2026-03-11 –, Main Hall
LAMMPS's complex, non-uniform GPU feature support creates a dilemma: users struggle to find a compatible build, leading to wasted time, admin overload, and underutilized GPUs (e.g., Intel Max). We present an intelligent, wizard-style utility that performs automatic three-way matchmaking between available LAMMPS software builds, GPU accelerators, and user input requirements. The utility suggests optimal, GPU-accelerated builds and clearly explains why others are incompatible. Built atop the Drona job launcher, the wizard uses dynamically repopulating menus to guide users through the selection process, and finally injects the full configuration directly into a ready-to-run Slurm job script.
The challenge of accelerating LAMMPS is defined by technical debt: a patchwork of one-off GPU backend implementations (CUDA, Intel, etc.) support a zoo of force calculations. For users, this "impossible to document" compatibility matrix means every execution is a gamble, and for administrators, it means constant, often unnecessary, requests for new software builds.
Our solution is an intelligent, rule-based matching engine designed for HPC application specialists and systems administrators.
The Talk will cover:
The Compatibility Logic: A deep dive into the rule-based logic used to match a user's LAMMPS input requirements (force styles, packages) against the capabilities discovered by inspecting the cluster's available LAMMPS executables.
The Discovery Process: How the utility identifies GPU devices (including underutilized ones like Intel Max 1100) and correlates them with the accelerated packages present in each LAMMPS build.
UI/UX for Complex Systems: The use of the Drona job-launcher framework for building the wizard. Drona enhances the traditional form interface with dynamically repopulating dropdown menus, allowing users to filter the solution space interactively. This avoids upfront cognitive overload and ensures only valid choices are ever presented.
The End Product: The utility generates a complete, human-readable, and immediately runnable job script (e.g., Slurm/PBS) with the correct configuration parameters automatically injected, transitioning the user seamlessly from configuration analysis to job submission. This saves admin time by reducing software requests and empowers users with full transparency on compatibility.
Target Audience: HPC System Administrators, Application Support Specialists, and Research IT staff.
Takeaway: A practical, open-source methodology for creating guided, compatibility-aware user interfaces for complex command-line scientific applications.
Sahil Vartak is a Master's student in Computer Engineering at Texas A&M University specializing in full-stack development and AI integration. As Lead Web Developer at Texas A&M High Performance Research Computing, he builds and maintains production software serving over 1,000 researchers daily. His work spans modern web applications, backend services, and machine learning pipelines. Previously, he developed ML-driven analytics dashboards at McDermott International and contributed to AI research published in PerCom 2025. Sahil is passionate about building reliable, scalable software and leveraging AI tools to accelerate development.
