Navigating the Challenges of Industrial Performance Simulation

Tuesday, October 16, 2018
3:30 PM to 4:30 PM
EER 0.806
Free and open to the public

As HPC applications move from single GPU to multi-GPU, there are growing challenges around how to accurately model and predict the performance of future GPU designs and systems. This talk will provide a retrospective on the development of a simulation workflow used within NVIDIA today and the fidelity versus simulation-time trade-offs needed to enable system level simulation of HPC and data center workloads. I will reflect on the realities of building and influencing industrial architectural models, while lobbying UT's future architects to develop soft skills, such as recognizing perception bias and appropriate salesmanship, to improve their research impact. Finally, I will describe how architectural models are being extended to support multi-GPU workloads and the areas where methodology improvements must occur to improve confidence in future performance projections and architectural innovation.


David Nellans


David Nellans currently manages the System Architecture Research group at NVIDIA where he has been for the last 6 years. His team provides co-designed SW and HW solutions to enable scalable GPU performance and energy efficiency in a post-Moore's world. By definition, system architecture and design spans multiple sub-specialties and his team looks for novel solutions that can vary from circuits to software. NVIDIA maintains a culture of open research with a strong track record of publishing and supporting the academic community, while also maintaining close ties with product groups to ensure that research positively effects the company within the next 5-10 years. Prior to NVIDIA, Dave was an early technical manager at Fusion-IO helping the company grow from 50 to 1000 employees, while successfully deploying the first PCIe attached NAND-Flash systems within Facebook and Apple data centers. Fusion-IO went public in 2011 with a valuation of 1.5B USD. Dave holds a BA from Colgate University and a PhD from the University of Utah, both in computer science.