A toolkit for discovering cluster network topology.
Tartan: Evaluating Modern GPU Interconnect via a Multi-GPU Benchmark Suite
Ulysses sequence-parallel all-to-all as a torch custom op, moved by the GPU copy engines into torch symmetric memory. Zero SM usage; 1.66-2.17x over torch.distributed on NVLink.
NUMA-aware multi-CPU multi-GPU data transfer benchmarks
This script collects some informations about NVLink and PCI bus traffic of NVidia GPUs. Results are published as prometheus metrics via a websocket.
Multi-GPU acceleration for MiniMax H3 video generation on NVIDIA V100 (sm_70). Ulysses sequence parallelism as a drop-in ComfyUI custom node — ~19 min to ~7 min on 8x V100.
Comprehensive NCA-AIIO exam prep: study notes, diagrams, screenshots, and field experience for the NVIDIA Certified Associate: AI Infrastructure and Operations certification.
Communication cost modeling for tensor parallel LLM inference with TP vs PP vs hybrid comparison, VRAM analysis, pipeline bubble modeling, regime detection, and cost-efficiency. Shows TP dominates on NVLink, PP has 47% bubble at 8 GPUs, and LLaMA-70B needs 8× A100 or 2× H100 for VRAM.
Running large LLMs on pre-Ampere NVIDIA hardware — Tesla V100 (sm_70), RTX 2080 Ti (sm_75), CMP 170HX. Measured benchmarks, vLLM forks, and the hardware side: NVLink on SXM2 carrier boards, driver traps, cooling, used-kit acceptance.
Ares: Multi-Cluster Kubernetes Scheduler with GPU Topology Optimization (Intra-Node, Inter-Node, Inter-Cluster) and Exactly-Once Execution Semantics
Real-time per-link NVLink bandwidth monitor + inter-GPU P2P benchmark for NVIDIA multi-GPU systems. Lightweight C++ — the monitor needs no CUDA toolkit.
Open hardware desktop AI node: 4× Tesla V100, 128GB HBM2, PCIe/NVLink topology and V-Core liquid/air cooling.
A plain-spoken, interactive course on NVIDIA's Blackwell architecture — the chips, engines, and rack-scale systems built for the age of AI reasoning.
LMPool: Distributed KV Cache Pooling for LLM Inference（面向 LLM 推理的分布式 KV Cache 池化方案）
Analytical benchmark connecting pipeline parallelism and KV cache management: bubble-time overlap for disaggregation, prefix reuse tax of PP stages, and tiering speedup across NVLink, InfiniBand, and PCIe
GPU-native agent-swarm orchestration for the NVIDIA AI stack — NeMo, NIM, Triton, DCGM, NGC, NIXL, OpenShell. Spawn GPU-pinned agent teams across DGX/HGX nodes with NVLink-aware scheduling, task DAGs, adaptive scheduling, and full observability.
From-scratch Ring-AllReduce on CPU (C++20) and GPU (CUDA, multi-device): real multi-GPU benchmarks, Nsight profiling, and a bandwidth-optimal ring implementation built without NCCL.
Close-to-metal C/CUDA lab for RL inference fast paths: persistent GPU workers, hugepage KV arenas, cacheline-aware command rings, and async reward handoff. Goal: remove page faults, malloc/free, scheduler wakeups, CPU round-trips, and KV migration from the per-token path.
To associate your repository with the nvlink topic, visit your repo's landing page and select "manage topics."