NVIDIA on Infrastructure -- Enabling the Backbone for AI and High-Performance Computing Environments and Hardware
High-Performance Computing Environments and the hardware used in them continue to evolve rapidly. The key to keeping up with the scale of change is to have an incredible software infrastructure. The Hardware Infrastructure organization develops critical workflows to enable engineers across the company to achieve the impossible - from training the newest deep learning models in data centers with thousands of GPUs to building the next-generation chip architectures.
One particularly challenging problem is to identify and resolve DL model performance issues at scale. Performance Monitors are built into GPU hardware and are critical to analyzing and improving the performance of applications. While the underlying hardware is the same, the approach to acquiring and analyzing the performance data is vastly different based on the scale. We will discuss how we enable analysis from thousands of GPUs in a cluster, down to a single GPU, and the supporting infrastructure that is required to ensure we can train the next great DL model.
Please join us to hear not only about this but other exciting opportunities at NVIDIA.
Speakers
Sharon Clay, VP of Hardware Infrastructure
Robert Hero, Senior Manager GPU Cluster Bringup
Nicole Magnus, Senior Technical Program Manager
Sharon Clay is VP of the Hardware Infrastructure organization and has been with Nvidia for 23 years,
previously working at SGI. She received her master's from UCSC in 1992 focused on Neural Networks and NLP.
Sharon's vision and passion for automating processes led to the formation of the infrastructure group within Nvidia.
The organization is now comprised of hundreds of engineers whose expertise spans all disciplines of computer science
and is innovating every aspect of NVidia HW and SW engineering.
Robert Hero is a Senior Manager within the Hardware Infrastructure organization, currently focused on building GPU
Datacenters and supporting tools for enabling NVIDiai's LLM training. Before that Robert built tools to enable
predictions of application performance through the chip design process, closely working with architects, hardware, and
software engineers across the company. He has been with Nvidia for 13 years and had done several NVidia internships
before that. Robert received his master's from UCSC in 2006 focused on Volume Visualization of Unstructured Data.
Nicole Magnus is a Senior Technical Program Manager with the Hardware Infrastructure organization focused on innovating
the component integration and CI/CD used by Nvidia's HW and SW teams. The Component Integration team creates tools that
enable the entire company to collaborate efficiently with "Speed of Light" code verifications and submissions. She has
been with Nvidia for 3 years. Nicole received her bachelor's in Computer Engineering from Prairie View A&M University in 1993.