A performance/cost evaluation for a GPU-based drug discovery application on volunteer computing.

Area of Science:

• Bioinformatics
• Computational Biology
• High-Performance Computing (HPC)

Background:

• Bioinformatics relies heavily on high-performance computing (HPC) for analyzing large biological datasets.
• Graphics Processing Units (GPUs) have made HPC more accessible, enabling powerful desktop computing.
• Scaling bioinformatics applications on GPU-based systems faces challenges with power consumption, heat, and total cost of ownership (TCO).

Purpose of the Study:

• To explore volunteer computing as a scalable and cost-effective alternative to owning large GPU-based HPC infrastructures for bioinformatics.
• To evaluate the feasibility of volunteer computing for demanding applications like drug discovery.

Main Methods:

• Utilized a GPU-based drug discovery application, BINDSURF, as a computational benchmark.
• Assessed volunteer computing as a method to scale computational tasks beyond single-machine capabilities.

Main Results:

• Volunteer computing provides a viable method for scaling bioinformatics applications that require significant data processing.
• It presents a cost-effective alternative to investing in and maintaining large, in-house GPU-based HPC systems.
• The approach is suitable for applications where rapid response times are not the primary concern.

Conclusions:

• Volunteer computing is a practical and economical solution for scaling bioinformatics workloads, particularly for computationally intensive tasks like drug discovery.
• It democratizes access to HPC resources for research institutions and scientists facing infrastructure limitations.
• This model addresses the TCO and environmental concerns associated with large GPU clusters.