DWDP: Distributed Weight Data Parallelism for High-Performance LLM Inference on NVL72

TL;DR

DWDP introduces a novel inference parallelization strategy for large language models that eliminates inter-rank synchronization, enabling more efficient multi-GPU execution and improved throughput.

Contribution

It proposes DWDP, a new parallelization method that offloads MoE weights and fetches experts on demand, reducing synchronization overhead in LLM inference.

Findings

DWDP improves end-to-end output throughput by 8.8% on NVL72 hardware.

It enables GPUs to progress independently, reducing workload imbalance effects.

Implemented in TensorRT-LLM, tested with DeepSeek-R1 on GB200 NVL72.

Abstract

Large language model (LLM) inference increasingly depends on multi-GPU execution, yet existing inference parallelization strategies require layer-wise inter-rank synchronization, making end-to-end performance sensitive to workload imbalance. We present DWDP (Distributed Weight Data Parallelism), an inference parallelization strategy that preserves data-parallel execution while offloading MoE weights across peer GPUs and fetching missing experts on demand. By removing collective inter-rank synchronization, DWDP allows each GPU to progress independently. We further address the practical overheads of this design with two optimizations for split-weight management and asynchronous remote-weight prefetch. Implemented in TensorRT-LLM and evaluated with DeepSeek-R1 on GB200 NVL72, DWDP improves end-to-end output TPS/GPU by 8.8% at comparable TPS/user in the 20-100 TPS/user serving range under 8K input sequence length and 1K output sequence length.