Calculon
An analytical performance model and open-source tool for high-level co-design of LLM training and inference systems.
I am a computer architecture research scientist and software/hardware engineer. I am a senior research scientist at NVIDIA Research working on next generation hardware and software systems for NVIDIA networks.
Ph.D. in Electrical Engineering, 2016
Stanford University
M.S. in Electrical Engineering, 2012
University of Utah
B.S. in Computer Engineering, 2010
University of Utah
An analytical performance model and open-source tool for high-level co-design of LLM training and inference systems.
A parallel discrete event simulation framework.
A parallel SSH-based remote machine management system.
An easy-to-use python package for parallel task execution.
A flit-level interconnection network simulator.
A project development find and goto system.
We present Calculon, a parameterized analytical performance model and open-source tool for high-level algorithm-architecture codesign of LLM training and inference systems.
We use an analytical performance model to design systems that train LLMs of up to 128 trillion parameters at 75%+ MFU.
We present ParaGraph, an open-source toolkit that bridges applications and high-fidelity network simulators to enable hardware-software co-design of supercomputer-scale systems.
We present two practical and efficient incremental adaptive routing algorithms for HyperX.
We present the first at-scale implementation of the HyperX topology and compare it side-by-side to a traditional Fat-Tree.
We present SuperSim, an open-source flit-level interconnection network simulator for large-scale high-performance networks.
This dissertation presents Sikker, a highly-scalable high-performance distributed system architecture for secure service-oriented computing.
NVIDIA
Salt Lake City, Utah
At NVIDIA Research, I work on AI system architecture. I study how accelerator design, memory hierarchy, and interconnect shape the performance and efficiency of modern language-model workloads. I build models and simulation tools to explore hardware/software trade-offs, evaluate new inference system designs, and investigate architectures for agentic AI and small language models. My work sits at the intersection of computer architecture, networking, and machine learning systems, with a focus on turning research ideas into practical directions for future AI platforms.
Google
Sunnyvale, California
I worked in the system infrastructure team specifically on tightly-coupled high-performance network technologies covering topologies, routing algorithms, network acceleration and offloading, congestion control, processing architectures, and more.
Hewlett Packard Labs
Fort Collins, Colorado
I was a lead architect on design of a new high-performance network designed for large-scale high-performance computing (HPC) systems and massively parallel memory-driven computing (MDC) systems. I was the key designer of the Gen-Z routing architectural specification. Using a data-driven simulation approach I guided the design of a novel multi-chip module (MCM) switch architecture that utilizes co-packaged integrated photonics.
L3 Communications
Salt Lake City, Utah
Developed digital processing architectures for encryption, networking, and waveform data processing. Designed systems with strict requirements for high-performance, low-power, and low-area. All designs were meticulously scrutinized for high-reliability and signal integrity.