uGDS is an open-source user-space library that handles NVMe command management on the host CPU while data moves by PCIe peer-to-peer DMA, supporting both NVIDIA CUDA and AMD HIP. On an A100 with a Samsung 990 PRO it reached up to 12.12x GDS's batch-read bandwidth and cut LMCache median time-to-first-token by up to 62.7%.

— via arXiv Distributed Computing, Jing Wei, Guanyi Chen, Qi Chen, Chuyu Zhang, Zheng Zhou, Chutian Wang, Jian Zhang