The YASARA Benchmarks
Since the first YASARA website went online in 1999, we have been providing benchmarks to allow an objective comparison with other programs. Our goal is however not to start 'benchmark wars' with other groups, as we are all trying to solve the same problems. So instead of comparing other programs with YASARA directly, we offer symbolic rewards of 100$ for the first one who can beat YASARA in one of the following benchmarks. You do not have to use your own program, just any software you find.
involve the enzyme dihydrofolate
reductase (DHFR), which is commonly used for this purpose, a PDB file
of the system
with 23786 atoms can be downloaded here.
The simulation is run with PME (grid spacing <1 A, 4th order
B-splines), 8.0 A
cutoff for Van der Waals and real-space Coulomb forces, correct atom
masses, reproducible trajectory (running the simulation again
gives the same results), Intel turbo boost disabled, and 8 AVX
registers. More details about recommended CPUs and
GPUs are available, the
simulation algorithms are described in:
The following restrictions apply:
Benchmark 1: 100$ or 1000$ for faster MD simulations on the CPUOn an Intel Core i7 5960X CPU running at 3.6 GHz (no turbo boost), without GPU, YASARA simulates DHFR with 160 ns/day. 100$ are yours if you can do it faster, or even 1000$ if you can do that with the same compiler options (GCC 4.8, -O3 -fno-strict-aliasing -march=core-avx2 -mavx2 -mfpmath=sse -ffast-math -m32). A YASARA power user reported 199 ns/day with the 5960X CPU clocked at 4.4 GHz.
Benchmark 2: 100$ for faster MD simulations on AMD Radeon GPUsOn an Intel Core i7 4770 CPU running at 3.4 GHz (no turbo boost), with an AMD Radeon R9-290X GPU, YASARA simulates DHFR with 150 ns/day. 100$ are yours if you can do it faster.
Benchmark 3: 100$
for faster MD simulations on nVIDIA GPUs with correct atom masses
an Intel Core i7 5960X CPU running at 3.6 GHz
(no turbo boost), with a Geforce GTX 980 GPU, YASARA simulates DHFR
with 253 ns/day. For reasons explained here, YASARA uses
the OpenCL industry standard to program the GPU. nVIDIA's OpenCL
implementation is currently still lacking some functions that are
available in its proprietary CUDA language, which gives CUDA-based
programs an advantage, making it a very tight race. So to avoid a 100$
payment, we added the restriction that the simulation must be run with
correct atom masses. A YASARA power user reported
290 ns/day with the 5960X CPU clocked at 4.4 GHz.