More than ever, scientists need new technologies to help solve many of the most pressing issues facing society like reversing climate change, addressing food insecurity, and developing lifesaving therapeutics. Fundamentally, these problems are chemistry and materials science challenges, and some will require the transformational power of a scaled quantum computer. However, given the advancements of classical computing services in the cloud, scientists can start to make rapid progress towards solving these problems today.

Molecular dynamics (MD)—the simulation of molecular interactions—is one computational problem that pushes the boundaries of what is possible with today’s high-performance computing platforms. More powerful platforms for molecular dynamics simulations could unlock the development of new materials, new drugs, and more efficient batteries, so a team of Azure Quantum scientists recently set out to ask a fundamental question: what are Azure’s capabilities for these types of simulations? Here’s what we learned and how any scientist can use Azure to drive similar results today.

Visual representation of Molecular Dynamics simulation (in this example, Satellite Tobacco Mosaic Virus– PBID 7M3T).  
Images generated using MOL* 3D viewer on RCSB PDB 3D View (source) 

Azure provides the latest in high-performance computing

Molecular dynamics calculations pose unique high-speed communication challenges which require state-of-the-art computing capabilities. The Microsoft Azure cloud architecture helps researchers overcome these hurdles by allowing them to take advantage of the latest software and hardware developments required for chemistry and materials science research. By simplifying the provisioning of the necessary high-performance computing (HPC) resources, Azure helps scientists rapidly deploy and execute complex simulations of the structure and dynamics of macromolecules. This significantly accelerates chemical and materials innovation, for example enabling the creation of better pharmaceuticals by modeling biomolecules and their relevant properties at a faster pace.

Azure high-performance computing engineers have made significant progress in advancing the scale and networking speeds of the cloud platform. The latest Azure virtual machines use InfiniBand for low-latency communication across distributed nodes for differentiated scalability and performance gains. Our team has demonstrated excellent parallel efficiency and an increase of over 200 percent in benchmark simulations compared to previous virtual machines, particularly for larger simulations.

Customers are already taking advantage of Azure cloud HPC capabilities today. You can read more here.

“With Azure HPC, we’ve seen about a 50 percent speedup on some of our chemistry calculations that we run—which is critical for R&D because every second counts, not just for getting the results quickly, but also in terms of cost and throughput.”— Glenn Jones, Research Manager at Johnson Matthey Technology Center.

Simulating complex molecular dynamics in Azure enables R&D acceleration

Molecular dynamics simulations translate atomic-scale forces and energies into molecular motion and are an important tool for both life sciences and materials science research. In life sciences, for example, molecular dynamics simulations are used to understand proteins, ligands, and their associated properties, which can be used to accelerate the discovery of better pharmaceuticals by modeling drug molecules and their relevant protein binding sites.

Molecular dynamics workloads stand to greatly benefit from specialized HPC systems whose architectures use both graphics processing units (GPUs) and central processing units (CPUs) with high-speed interconnects between them. Optimized MD simulations pose unique computational challenges related to time scale, sampling, and analysis, requiring powerful computing nodes with low-latency communications—a task for which Azure is uniquely suited.

Time scale and sampling

Biologically relevant events occur across a wide range of timeframes from fractions of a second to decades. Consider an example event that occurs within 1/1,000th of a second. While this may seem short in real time, it represents a massive computational workload. To capture relevant chemical properties in these simulations, the system needs to compute the position and momentum of all the atoms very frequently. Often, these properties are calculated every 10-12 seconds for all atoms in the system. This means one would have to carry out at least a billion calculation steps to perform simulations on the same timescales as the event of interest.

Powerful computing resources in Azure make this computational problem more tractable. In addition to requiring state-of-the-art processors—including both CPUs and GPUs—to accurately evaluate energies and forces, scalable molecular dynamics simulations need high-performance communication networks, because each calculation step requires messages to be passed between processors to communicate force and energy information. As the number of processors used in the simulation increases, so too does the need for faster communication, since simulation performance is extremely sensitive to the speed at which information can be passed between distributed nodes.

Analysis

The analysis of molecular dynamics simulation trajectories also requires high-performance computing methods. To understand the simulation results, scientists must apply compute-intensive methods to analyze large volumes of trajectory data. This analysis requires advanced statistical methods, high-performance computing platforms, and chemists’ expert knowledge to interpret results.

InfiniBand acceleration is critical to molecular dynamics simulations

Microsoft is committed to making the most advanced computing resources available through Azure cloud services. The Azure cloud HPC platform allows researchers to take advantage of InfiniBand on HB Series virtual machines, which enables low-latency communication across distributed nodes for differentiated scalability and performance boosts.

We’ve seen strong results using Azure for molecular dynamics:

  • When simulating a benchmark model for Satellite Tobacco Mosaic Virus (STMV) with 20 million atoms, our latest high-bandwidth (HB) VMs (v3) outperformed previous versions of HB VMs by 218 percent to 251 percent, while also reducing the cost per nanosecond by a third (32 percent to 36 percent).
  • And while these simulations scale well on CPUs, they also scale on GPUs—a configuration also supported by Azure. As we continue to bring new hardware and software to Azure HPC, further benchmark details will be published.
  • Overall, these benchmarks illustrate Azure’s continued capability to reduce the time and cost of complex simulation workloads by utilizing state-of-the-art configurations, such as VMs with InfiniBand technology.

Azure’s unique capabilities extend far beyond the life sciences and can be applied to many other high-performance and computing-intensive workflows, including those in materials science and chemical physics.

The importance of AI and future impact of scaled quantum computing

Increasingly, we see great potential to accelerate chemistry and materials advances by integrating Azure’s scaled HPC solutions with the speed of groundbreaking AI models tuned for scientific research. At Microsoft, we have been exploring a full breadth of AI capabilities for decades with our internal research teams. With the broad range of AI tools in Azure, innovators can design workflows which harness AI models to sort through massive data sets and subsequently use HPC-based simulation insights to narrow those results. These scenarios are only possible with the deep integration of AI and HPC in Azure today, which will also include the power of quantum at scale to help researchers improve model accuracy in the future.

Many of the world’s most pressing problems require advanced computing and the ability to simulate complex systems, because many physical interactions and natural processes are too difficult to study with classical computation at sufficient levels of accuracy. For this reason, scaled quantum computers must be part of the architecture of the future. Since quantum mechanics explains the behavior or matter and energy on the smallest possible scale—the scale of atoms and subatomic particles—quantum computers are inherently capable of understanding and predicting the complexities of nature, like those in chemical and materials science.

Scaled quantum computing will deliver breakthrough accuracy in modeling the forces and energies of such systems, allowing insights into spaces that are currently intractable to explore. Microsoft is focused on engineering a scaled quantum machine with these capabilities right now. However, with the world’s future possibly in the balance, we constantly ask the question: “How can we empower scientists to accelerate progress today?”

CLO17_azure_010

Azure Quantum

Accelerating scientific discovery.

Start accelerating innovation today with Azure

At Microsoft, we continue to invest more in our cloud HPC infrastructure so that innovators can accelerate the pace of research and discovery—both within chemical and materials applications and beyond.

We’re excited to see how the power of the Azure cloud will help you.