AWS Public Sector Blog
Old Dominion University helps to modernize quantum chemistry software for 140,000 researchers with AWS
For more than four decades, the General Atomic and Molecular Electronic Structure System (GAMESS) has helped over 140,000 researchers worldwide model molecular structure and behavior across chemistry, biology, and materials science. The comprehensive quantum-chemistry software package, now comprising more than two million lines of code, supports everything from exploring new materials to drug discovery through protein-ligand binding predictions.
Despite its long history and global reach, one challenge has limited its potential. Some researchers run GAMESS on large supercomputing clusters, while others use laptops or shared departmental servers. This variation in computational environments affects reproducibility, slows collaboration, and limits who can fully participate in advanced computational chemistry.
Old Dominion University (ODU) set out to solve this problem. In collaboration with Iowa State University, the longtime home of GAMESS development, and with support from Amazon Web Services (AWS), ODU built CPU- and GPU-optimized GAMESS containers and began deploying them on AWS High-Performance Computing (HPC) services. Their work offers a practical roadmap for other institutions looking to bring legacy HPC applications into a modern, cloud-native environment.
A collaborative effort to modernize a landmark HPC application
The partnership between ODU and Iowa State is rooted in long-standing scientific connections. Masha Sosonkina, now a professor of electrical and computer engineering at ODU, previously taught at Iowa State and worked at the Ames National Laboratory under Mark Gordon, the creator and longtime steward of GAMESS. Her background in high-performance computing positioned her as a natural bridge between the two institutions. Together with Yuzhong Shen, professor and co-chair of electrical and computer engineering at ODU, and Peng Xu, adjunct assistant professor of chemistry at Iowa State, the team set out to make GAMESS more accessible to its growing global user community.
GAMESS performs a wide range of calculations related to electronic structure, molecular properties, and reaction pathways. One of its unique capabilities is the Fragment Molecular Orbital (FMO) method, which partitions large molecular systems like proteins into smaller fragments that can be treated independently. This makes GAMESS particularly well-suited for drug discovery, where researchers need to model how potential drug compounds interact with target proteins.
With that capability comes complexity and significant resource demand. University clusters often have physical space for new hardware but lack the power infrastructure to support it. “Moving to the cloud allows us to balance performance and power consumption in an environmentally conscious way,” said Sosonkina. Making GAMESS cloud-native would also enable scalability and automate the numerous manual steps traditionally required for complex molecular simulations.
The variety of setups among global users also creates challenges for reproducibility. As Xu explained, “When we do a GAMESS calculation on one system, it needs to be reproducible no matter what kind of architecture or hardware you run the calculations on. That’s a scientific requirement.”
Building a modern HPC workflow with AWS
The transition to the cloud took shape after ODU connected with AWS. The team received assistance with testing, hands-on workshops, and recurring technical meetings with cloud HPC specialists. AWS provided support at two levels: working directly with the research team and supporting ODU’s research cloud computing group, whose systems engineer, Terry Stilwell, led the container development effort.
“We learned how to create the cluster and use it from our desktops,” said Sosonkina. “The AWS engineers really listened to what we wanted and helped us find ways to make it happen.”
Technical work on the containers began in April 2025. After an AWS workshop in June 2025, the ODU team had a functional containerized version of GAMESS by August 2025. They built separate CPU and GPU containers instead of a single combined version, giving researchers flexibility depending on their workload.
“Matching versions of the tools used to compile GAMESS was our biggest challenge,” said Sosonkina. “Everything had to be reconciled into one container that could run consistently across different environments.”
For orchestration, the workflow uses AWS Parallel Computing Service (AWS PCS) with Slurm, allowing researchers to run HPC jobs on Amazon Elastic Compute Cloud (Amazon EC2) instances using familiar commands and job-queuing processes. The team also explored AWS Batch for managed batch workloads and Amazon Elastic Kubernetes Service (Amazon EKS) for container orchestration. To support shared storage across compute nodes, the team used Amazon Elastic File System (Amazon EFS).
Validating results and preparing for scale
To confirm that the containerized version of GAMESS maintained computational accuracy, the ODU team ran validation tests using 48 protein-ligand conformers from Iowa State’s published research on a four-node deployment, comparing outputs directly with the published values. The results matched. Using Tuning and Analysis Utilities (TAU) performance analysis software, they also confirmed that the containers introduced no performance overhead compared to traditional installations.
This validation step is essential for any scientific application migrating to the cloud. Researchers depend on consistency, especially when comparing data across labs or running long simulations on different hardware. Demonstrating reproducible results gives the broader GAMESS community confidence that the cloud-native version is reliable.
Why a hybrid workflow matters for drug discovery
Cloud infrastructure is particularly valuable for GAMESS because of the hybrid approach researchers use to balance speed and accuracy. Using molecular mechanics software like VeraChem, researchers can quickly generate roughly 1,000 molecular conformations on a single compute node. Quantum mechanical refinement, however, costs several orders of magnitude more in time and resources.
Rather than running expensive quantum-mechanical calculations for every conformation, researchers select only the most promising candidates, typically around 30, based on probability ranking. Automating and orchestrating this selective workflow on AWS is the team’s next goal, with the potential to dramatically reduce both time and cost for drug discovery applications.
Looking ahead: Scaling, energy efficiency, and AI integration
With a validated container now in place, the team is preparing to scale. Full production workloads and comprehensive performance benchmarks are next, but the foundation is set. An ARM-optimized container is planned for AWS Graviton Processors, enabling researchers to run quantum chemistry calculations with reduced environmental impact and potentially lower costs. Prior research by the team showed the potential to improve energy efficiency by 30 percent with only a 10 percent reduction in computational speed, a trade-off that could yield substantial savings across 140,000 users.
The validated containers can also be tested across larger node counts using Elastic Fabric Adapter (EFA) networking. The team is also exploring the use of artificial intelligence (AI) to prioritize promising molecular conformations before performing full quantum-mechanical refinement. “We hope that GAMESS can be adapted to an AI-based selection strategy, so it’s not limited to the current hardware and environment,” said Xu.
A blueprint for modernizing legacy HPC applications
This project demonstrates that, with AWS support, even long-standing, complex HPC software can be successfully modernized through containerization and cloud-native workflows. For other research institutions considering a similar migration, the team recommends:
- Starting with containerization
- Selecting managed services like AWS PCS
- Monitoring cloud budgets closely
- Choosing workloads suited for diverse compute architectures and automatic scaling
By overcoming traditional computational and infrastructure limitations, researchers can explore more complex models and accelerate discovery across scientific disciplines. Institutions interested in similar modernization efforts can visit AWS to learn how AWS can help.