Competing against 49 top university teams across the Asia-Pacific region, Monash DeepNeuron secured a podium finish.
It was a rare opportunity to work with next-generation supercomputing hardware as a first-year student, contributing to a mixed team of undergraduates and final-year engineers.
We were tasked with solving high-stakes engineering problems at scale, optimizing for both computational chemistry and generative AI.
Minimising execution time for NWChem simulations across multi-node CPU clusters, requiring precise resource allocation and parallelization strategies.
Maximising inference throughput for the massive DeepSeek-R1 (671B) reasoning model using SGLang on NVIDIA H200 infrastructure.
I served as the Lead on the AI track. My focus was not just on running the model, but on understanding the system-level bottlenecks of the SGLang framework on the H200 architecture.
I worked on tuning concurrency parameters, memory management, and configuration best practices to push the inference speed of the 671B parameter model to its maximum potential.
Josh Riantoputra
Team Captain
Luca Lowndes
AI Lead
Nathan Culshaw
HPC Lead
Isaac Barnes
Engineer
Giacomo Bonomi
Engineer
The work continues beyond the competition. I am collaborating with the HPC-AI Advisory Council and Firmus Technologies to develop official deployment best practices.
My goal is to contribute technically to the SGLang ecosystem, ensuring that high-performance inference on H200 clusters is accessible and reproducible.