Golden Gate Bridge
A young high performance enthusiast

Haoquan Chen陈淏泉

I build the infrastructure between ambitious workloads and the powerful machines that run them.

Research Interests

Where I want to push the limits

The questions I keep coming back to when hardware, compilers, and distributed systems meet.

GPU ArchitectureMulti-GPU execution, accelerator memory systems, and architecture-aware performance.
AI InfraTraining, serving, scheduling, and resource management for AI clusters.
HPC SystemsSystem software, scheduling, and runtime support for large-scale scientific computing.

Experience

How I learned systems from SUPERCOMPUTERS

Labs taught me abstractions; supercomputers taught me constraints; workloads taught me to measure.

MiniMax

Developing a Megatron-based training framework for frontier video models, scaling workloads across large GPU clusters while improving parallel efficiency, load balancing, and fault recovery.

MegatronPyTorchNCCLDistributed Training

arcSYSu Lab, Sun Yat-sen University

Worked around HPC containers, compiler-assisted deployment, and distributed system support for heterogeneous clusters. The experience shaped my understanding of container runtimes, compiler toolchains, reproducible evaluation, and production systems on Tianhe-Xingyi supercomputer.

HPC ContainersDistributed SystemsDockerKubernetes

SYSU Supercomputing Team

The team gave me full-stack HPC training from the ground up, from hardware setup and cluster configuration to operator development, GPU tuning, application optimization, and performance profiling, with hands-on experience on self-built clusters and Bridges-2 supercomputer.

Parallel ProgrammingGPUKernel OptimizationPerformance Profiling

Research

What came out of the experiments

Artifacts from turning systems ideas into papers, prototypes, and evaluations that have to survive real machines.

SC'25 · CCF-A

coMtainer

Compilation-assisted HPC Container Images with Enhanced Adaptability

Built the container transformation and evaluation workflow for heterogeneous HPC clusters, preserving near-native execution while improving deployment adaptability.

coMtainer build graph and image model

Projects

Things I built along the way

Small enough to ship, technical enough to teach me something about performance, platforms, or play.

OpenMX · Performance optimization

Performance Optimization of OpenMX

Optimized the nanomaterials simulation package on Bridges-2 using profiling, kernel refactoring, and hybrid MPI + OpenMP parallelization. Achieved 2.2x end-to-end performance improvement.

C/C++FortranMPI+OpenMPVTuneuProfScaLAPACK/ELPA
YatCC-OL · HPC code-servers

Lightweight HPC Development Platform

Built an online development platform above the Tianhe-Xingyi supercomputer for the YatCC-AI course, with Flask backend services and Kubernetes integration for standard cloud clusters.

FlaskKubernetesRedisDocker
WebGL · Space fighter game

PlanetWar: Maverick

A WebGL space fighter game, playable directly in your browser.

WebGLJavaScriptGame
Play in browser
PlanetWar: Maverick game logo

Awards

A few milestones from the journey

World ChampionISC25 Student Cluster CompetitionJune 2025
Top PrizeTencent Scholarship, 1 / 424Nov 2025
Third PrizeCompiler Design Competition 2024Aug 2024
Outstanding StudentsScholarship, Sun Yat-sen University2023-2025

Technical Skills

Tools I reach for when things get REAL

ProgrammingPython, C/C++, Shell scripting
Parallel ComputingMPI, OpenMP, SIMD
Performance ToolsNsight, VTune, uProf, Perf
HPC & CloudDocker, Kubernetes, Slurm
AI InfrastructureMegatron, PyTorch, NCCL

Writing

What I noticed while building and traveling