeepEP

DeepEP: an efficient expert-parallel communication library

9,6961,279+1

GPUMD

#16

Graphics Processing Units Molecular Dynamics

782186

cudagpugpumdheat-transporthigh-performance-computing

cub

[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl

1,833462

algorithmscppcpp11cpp14cpp17

Thu, June 4, 2026

llm.c

#15

LLM training in simple, raw C/CUDA

30,1183,624+13

cuopt

GPU accelerated decision optimization

929188+11

ThunderKittens

Tile primitives for speedy kernels

3,409290+4

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,410426+3

attentioncudaefficient-attentioninference-accelerationllm

nccl-tests

NCCL Tests

1,541375+2

causal-conv1d

Causal depthwise conv1d in CUDA, with a PyTorch interface

895189+2

mirage-project /

mirage

#14

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

2,290214+1

cuvs

cuVS - a library for vector search and clustering on the GPU

775193+1

annsclusteringcudadistancegpu

CUDALibrarySamples

CUDA Library Samples

2,424459+1

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,179204+1

DeepEP

DeepEP: an efficient expert-parallel communication library

9,6951,278+1

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3311,021+1

nvbench

CUDA Kernel Benchmarking Library

868109

benchmarkcudacuda-kernelsgpukernel-benchmark

cub

[ARCHIVED] Cooperative primitives for CUDA C++. See https://github.com/NVIDIA/cccl

1,833462

algorithmscppcpp11cpp14cpp17

GPUMD

Graphics Processing Units Molecular Dynamics

782186

cudagpugpumdheat-transporthigh-performance-computing

Wed, June 3, 2026

llm.c

LLM training in simple, raw C/CUDA

30,1073,623+9

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,178204+8

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3301,019+6

cuopt

#14

GPU accelerated decision optimization

918184+5

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,407425+4

attentioncudaefficient-attentioninference-accelerationllm

DeepEP

DeepEP: an efficient expert-parallel communication library

9,6941,276+4

mirage-project /

mirage

#15

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

2,289214+2

GPUMD

Graphics Processing Units Molecular Dynamics

782186+2

cudagpugpumdheat-transporthigh-performance-computing

instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more

17,4142,065+1

3d-reconstructioncomputer-graphicscomputer-visioncudafunction-approximation

HenryHuYu /

DiffPhysDrone

Published on Nature Machine Intelligence! The first real robot(quadrotor) based on differentiable physics training.

55982+1

droneend-to-endreinforcement-learningrobotics

causal-conv1d

Causal depthwise conv1d in CUDA, with a PyTorch interface

893188+1

CUDALibrarySamples

CUDA Library Samples

2,423459+1

ThunderKittens

Tile primitives for speedy kernels

3,405290

cugraph

cuGraph - RAPIDS Graph Analytics Library

2,187357

complex-networkscudagpugraphgraph-algorithms

nccl-tests

NCCL Tests

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,170204+16

llm.c

LLM training in simple, raw C/CUDA

30,1033,622+13

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3251,014+6

ThunderKittens

Tile primitives for speedy kernels

3,405290+4

DeepEP

DeepEP: an efficient expert-parallel communication library

9,6931,273+4

CUDALibrarySamples

CUDA Library Samples

2,422458+3

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,403425+3

attentioncudaefficient-attentioninference-accelerationllm

instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more

17,4142,065+2

3d-reconstructioncomputer-graphicscomputer-visioncudafunction-approximation

causal-conv1d

Causal depthwise conv1d in CUDA, with a PyTorch interface

892188+2

cuvs

cuVS - a library for vector search and clustering on the GPU

772192+2

annsclusteringcudadistancegpu

nccl-tests

NCCL Tests

1,539375+2

cuopt

GPU accelerated decision optimization

915183+2

GPUMD

Graphics Processing Units Molecular Dynamics

780186+1

cudagpugpumdheat-transporthigh-performance-computing

Mon, June 1, 2026

llm.c

LLM training in simple, raw C/CUDA

30,0903,621+13

cuopt

GPU accelerated decision optimization

913183+8

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3201,014+6

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,154200+5

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,401425+4

attentioncudaefficient-attentioninference-accelerationllm

Sun, May 31, 2026

llm.c

LLM training in simple, raw C/CUDA

30,0813,618+7

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,149199+4

cuopt

GPU accelerated decision optimization

905181+2

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,397426+2

attentioncudaefficient-attentioninference-accelerationllm

psmarter /

CUDA-Practice

CUDA编程练习项目-Hands-on CUDA kernels and performance optimization, covering GEMM, FlashAttention, Tensor Cores, CUTLASS, quantization, KV cache, NCCL, and profiling.

15312+2

cudacuda-kernelscutlassflash-attentiongemm

DeepGEMM

DeepGEMM: clean and efficient FP8 GEMM kernels with fine-grained scaling

7,3171,011+2

ThunderKittens

Tile primitives for speedy kernels

3,399288+1

instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more

17,4112,064

3d-reconstructioncomputer-graphicscomputer-visioncudafunction-approximation

Sat, May 30, 2026

llm.c

LLM training in simple, raw C/CUDA

30,0743,616+12

rtp-llm

RTP-LLM: Alibaba's high-performance LLM inference engine for diverse applications.

1,145197+7

ThunderKittens

Tile primitives for speedy kernels

3,398288+4

instant-ngp

Instant neural graphics primitives: lightning fast NeRF and more

17,4122,064+3

3d-reconstructioncomputer-graphicscomputer-visioncudafunction-approximation

mirage-project /

mirage

Mirage Persistent Kernel: Compiling LLMs into a MegaKernel

2,284212+2

SageAttention

[ICLR2025, ICML2025, NeurIPS2025 Spotlight] Quantized Attention achieves speedup of 2-5x compared to FlashAttention, without losing end-to-end metrics across language, image, and video models.

3,395426+2

attentioncudaefficient-attentioninference-accelerationllm