Beaver: Elastic GPU Sharing between ML and Latency-Critical vRAN Workloads

2026-10-01

AI Summary

This study addresses the resource contention challenge of sharing GPUs between latency-sensitive vRAN and throughput-oriented ML workloads in AI-RAN by proposing Beaver. This system introduces a novel slot-granularity dynamic SM repartitioning mechanism coupled with HBM bandwidth-aware kernel rewriting to elastically yield resources during vRAN memory-critical phases, enabling synergistic management of compute and memory. Experimental results on NVIDIA H200 demonstrate that Beaver strictly guarantees vRAN p99.9 latency with zero deadline violations while preserving 74% of Llama-3 serving throughput, further exhibiting strong cross-hardware generalizability.

Abstract

Within the shared industry vision of AI-RAN, AI-and-RAN seeks to co-locate virtualized radio access network (vRAN) workloads and AI services on shared GPUs. This sharing is inherently asymmetric: vRAN workload is latency-critical, whereas the machine learning (ML) workload is a throughput-oriented, best-effort co-tenant. We present Beaver, a GPU sharing system that jointly manages compute and memory resources while protecting the vRAN's strict processing deadline. Beaver sizes the vRAN's streaming multiprocessor (SM) allocation from each slot's scheduled workload, repartitions SM allocations at slot granularity, and rewrites compiled ML kernels to yield HBM bandwidth during the vRAN's memory-critical phases. We implement Beaver and evaluate it using NVIDIA Aerial with heterogeneous multi-cell workloads, real-world cellular traces, production inference kernels, and full-stack LLM serving. On an H200 GPU, Beaver keeps the vRAN's p99.9 latency within its 1.5ms uplink deadline while retaining 74% of Llama-3.3-70B serving throughput. It also incurs no observed deadline misses under replayed cellular traces, protects a 375us downlink deadline, and generalizes to other GPUs including A100, GB10 and GH200.

Problem

• GPU sharing
• vRAN
• AI-RAN
• latency-critical
• ML workloads

Innovation

• Elastic GPU Sharing
• AI-RAN
• Latency-Critical vRAN
• Kernel Rewriting
• Resource Partitioning