HPC Consulting
We improve the performance of numerical computing programs such as CFD codes on HPC environments
VINAS has been involved in various subjects of R&Ds including performance improvement of parallel programs, development of fast and robust linear solvers, etc.,with a number of national research institutes and universities.
Since linear solvers consume main proportion of computational time in scientific simulations such as CFD, structural analysis, electromagnetic analysis, acoustic analysis, etc.,high stability and high performance are mandatory.
We will improve performance of your scientific simulation programs with our experience and knowledge.
TUNING FOR IMPROVEMENT OF PERFORMANCE
Code tuning for improvement of parallel performance
●Parallelize your numerical analysis programs in CFD, structural analysis, electromagnetic analysis, etc., and realize high speed and large scale computing.
●Improve communication efficiency of large scale parallel computation programs to maximize parallel scalability.
Performance improvement of matrix solvers
●R&D in the field of scientific simulations, such as development of numerical computing routines (linear solvers, etc.)
●Investigate existing numerical computing programs, and improve performance by introducing better algorithms
●Development of special purpose algorithms that satisfy customer's requirements
Consulting on HPC hardware systems
●Design and propose hardware configuration on which user's application runs at best performance, considering hardware specifications such as memory bandwidth, network bandwidth, disk I/O, etc.
PORTING OF IN-HOUSE PROGRAMS ONTO HPC ENVIRONMENT
●Propose HPC environment on which user's program performs fast enough at minimal cost
●Make in-house programs executable on destination environment by replacing or newly creating libraries that are lacking on the destination environment
●Parallelize with the best possible parallelization scheme considering the property of both the in-house program and the hardware configuration
●Tune the program to improve parallel scalability considering memory bandwidth, network bandwidth, disk I/O, etc.
