Run Molecular Dynamics Simulations in the Cloud as Fast as 1139 ns/day, as Cost-effective as $29/μs

Benchmarking GROMACS on Fovus

Photo by ANIRUDH on Unsplash

GROMACS has revolutionized molecular dynamics (MD) simulations by empowering researchers to simulate molecular interactions across biology, chemistry, and materials science, driving advancements in protein folding, drug discovery, material design, and nanotechnology.

However, running GROMACS on cloud GPUs presents significant scalability and cost challenges. GPUs are among the most in-demand resources in the cloud, making them difficult to procure, expensive to use, and challenging to scale efficiently. Due to high demand from AI, high-performance computing (HPC), and other compute-intensive workloads, cloud providers often face GPU shortages, leading to limited availability and long provisioning times. Scaling up GPU resources dynamically can be unpredictable, as instances may not be readily available when needed in any particular cloud region and availability zone.

Additionally, GROMACS achieves optimal performance through close CPU-GPU collaboration, offloading many calculations to the GPU while relying on significant CPU processing power for task management, communication, and I/O. Thus, choosing the proper GPU and CPUs is essential. Yet, diverse cloud GPU offerings, flexible system configurations, and complex pricing models make it difficult for computational scientists to find the optimal choice. Without proper hardware choice or HPC strategy, GROMACS runs have been perceived as unaffordable at scale, given the high cost of cloud GPUs.

With Fovus, these challenges are now eliminated! Fovus is an AI-powered, serverless HPC platform delivering intelligent, scalable, and cost-efficient supercomputing power at the scientists’ and engineers’ fingertips.

Free Automated Benchmarking

Fovus auto-benchmarks your GROMACS workloads for free to evaluate how different HPC strategies (GPU, CPUs, memory configurations, and more) quantitively impact the overall runtime and cost of your production runs, gaining the visibility of HPC strategy space for your particular GROMACS use cases.

AI-driven Strategy Optimization

Driven by the benchmarking data, Fovus automatically determines the optimal HPC strategy (GPU, CPUs, memory configurations, and more) to optimize runtime and costs for GROMACS runs at scale autonomously.

Dynamic Multi-Cloud-Region Auto-Scaling

Fovus dynamically allocates optimal GPUs to distribute your GROMCAS runs across multiple cloud regions and availability zones according to their availability dynamics. This multi-cloud strategy ensures superior availability and scalability for efficient distributed computing of large-batch GROMACS simulations.

Intelligent Spot Instance Utilization

Fovus intelligently auto-leverages the optimal spot GPU instances across multiple cloud regions and availability zones to slash the cost of GPU-accelerated GROMACS runs. Using GROMACS’ checkpoint function, Fovus’ spot-to-spot failover capability ensures result integrity by seamlessly recovering from spot instance interruptions, making large-scale GROMACS simulations affordable and reliable.

Continuous Improvement

Fovus auto-updates benchmarking data and auto-refines your HPC strategies for GROMACS as cloud infrastructure and GPU technology evolve. This continuous improvement ensures sustained efficiency and cost-effectiveness and keeps you ahead of the curve.

Serverless HPC Model

Fovus simplifies cloud HPC with AI-driven autonomous automation for cluster setup and infrastructure provisioning, allowing chemists to focus on discovery. Workloads deploy in a single CLI command or a few clicks via web UI, with Fovus handling logistics, optimization, scaling, maintenance, and many more. No cloud or cluster management is needed — pay only for workload runtime.

To reveal the real-world impact of these powerful features, we present a benchmarking study for GROMACS on Fovus, highlighting its superior speed and cost-effectiveness for running GPU-accelerated GROMACS simulation across small, medium, and large protein systems with 20,000 to 1 million atoms. Fovus intelligently optimizes HPC strategies and orchestrates cloud logistics according to spot GPU availability and pricing dynamics as well as user-defined objectives — whether minimizing cost, runtime, or both.

The Benchmarks

We evaluated six molecule systems of varying sizes to demonstrate Fovus’ performance and cost-efficiency:

• System 1: R-143a in hexane, 20,248 atoms — small
• System 2: RNA piece with explicit water, 31,889 atoms—small
• System 3: Protein inside a membrane, 80,289 atoms — medium
• System 4: Protein in explicit water, 170,320 atoms — medium
• System 5: EAG1 membrane protein embedded in a lipid bilayer, 615,924 atoms — large
• System 6: STMV Virus Protein, 1,066,628 atoms — large

Each system was simulated with GROMACS 2023.2 within a Docker container, with checkpointing enabled (-cpi) using the following command:

docker run --volume=$PWD:/container_workspace --workdir=/container_workspace 
--gpus=all nvcr.io/hpc/gromacs:2023.2 gmx mdrun -v -s $benchmark_name.tpr 
-nb gpu -pme gpu -bonded gpu -update gpu  -ntmpi 1 -pin on -pinstride 1 
-nsteps 2000000 -deffnm $file_name -cpi

Each GROMACS simulation is deployed to Fovus and runs on an optimal spot GPU determined by Fovus’ HPC strategy optimization. In the event of a spot instance interruption, Fovus automatically fails over the GROMACS run to a new optimal spot GPU, resuming the simulation from the last saved checkpoint. Additionally, Fovus exclusively utilizes datacenter-grade GPUs with Error-Correcting Code (ECC) memory, ensuring reliable computation and trustworthy results by eliminating the risk of random errors inherent in consumer-grade GPUs.

For each protein system, benchmarking was conducted three times on Fovus, each with a different objective specified for HPC strategy optimization:

1. Minimizing Cost: Prioritize cost over performance. Get the most cost-efficient strategy.
2. Minimizing Cost and Time: Prioritize cost and time minimization equally. Optimizing for both cost-efficiency and speed.
3. Minimizing Time: Prioritize performance over cost. Get the fastest strategy.

Three key performance metrics were analyzed:

• ns/day (nanoseconds per day): Measures simulation speed, indicating how much simulated time can be computed in one real-time day.
• $/µs (dollars per microsecond): Evaluates the unit cost of running a 1-µs long simulation in simulated time.
• ns/$ (nanoseconds per dollar): Assesses cost efficiency, representing how much simulated time can be computed for each dollar spent.

Together, these metrics clearly show the performance and cost efficiency of running GPU-accelerated GROMACS simulations on Fovus.

Benchmarking Results

Below are the performance and cost-efficiency results achieved on Fovus under each objective:

Summary of Results

The benchmarking results demonstrate Fovus’s ability to deliver high-performance and cost-effective GPU computing power for GROMACS simulations while optimizing for the best performance-cost tradeoffs achievable in the cloud according to user preferences. Key takeaways include:

• Cost Efficiency: Researchers can achieve significant cost savings, with simulations costing as little as $28.64/µs (34.9 ns/$) for small molecule systems, and as little as $81.3/µs (12.3 ns/$) and $754.85/µs (1.3 ns/$) for medium and large molecule systems, respectively.
• Speed: Researchers can drastically accelerate time to insight by running simulations at speeds of up to 1,139.4 ns/day for smaller molecule systems and up to 428.3 ns/day and 65.3 ns/day for medium and large molecule systems, respectively.
• Optimal performance-cost tradeoff: If time and cost are both critical to your mission, Fovus offers the optimal tradeoff, achieving 491.7 ns/day at $61.01/µs (16.4 ns/$) for small-molecule systems, 162.8 ns/day at $137.91/µs (7.3 ns/$) for medium-molecule systems, and 20.8 ns/day at $1,078.49/µs (0.9 ns/$) for large-molecule systems, respectively. This strategy provides the ideal solution for researchers to balance time and cost budgets.
• Sustainability: With Fovus, all these metrics will continue to improve as we automatically upgrade your HPC strategy in response to the rapid evolution of cloud infrastructure.

These results highlight Fovus’s ability to handle diverse GROMACS workloads, from small to large, complex molecule systems, making it an ideal platform for accelerating molecular dynamics research.

Focusing on Discovery, Not Logistics

At its core, GROMACS enables researchers to tackle some of the world’s most pressing scientific challenges, from drug discovery to materials design. However, scientists’ time is best spent interpreting results and advancing their research — not managing cloud resources or troubleshooting computational bottlenecks.

Fovus ensures that researchers remain at the forefront of innovation by streamlining cloud HPC workflows and autonomously optimizing cloud runtime and cost. By shifting the focus from dealing with cloud hassles to discovery, Fovus empowers scientists to achieve breakthroughs faster than ever.

For more information on how Fovus can transform your GROMACS workflows, visit fovus.co.

Get a free trial today!