San Diego Supercomputer Center
In this latest blog on the oneAPI Centers of Excellence and the projects they are working on, I’d like to focus on the San Diego Supercomputer Center (SDSC) at the University of California San Diego (UCSD). Dr. Andreas Goetz is a research scientist at SDSC and a group leader for Data-Driven and High-Performance Computational Chemistry. He and his team focus on enabling high-performance molecular dynamics (MD) simulations in Amber*, a biomolecular simulations software package used by thousands of scientists in academia, national labs, and industry. They are using the open standards-based cross-architecture oneAPI programming model to modernize Amber’s code for portability, making it multiarchitecture-ready, and ensuring that molecular dynamics simulations can take full advantage of the power of upcoming graphics processing units (GPUs). [1]
Andreas and his team also lead workshops and tutorials on the code migration of molecular dynamics simulations from CUDA* to SYCL*.
The current focus is on achieving performance portability and functional completeness, so the research community can take full advantage of oneAPIs vision of a single source codebase that developers can use to gain flexibility in targeting the best hardware for solutions rather than being confined to a single-vendor proprietary architecture.
Having this available in a ready-to-use code package as part of The Amber Project is the immediate goal of Andreas’ team and their close collaboration with application engineers at Intel.
Amber Biomolecular Simulation
Amber consists of two major parts:
- a comprehensive package of molecular simulation programs with source code, examples, revision, and continuous integration control.
- a set of molecular mechanical force fields for the simulation of biomolecules.
The simulation programs cover a diverse set of functionality including, but not limited to:
- Membrane Systems
- Ionic Liquids
- Material Systems such as Polyethylene Terephthalate (PET) Polymers
- Pharmaceutical Compound Force Field Simulation
- System Stabilization and Relaxation
- Thermodynamics
- Trajectory Analysis
- Free Energy Calculations
- Chemical Reactions and Equilibria
The molecular force fields covered include:
- Protein force fields
- Nucleic acid force fields
- Carbohydrate force fields
- Lipid force fields
- Ligands and Cofactors force fields
- Force fields for ions
- Force fields pantetheine-containing ligands (PCLs)
- Coarse Grain Models
- Fluorescent Dyes force fields
- Modified Amino Acid force fields
- Contributed Parameters
The Computational Challenge of Understanding Molecular Dynamics
The use of molecular dynamics simulation in science, materials research, and pharmaceutical pathfinding has exploded over the last two decades.
The underlying calculations for these types of simulations are extremely compute-intensive, which is why the wide adoption of computational molecular simulation has only recently taken off.
SYCL and oneAPI help democratize offload compute and increase accessibility to compute acceleration by giving developers the flexibility to target the best hardware for solutions rather than being confined to a single-vendor proprietary architecture, all with a common code base.
To gain insight into the challenge molecular mechanics pose for computational requirements, let us look at the nature of the problem Amber and thus Andreas and team are tackling.
At the heart of molecular dynamics is understanding the interactions between all the different atoms and bonds within a molecular system, whether within a single molecule or a system of interacting molecules. These interactions are expressed by Schrödinger Equations. Differential wave equations as such do, however not easily lend themselves to a computational approach.
The Born-Oppenheimer approximation states that due to the difference in relative mass between nuclei and electrons, their wave functions can be treated separately, and the spatial position of the nuclei can be assumed as relatively fixed. This allows for the energies and, thus the force fields present in a molecular system to be expressed as the sum of individual forces.
For the purposes of computational simulation, they are therefore described as a sum of force fields acting on said molecules:
Figure 1: Molecular Mechanics Force Fields
This equation adds up the bonding strength between two atoms, the angle bending due to the proximity of bonded atoms, and the torsion applied to a bond. The last expression in this equation calculates the sum of all the additional non-bonded interactions with atoms that are not part of the immediate bond under consideration.
Figure 2: Simplified Molecular Interaction Model
If we look at molecular interactions in the real world, we notice that bond vibrations occur on a femtosecond (10^-15 sec) time scale. The resulting folding process , however, leads to a full 3-dimensional conformation on a microsecond (10^-6 sec) or even millisecond (10^-3 sec) scale.
In other words, modeling a single molecular dynamics interaction in a useful way requires calculation steps representing the femtosecond scale resulting in the requirement to perform between 10^9 and 10^15 calculations in a reasonable time frame.
This is where high-performance accelerated offload compute, and the use of GPUs, in addition to CPUs and other hardware accelerators, comes to the rescue. This is where having the choice of an open multi-architecture approach drives innovation forward.
Porting Amber to SYCL
In May 2023, San Diego Supercomputer Center held a comprehensive workshop on the benefits of oneAPI and SYCL and the fundamentals of porting CUDA code to C++ with SYCL.
I highly recommend checking out the full recording available on YouTube for additional background and to get started on your journey.
Porting Amber to SYCL: A Brief Overview San Diego Supercomputer Center oneAPI Workshop, May 15, 2023
|
Performance and Portability
Andreas Goetz and his colleagues at the oneAPI Center of Excellence at the San Diego Super Computer Center are aggressively pursuing and completing their ambitious endeavor of enabling the Amber code base with SYCL ensuring not only code portability but also performance portability, readying this essential computational research tool and pharmaceutical design tool for the new age of ubiquitous heterogeneous compute - free from vendor lock and ready to be deployed on the compute platform best suited for the task at hand.
Bringing Choice to Accelerated Compute
Become part of the effort to make high-performance cross-architecture compute transparent, portable, and flexible. Include SYCL as the accelerator and GPU offload solution in your code path. Adopt oneAPI as the means to implementations free from vendor lock.
Get started with The Amber Project, Amber GitHub, and the Intel® DPC++ Compatibility Tool.
Please stay tuned for next week’s oneAPI project focus.
SYCL Resources
- Academic oneAPI Centers of Excellence
- Essentials of SYCL*
- CUDA* to SYCL* Catalog
- Migrate from CUDA* to C++ with SYCL*
- Workflow for a CUDA* to SYCL* Migration
- An Awesome List of oneAPI Projects
- oneAPI Centers of Excellence: SYCL* Adoption Ready for You
- SYCLomatic
Amber Resources
References
[1] UC San Diego oneAPI Center of Excellence to Bring High-performance Simulations to Amber
Notices and Disclaimers
Performance varies by use, configuration, and other factors. Learn more at www.Intel.com/PerformanceIndex. Results may vary.
Performance results are based on testing as of dates shown in configurations and may not reflect all publicly available updates. No product or component can be absolutely secure.
Your costs and results may vary.
Intel technologies may require enabled hardware, software or service activation.
Intel does not control or audit third-party data. You should consult other sources to evaluate accuracy.
© Intel Corporation. Intel, the Intel logo, and other Intel marks are trademarks of Intel Corporation or its subsidiaries.
*Other names and brands may be claimed
You must be a registered user to add a comment. If you've already registered, sign in. Otherwise, register and sign in.