Presentation
Reconfigurable Computing in HPC, ML and Data Analytics
BoF Organizer/Speakers
Event Type
Birds of a Feather
AI/Machine Learning/Deep Learning
HPC Accelerators
Heterogeneous Systems
Post Moore’s Law Computing
TimeWednesday, June 27th8:30am - 9:30am
LocationPikkolo
DescriptionHPC and machine learning are increasingly stretching conventional, von-Neumann style computer architectures to their limits. The need for substantially enhanced energy efficiency, in particular, is driving significant research and development towards including reconfigurable computing devices into systems and exploring their performance potential by applications.
The field of FPGAs, in particular, has made very significant progress, with the integration of CPU cores and highly efficient floating point units, close coupling of FGPAs between each other and towards CPUs, and integration of NIC and compute functions to achieve in-transit data processing.
On the application side, the emergence of Deep Learning has changed the playing field in many areas of data analytics; DL approaches have been shown to run very well on FPGA infrastructures, and the prevalent use of high-level SW environments eases the transition to such platforms. Uptake of FPGAs is also on the rise for processing the data from large scientific instruments, with more and more complex data filtering and triggering handled by reconfigurable computing devices. Finally, FPGAs have been used with good success to run combinations of network stacks and compute payloads, paving the way to effective in-transit computing.
This BoF will continue the sequence of similar events at past ISC conferences and report on latest results in reconfigurable computing technology and actual use of such in the areas mentioned above.
Presenters include John Romein (Astron), Zoran Jaksic (BSC), Niko Neufeld (CERN), Mondrian Nuessle (EXTOLL), Suleyman Demirsoy (Intel), and a person from Xilinx.
The field of FPGAs, in particular, has made very significant progress, with the integration of CPU cores and highly efficient floating point units, close coupling of FGPAs between each other and towards CPUs, and integration of NIC and compute functions to achieve in-transit data processing.
On the application side, the emergence of Deep Learning has changed the playing field in many areas of data analytics; DL approaches have been shown to run very well on FPGA infrastructures, and the prevalent use of high-level SW environments eases the transition to such platforms. Uptake of FPGAs is also on the rise for processing the data from large scientific instruments, with more and more complex data filtering and triggering handled by reconfigurable computing devices. Finally, FPGAs have been used with good success to run combinations of network stacks and compute payloads, paving the way to effective in-transit computing.
This BoF will continue the sequence of similar events at past ISC conferences and report on latest results in reconfigurable computing technology and actual use of such in the areas mentioned above.
Presenters include John Romein (Astron), Zoran Jaksic (BSC), Niko Neufeld (CERN), Mondrian Nuessle (EXTOLL), Suleyman Demirsoy (Intel), and a person from Xilinx.
BoF Organizer/Speakers
Principal Engineer
Lab Director