4 papers
NCCL EP: Towards a Unified Expert Parallel Communication API for NCCL
Amos Goldman, Nimrod Boker, Maayan Sheraizin +15
Mixture-of-Experts (MoE) architectures have become essential for scaling large language models, driving the development of specialized device-initiated communication libraries such…
GPU-Initiated Networking for NCCL
Khaled Hamidouche, John Bachan, Pak Markthub +6
Modern AI workloads, especially Mixture-of-Experts (MoE) architectures, increasingly demand low-latency, fine-grained GPU-to-GPU communication with device-side control. Traditional…
SDR-RDMA: Software-Defined Reliability Architecture for Planetary Scale RDMA Communication
Mikhail Khalilov, Siyuan Shen, Marcin Chrapek +16
RDMA is vital for efficient distributed training across datacenters, but millisecond-scale latencies complicate the design of its reliability layer. We show that depending on long-…
Bento and the Art of Repeated Research
Peter-Jan Gootzen, Animesh Trivedi
Bento provides a new approach to developing file systems, with safety and high-velocity development in mind. This is achieved by using Rust, a modern and memory-safe systems progra…