Intel Agilex 7 FPGAs

CPUs or GPUs are usually the primary targets of workloads in high performance computing (HPC)

This case study details the application, tool flow, and upgrade to Intel Agilex 7 FPGA technology

BSC is noted for its heterogeneous data center architecture with CPUs, GPUs, and FPGAs with world-class computing resources

FPGAs execute algorithms on proprietary hardware, while software developers may use high-level tool pipelines like oneAPI to access FPGA performance

On the Intel PACs, they computed the computational spectra using the OpenCL code that was already in place

Initial Intel PAC findings utilizing the Intel Arria 10 GX FPGA reduced kernel execution time from 10K seconds to 540.457 seconds

FPGA architecture, BSC turned a 64-bit double-precision floating point accumulator into a 40-bit integer data type with acceptable accuracy loss