byteLAKE Collaborates with Xilinx
byteLAKE Collaborates with Xilinx to Boost Performance and Power Efficiency of Computational Fluid Dynamics.
CDF kernels optimized for Xilinx® Alveo™ Datacenter accelerator cards run four-times faster, use 80% less energy.
Frankfurt (GER), 10 June, 2019 – byteLAKE, a Polish company specializing in building AI and HPC software solutions, has created a set of highly optimized Computational Fluid Dynamics (CFD) kernels that leverage Alveo accelerator cards from Xilinx, Inc. to create a high-performance platform for fast and efficient complex engineering analysis.
A range of industries such as automotive, chemical, aerospace, biomedical, power and energy, and construction rely on fast CFD analysis turnaround time as a key part of their design workflow to understand and design how liquid and gas flows interact with surfaces.
The high algorithmic complexity of CFD analysis maps naturally to the adaptable and reconfigurable Alveo platform, designed to dramatically accelerate time to insight on a broad range of challenging compute workloads in the Cloud and on-premise data centers. The adaptability of Alveo accelerator cards, powered by Xilinx UltraScale+ FPGAs, is also an ideal platform for byteLAKE to leverage and take their deep algorithmic expertise to solve critical industry problems.
“Moving from CPU-only compute architectures to heterogeneous configurations is essential to meet the ever-increasing demand for accuracy and capabilities of HPC workloads, not only delivering faster performance but also taking into account the energy budget,” said Marcin Rojek, byteLAKE’s co-founder. “FPGA-based accelerators like the Alveo U250 offer an excellent combination of performance and energy efficiency. We achieved almost four-times faster execution of Advection algorithms and up to six-times better performance per Watt, ultimately cutting energy consumption by more than 80% (*).”
“byteLAKE saw the potential to make CFD analysis even faster and more easily accessible leveraging the flexibility and superior parallelism of Alveo accelerator cards,” said Viraj Paropkari, senior manager, Data Center Marketing, Xilinx. “Our teams worked together to create finely tuned kernels for Alveo accelerators that are ready to deploy on-premise or in the cloud.”
The two companies will jointly demonstrate the CFD kernels running on Alveo accelerators during the upcoming ISC High Performance conference in Frankfurt, Germany, at booth D-1055, June 17-20, 2019.
byteLAKE has a long experience in delivering highly optimized CFD algorithms for supercomputers. While working mainly with the MPDATA (Multidimensional Positive Definite Advection Transport Algorithm) and EULAG (Eulerian/semi-Lagrangian fluid solver) models, the team is now bringing the key kernels to the FPGA world. byteLAKE specializes in:
- AI (machine learning, deep learning) – building highly optimized AI engines to analyze text, image, video, time series data,
- HPC (high-performance computing) – building highly optimized applications and kernels for massively parallel, heterogeneous HPC architectures.
PR Contact Details:
The effort makes it possible to import the kernels directly into the geophysical models. All algorithms have been customized for real scientific scenarios and redesigned and optimized for Alveo. Configurations with U250 produced a performance speedup of almost 2x versus high-end CPU-only configurations (**) and up to 4x for other CPU-only configurations (***).
* Energy measurement includes: CPU, Alveo, RAM.
** 2x running Pseudo Velocity on CPU + Alveo U250 vs. Intel ® Xeon™ Gold 6148
*** 4x running Pseudo Velocity on CPU + Alveo U250 vs. Intel ® Xeon™ E5-2995