7 papers
Design Space Exploration of DMA based Finer-Grain Compute Communication Overlap
Shagnik Pal, Shaizeen Aga, Suchita Pati +2
Modern ML workloads demand distributing training and inference across multiple GPUs. However, these parallelization techniques often suffer from exposed critical-path communication…
CompPow: A Case for Component-level GPU Power Management
Shaizeen Aga, Mohamed Assem Ibrahim
The ever increasing demand for ML-driven intelligence in a wide spectrum of domains has led to ubiquity of GPUs. At the same time, GPUs are notorious for their power consumption ne…
Lit Silicon: A Case Where Thermal Imbalance Couples Concurrent Execution in Multiple GPUs
Marco Kurzynski, Shaizeen Aga, Di Wu
GPU systems are increasingly powering modern datacenters at scale. Despite being highly performant, GPU systems can exhibit performance variation at the node and cluster levels. Su…
DMA-Latte: Expanding the Reach of DMA Offloads to Latency-bound ML Communication
Suchita Pati, Shaizeen Aga, Mahzabeen Islam +3
Offloading communication to existing direct memory access (DMA) engines, available on most state-of-the-art commercial GPUs, has emerged as an interesting and low-cost solution to…
Chopper: A Multi-Level GPU Characterization Tool & Derived Insights Into LLM Training Inefficiency
Marco Kurzynski, Shaizeen Aga, Di Wu
Training large language models (LLMs) efficiently requires a deep understanding of how modern GPU systems behave under real-world distributed training workloads. While prior work h…
Optimizing ML Concurrent Computation and Communication with GPU DMA Engines
Anirudha Agrawal, Shaizeen Aga, Suchita Pati +1
Concurrent computation and communication (C3) is a pervasive paradigm in ML and other domains, making its performance optimization crucial. In this paper, we carefully characterize…