switchless-ring

Here are 3 public repositories matching this topic...

yunwei37 / dgx-spark-4-ring-no-switch

Reproducible LLM inference on four NVIDIA DGX Spark (GB10) nodes wired as a direct switchless ConnectX ring — serving profiles, ring launchers, pinned ARM64 runtime images and measured results, newest tested model first.

arm64 llm-serving vllm sglang distributed-inference gb10 dgx-spark tensor-parallel switchless-ring nccl-mesh

Updated Oct 2, 2026
Python

ZackO2o / DeepSeek-V4.1-Flash-4x-GB10-1M-Full-Recipe

Star

Measured serving recipes for DeepSeek-V4.1-Flash on 4x NVIDIA DGX Spark (GB10): 1M context on vLLM (CUDA graphs, vision, tools, DSpark) and a switchless-ring SGLang TP4 lane, plus a cross-project reference table. EN + 中文.

vision roce llm-serving vllm speculative-decoding deepseek sglang tool-calling gb10 dgx-spark deepseek-v4 cuda-graphs sm121 1m-context tensor-parallel switchless-ring

Updated Sep 28, 2026
Python

ondigo-winder / SWITCHLESS-NCCL-RING-4-GB10

Star

NCCL over a switchless 4-node DGX Spark / GB10 ring using both PCIe halves of every QSFP cable: ~193 Gb/s per cable instead of ~112, up to +34% vLLM prefill. One patch on top of switchless-nccl.

Updated Oct 2, 2026
Cuda

Add this topic to your repo

To associate your repository with the switchless-ring topic, visit your repo's landing page and select "manage topics."

Learn more