Tartan: Evaluating Modern GPU Interconnect via a Multi-GPU Benchmark Suite
-
Updated
Sep 12, 2018 - Cuda
Tartan: Evaluating Modern GPU Interconnect via a Multi-GPU Benchmark Suite
Giddy - A lightweight GPU decompression library
Deterministic GPU PCIe link and bandwidth diagnostic using observable hardware data (NVML + CUDA memcpy).
A CUDA host↔device scatter-gather transfer-orchestration engine — implements the full strategy space and honestly characterizes which strategy wins at which granularity/sparsity (bare-metal: Blackwell + Hopper).
To associate your repository with the pcie topic, visit your repo's landing page and select "manage topics."