Adam DeConinck

Summary

Senior systems architect and engineer specializing in the design, build, and operation of large-scale, high-performance GPU clusters for AI and machine learning workloads. Proven track record of architecting and troubleshooting large-scale systems for both internal R&D and major hyperscale customers.

Professional Experience

NVIDIA, 2022 - Present: Architecture of Future GPU Clusters

Director of Next-Gen Cluster Architecture, 2025 - Present
Senior Manager, 2023 - 2025
Principal Systems Software Engineer, 2022 - 2023

• Led the architecture process for future cluster architectures based on the Grace-Hopper, Grace-Blackwell and Vera-Rubin platforms
  • Reference designs developed for clusters based on DGX B200, GB200 NVL36x2, GB200 NVL72, and GB300 NVL72 products
  • Published internal reference architectures to provide a validated design for both internal and customer builds
• Designed and led proof-of-concept projects for next-generation compute fabric architectures, including initial deployments for XDR InfiniBand and multi-plane SpectrumX (RoCE) Ethernet fabrics
• Developed a systematic approach to design of new architectures to streamline development for upcoming builds
• Initiated a new team focused on cluster architecture and datacenter design, and grew that team from an initial group of 4 engineers to a team of 10
• Developed automation and performed at-scale troubleshooting for Eos, a 4,000+ GPU DGX H100 SuperPOD ranked at #9 on the Top500 list (Nov 2023)
• Collaborated directly with key customers on hyperscale AI training clusters based on NVIDIA reference architectures
• Track chair for datacenter and cloud papers at NTECH 2025, an internal technology conference

NVIDIA, 2019 - 2022: DGX SuperPOD and Customer Builds

Senior Product Architect, 2021 - 2022

• Collaborated closely with enterprise product management teams for large-scale HPC projects
• Design and project management for initial customer deployments of NVIDIA Base Command Manager, a tool for cluster deployment and system management
• Developed automation and content for early NVIDIA Launchpad projects

Senior Solutions Architect, 2019 - 2021

• Customer-facing systems architect for design of supercomputing systems
• Systems software lead for several initial customer deployments of DGX A100 SuperPOD, including developing the deployment automation system for these clusters
• Automation and hardware deployment for Selene, a 4,000+ GPU DGX A100 SuperPOD raked at #5 on the Top500 list (June 2020)
• Technical lead for multiple RFP responses for supercomputing projects at Department of Energy sites
• Technical lead for initial customer deployments of DRIVE Constellation, a datacenter product for validation of autonomous vehicle software stacks
• System management of multiple internal clusters geared toward benchmarking and application development

Facebook, 2017 - 2019: Exabyte-Scale Blob Storage

Senior Production Engineer

• Production Engineering lead for metadata services of Facebook Tectonic, an exabyte-scale distributed filesystem used for data warehouse and blob storage applications (known internally as “warm storage”)
• Migrated Tectonic control plane services to shared container management platform, reducing usage of high-cost instances by >60% and reducing on-call burden for hardware maintenance tasks
• Developed and deployed automated capacity planning tools based on actual and projected traffic requirements
• Analyzed and improved database replication strategy for metadata services, mitigating an impactful class of production incidents
• Characterized the performance of Tectonic storage service on a next-generation storage server (``Bryce Canyon’’)
• Onboarded and mentored several new team members, including both new college grads and senior hires
• Organized multiple team events for Seattle Production Engineering team

Los Alamos National Laboratory, 2014 - 2017: Top-10 Supercomputing

HPC Cluster Administrator

• Production team lead for intial deployment of Trinity, a 20,000-node Cray XC40 supercomputer ranked at #6 on the Top 500 list (Nov 2015)
• Developed system automation for configuration and production operations of LANL capability systems, including leading a migration from Cfengine to Ansible
• Coordinator for production change control process for LANL HPC systems
• Derivative classifier (Q clearance) for review of HPC publications, including conference presentations and internal reports

NVIDIA, 2012 - 2014: HPC Development Platforms

HPC/Cloud Systems Engineer

• Supported an early internal HPC development platform for the Solution Architect team, including ongoing hardware and software refresh
• Collaborated closely with developer technology engineers and third-party developers to support specific hardware tests
• Support for customer evaluation workloads on a dedicated NVIDIA-hosted HPC platform
• Developed internal and cloud-hosted training platform for customer tutorials, including at conferences such as Supercomputiong and GTC

R Systems NA, 2010 - 2012: Custom HPC Clusters

Systems Engineer

• Designed, deployed, and supported hosted HPC clusters for industrial customers
• Developed custom solutions to support particular customer workloads, including climate simulations, financial modeling, and Formula-One racing
• Experience included a wide variety of software stacks, including Red Hat Enterprise Linux, Ubuntu, and Windows 2008 HPC

Skills and Technologies

• Technical leader with experience planning large, complex projects and setting technical direction, both as a manager and as a senior/staff individual contributor
• Software development experience focused on development of operational tooling and infrastructure services, primarily using Python, Go, and C++
  • Scientific software and data analysis experience using Python, Julia, Matlab, and Fortran
  • Debugging and hobby projects for applications written in Rust, Clojure, and OCaml
• Experienced SRE responsible for large-scale distributed systems (10,000+ nodes) including compute clustgers, distributed storage systems, and microservice-based architectures
• Familiar with a wide variety of bare-metal deployment toolkits based on core principles of network boot (PXE) followed by either stateless operation with RAM-based root, or installation of the OS on a local disk
  • Tools include Canonical MaaS, NVIDIA BCM, Warewulf, Perceus, xCAT, Cobbler, …
• Configuration management using tools such as Ansible, Cfengine, Chef, or Puppet to converge a running system to a desired state
• Deployment and performance tuning of parallel and distributed filesystems such as Lustre, Ceph, HDFS, and S3-compatible object storage
• Extensive experience with the design and operation of complex monitoring and observability systems, both using custom implementations and with open-source tooling such as Prometheus, Grafana, Loki, and the ELK stack

Education

M.S., University of Illinois at Urbana-Champaign, Materials Science and Engineering.
Thesis title: Fabrication, dynamics, and self-assembly of anisotropic colloidal particles.

B.S., Michigan Technological University, Physics.
Minors in Mathematics and Electronic Materials.

Selected Publications

• A. DeConinck, “Architecting and deploying compute clusters for large language models”, 2nd Workshop on Advancing Neural Network Training, International Conference on Machine Learning, July 2024.
• A. DeConinck, A. Bonnie, K. Kelly, S. Sanchez, C. Martin, M. Mason, J. Brandt, A. Gentile, B. Allan, A. Agelastos, M. Davis and M. Berry. “Design and implementation of a scalable monitoring system for Trinity”, Proc. Cray User’s Group, May 2016.
• S. Sanchez, A. Bonnie, G. Van Huele, C. Robinson, A. DeConinck, K. Kelly, Q. Snead and J. Brandt, “Design and Implementation of a Scalable HPC Monitoring System”, Wrk. on Monitoring and Analysis for High Performance Computing Systems Plus Applications (HPCMASPA), 2016 IEEE International Parallel and Distributed Processing Symposim (IPDPS), May 2016.
• A. DeConinck and K. Kelly. “Evolution of Monitoring Over the Lifetime of a High Performance Computing Cluster”. Wrk. on Monitoring and Analysis for High Performance Computing Systems Plus Applications (HPCMASPA), 2015 IEEE International Conference on Cluster Computing (CLUSTER), September 2015.
• A. J. DeConinck. “Tools and Tips for Managing a GPU Cluster”. GPU Technology Conference, March 2014.