arxivcs.ARcs.DCcs.PF2026-07-30
Characterizing LLM Kernel Access and Memory Interaction in Multi-Partition NUMA GPUs
Donghyeon Joo, Sooraj Puthoor, Nuwan Jayasena, Bahar Asgari
Large language model (LLM) workloads motivate multi-partition GPUs as a path to scaling compute and memory capacity, but their non-uniform memory access characteristics and inter-partition communication can amplify contention and degrade locality, leading to suboptimal kernel lat…