857.namd_s
Scientific, Structural Biology, Classical Molecular Dynamics Simulation
NAMD was written by Jim Phillips and many others at the Theoretical and Computational Biophysics Group at the Beckman Institute for Advanced Science and Technology at the University of Illinois, Urbana-Champaign.
857.namd_s was submitted to the SPEC CPU v8 Benchmark Search Program by David J. Hardy of the Beckman Institute, <dhardy[at]illinois [at] edu>
NAMD is a widely used biomolecular dynamics application designed to simulate the interactions of DNA and proteins for a variety of use cases, including virology, drug discovery, and structural biology. NAMD was the recipient of the 2020 ACM Gordon Bell Special Prize for development of an AI-driven workflow to model the SARS-CoV-2 spike protein and its interaction with the human receptor ACE2. NAMD is capable of scaling to over 200,000 cores for very large systems, however serial performance is equally important to the 50,000+ users who have downloaded the program over the past decade.
The SPEC CPU NAMD benchmark is derived from the full production NAMD code, with most of the runtime spent calculating inter-atomic interactions in a small set of functions. This key set of functions was separated from the bulk of the code to form a compact benchmark for CPU 2026. The benchmark encompasses twelve different flavors of force calculations, designed to inline as many branches and function calls as possible. This includes intra-patch (self) and inter-patch (pair) calculations, covering both force and energy computations. It also applies multiple time-stepping of long-range electrostatic interactions, categorized as fast only, fast and slow separated, and fast and slow merged. The benchmark uses all of these for variety, and checks that combinations of them add up correctly.
The input dataset is the Tobacco Mosaic Virus (STMV), which is a well studied virus that can infect a wide range of plants, including tobacco and other members of the Solanaceae family (e.g., tomatoes, peppers, potatoes, etc). STMV's structure and behavior at an atomic level is studied for developing strategies to prevent and manage viral diseases in crops.
The file stmv.input is a 1,066,628 atom simulation of Tobacco virtovirus 1 and is derived from the published project Molecular Dynamics of Viruses . It contains the force field parameters, molecular structure and atomic positions necessary to run the STMV simulation. Test, train and refspeed workloads read from the same input file, but run the code for different number of iterations. For refspeed, the code is run for 399 iterations. For training 49 iterations are used to properly weight pairlist creation and usage. OpenMP is enabled for parallelization.
The output contains calculations for the potential energy, net force, and pressure virial of the system for each set of positions in which the forces are evaluated during the run. The output file also contains various checksums on the force calculations, which will be consistent across platforms, within roundoff error. An absolute tolerance of 0.00005 was chosen to allow for floating point accumulation order changes, while still causing the test to fail in the case of compiler optimization errors.
The SPEC CPU 2026 benchmark 857.namd_s differs from 444.namd (SPEC CPU®2006) and 508.namd_r (SPEC CPU®2017) in the following ways:
C++
The benchmark uses OpenMP 3.1 for parallelization.
None known. The benchmark is written in conservative C++, is quite portable, and the inner loop code (module ComputeNonbondedStd.C ) contains no aliasing.
NAMD was contributed and licensed directly to SPEC® by the University of Illinois. This benchmark was derived from the original v2.14 source code, which can be retrieved from the NAMD Downloads Page under "Version 2.14 (2020-08-05) Platforms Source Code". Users must register a username and password first to download the source code.
In terms of open source licensing, NAMD contains an erfc() function for Windows portability that is distributed with the source code. This code is licensed under the SunPro license, described in erfc.license.txt. This file has been kept for reference, however it not needed for most platforms and has been disabled for SPEC.
Copyright © 2026 Standard Performance Evaluation Corporation (SPEC®)