HPCA 2026
Sat 31 January - Wed 4 February 2026 Sydney, Australia
co-located with HPCA/CGO/PPoPP/CC 2026
Wed 4 Feb 2026 12:30 - 12:50 at Cronulla - GPU Memory Management and Multi-Chiplet Systems Chair(s): EJ Kim

Artificial intelligence (AI) has been increasingly applied to gaming, with large language models (LLMs) playing a key role in character control. However, efficiently co-locating game rendering and LLM inference on one GPU presents challenges due to resource constraints, diverse latency requirements, and fine-grained task scheduling. We propose LEGO, an algorithm-system co-design that enables the efficient colocation of LLM inference and game rendering tasks. Algorithm-wise, LEGO features a resource-oriented layer-skipping adaptor, which distills knowledge from skipped layers to reduce computational demand while maintaining inference accuracy. System-wise, LEGO proposes a headroom-maximizing LLM scheduler, which dynamically partitions inference tasks to utilize available rendering headroom. Evaluations on an Nvidia RTX 4090 show that LEGO meets latency targets in all scenarios, improves rendering headroom utilization by 28.8%, and enhances LLM inference accuracy by 51.4% compared to current approaches.

Wed 4 Feb

Displayed time zone: Hobart change

11:30 - 12:50
GPU Memory Management and Multi-Chiplet SystemsMain Conference at Cronulla
Chair(s): EJ Kim Texas A&M University
11:30
20m
Talk
Exploration of LLM Workload Reliability based on di/dt effects and Voltage Droops
Main Conference
Zhixing Jiang University of Texas at Austin, Justin Garrigus University of Texas at Austin, Allison Seigler University of Texas at Austin, Ethan Syed University of Texas at Austin, Yan-Lun Huang University of Texas at Austin, Mehdi Sadi Advanced Micro Devices, Tawfik Rahal-Arabi Advanced Micro Devices, Lizy John University of Texas, Austin
11:50
20m
Talk
ARIADNE: Adaptive UVM Management for Efficient GPU Memory Oversubscription
Main Conference
Hyunkyun Shin Yonsei University, Seongtae Bang DGIST, Hyungwon Park DGIST, Daehoon Kim Yonsei University
12:10
20m
Talk
LRM-GPU: Alleviating Synchronization Overhead for Multi-Chiplet GPU Architecture
Main Conference
Baiqing Zhong Sun Yat-Sen University, Zhirong Ye Sun Yat-Sen University, Xiaojie Li Sun Yat-Sen University, Peilin Wang Sun Yat-Sen University, Haiqiu Huang Sun Yat-Sen University, Zhaolin Li Tsinghua University, Zhiyi Yu Sun Yat-sen University, Mingyu Wang Sun Yat-Sen University
12:30
20m
Talk
LEGO: Supporting LLM-enhanced Games with One Gaming GPU
Main Conference
Han Zhao Shanghai Jiao Tong University, Weihao Cui Shanghai Jiao Tong University, Zeshen Zhang Tongji University, Wenhao Zhang Shanghai Jiao Tong University, Jiangtong Li Tongji University, Quan Chen Shanghai Jiao Tong University, China, Youmin Chen Shanghai Jiao Tong University, Pu Pang Shanghai Jiao Tong University, Zijun Li Shanghai Jiao Tong University, Zhenhua Han The University of Hong Kong, Yuqing Yang Microsoft Research, Minyi Guo Shanghai Jiao Tong University