Build, load, and verify NCCL for switchless DGX Spark RoCE cycles
-
Updated
Oct 25, 2026 - Shell
#
switchless
Here are 13 public repositories matching this topic...
Another Mega Drive switchless MOD
-
Updated
Jul 4, 2022 - HTML
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.
nvidia rdma roce nccl dgx blackwell switchless tensor-parallelism vllm llm-inference rocev2 gb10 blackwell-gpu nvfp4 dgx-spark dgxspark connectx-7 nvfp4-native dgx-sparks switchless-ring
-
Updated
Oct 2, 2026 - Cuda
GLM-5.3-Flash NVFP4 TP4 on 4x DGX Spark switchless ring
self-hosted nvidia moe roce glm tp4 ring-network switchless llm-serving vllm speculative-decoding gb10 nvfp4 dgx-spark
-
Updated
Sep 6, 2026
Four DGX Spark switchless TensorFold TP4 deployment for GLM-5.3-Flash EXL3, with ring exchange and measured benchmarks
-
Updated
Oct 6, 2026 - Shell
Switchless region mod for the SEGA Mega Drive
-
Updated
Jul 27, 2025 - C
GLM-5.3-Flash on four switchless DGX Sparks with Mia TensorFold, prepared RoCEnante and patched NCCL: builds, recovery, qualification and benchmarks.
benchmarks rdma roce multi-gpu ring-topology nccl switchless llm-inference gb10 dgx-spark dflash exl3 glm-5-3-flash tensorfold rocenante
-
Updated
Oct 25, 2026 - Python
self-hosted nvidia vision roce tp4 multimodal ring-network switchless llm-serving vllm deepseek gb10 dgx-spark
-
Updated
Sep 13, 2026 - Python
TensorFold v0.6.0 GLM-5.3-Flash engine generalized from 2 to 4 DGX Sparks (TP4) over a switchless RoCE fiber ring - uncensored EXL3 checkpoint, overlay + tooling. Upstreams: ashhart/TensorFold, MiaAI-Lab 2x-Spark stack, alexellis/switchless-nccl.
roce glm quantization nccl uncensored switchless tensor-parallelism llm-inference speculative-decoding gb10 dgx-spark exl3 dflash2 glm-5-3-flash tensorfold
-
Updated
Oct 16, 2026 - Python
Switchless-first eugr recipes and qualification gates for DeepSeek-V4.1-Flash across four NVIDIA DGX Sparks using TP4, disk-backed Engram, and hardened NCCL.
roce nccl b12x engram switchless tensor-parallelism vllm distributed-inference mxfp4 dgx-spark deepseek-v4 eugr deepseek-v4-1
-
Updated
Sep 19, 2026 - Shell
GLM-5.3-Flash EXL3 on a four-DGX-Spark switchless ring with pinned TensorFold, dual-PCIe RoCE and SparkDash/RigMark comparisons.
benchmarks rdma roce multi-gpu ring-topology nccl switchless llm-inference gb10 dgx-spark dflash exl3 glm-5-3-flash tensorfold
-
Updated
Oct 10, 2026 - Python
Switchless TP4/TP2 for DeepSeek-V4.1-Flash-1017 (uncensored) on DGX Spark (GB10). TensorFold Zig TP4 on a switchless 200G fibre ring (AOT, no CUDA JIT). Plus MiaAI-Lab TensorFold TP4 native FP8 with prompt cache on disk, SGLang TP4, TP2 Zig + EXL3 2.9bpw.
zig roce tp2 tp4 model-serving nccl uncensored switchless llm-inference deepseek sglang gb10 dgx-spark deepseek-v4 exl3 tensorfold switchless-tp4 fibre-ring engine-patches tensorfold-zig
-
Updated
Oct 30, 2026 - Shell
Saturn Switchless Mod PIC Code for common anode r-g led
-
Updated
Nov 19, 2020 - C
Add this topic to your repo
To associate your repository with the switchless topic, visit your repo's landing page and select "manage topics."