Implement GPU collective communication using a custom PyTorch c10d backend built on libibverbs and softRoCE for distributed model training.
-
Updated
May 21, 2026 - C++
#
nccl
Here are 67 public repositories matching this topic...
AI cluster debugging lab for distributed LLM and HPC workloads: GPU, NCCL, Kubernetes, failure analysis, and tuning recommendations.
kubernetes performance-engineering hpc nvidia roce observability distributed-training nccl gpu-cluster failure-analysis llm dcgm
-
Updated
May 20, 2026 - Python
NYDUX — AI Infrastructure Intelligence | GPU Cluster Diagnostics | ML Systems Mastery
machine-learning hpc distributed-training nccl gpu-cluster llm-training ai-infrastructure mfu-optimization
-
Updated
May 16, 2026 - HTML
Safe rust wrapper around CUDA toolkit
rust gpu cuda cublas gpu-acceleration cuda-kernels cudnn cuda-toolkit nccl curand cuda-programming nvrtc
-
Updated
May 15, 2026 - Rust
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.
parallel-computing cuda high-performance-computing cuda-kernels quantization cutlass gemm performance-optimization nccl gpu-programming roofline-model tensor-core llm-inference flash-attention nsight-compute
-
Updated
May 11, 2026 - Cuda
Actor-shaped face for compute acceleration backends (NVIDIA CUDA shipping; ROCm / Metal / oneAPI / Vulkan future). Built on the atomr actor runtime.
python rust machine-learning actors akka gpu actor-model concurrency cuda cublas cudnn nccl nvrtc agentic-systems
-
Updated
May 10, 2026 - Rust
Practical AI homelab setup guides for GB10, Mac Studio Ultra, RoCE/RDMA, MikroTik switching, NCCL, and heterogeneous workload experiments.
mikrotik dell inference nvidia rdma roce distributed-training nccl kv-cache gb10 dgx-spark grace-blackwell connectx-7
-
Updated
May 10, 2026
Mini Distributed Training Framework using NCCL
-
Updated
May 9, 2026
Simple quick test to benchmark your pytorch + nccl/ncclx setup
-
Updated
May 3, 2026 - Python
From-scratch RDMA-based PyTorch backend in C++; trained a char-level GPT on TinyShakespeare via DDP through code I wrote.
-
Updated
May 1, 2026 - C++
Experimental Explicit Communications API for Kokkos
-
Updated
Apr 30, 2026 - C++
Fixes NCCL hangs on NVIDIA L40S GPUs by resolving IOMMU-induced PCIe P2P communication issues. Includes reproducible tests, architecture explanation, and production-ready solution.
machine-learning performance deep-learning hpc gpu linux-kernel cuda nvidia pcie distributed-training nccl iommu ai-infrastructure l40s
-
Updated
Apr 30, 2026 - Shell
A Reliable and Resilient Collective Communication Library for NCCL and others
-
Updated
Apr 29, 2026 - C++
Multi-threaded Layer-2 stress and hardware evaluation tool for GPU cluster fabric validation
-
Updated
Apr 26, 2026 - C
Real multi-node MPI benchmarks for AI infrastructure teams. By NYDUX — nydux.ai
-
Updated
Apr 22, 2026
A hybrid testbed for evaluating top open-source LLMs (like gpt-oss-20b and Llama 3.3) on local, cloud GPUs, and AWS Inferentia2/Trainium instances, focusing on vLLM optimization, capacity management, kernel bypass, hardware-software co-design, as well as supporting infrastructure such as NCCL, RDMA, NVMeoF.
aws gpu rdma nvme kernel-bypass nccl gpudirect nvlink nvmeof llm vllm trainium vllm-serve inferentia2 software-hardware-co-design aws-ofi-nccl
-
Updated
Apr 21, 2026 - Python
Improve this page
Add a description, image, and links to the nccl topic page so that developers can more easily learn about it.
Add this topic to your repo
To associate your repository with the nccl topic, visit your repo's landing page and select "manage topics."