At IDF14, Altera demonstrated some useful technology for FPGA and OpenCL FPGA developers including: (1) A Stratix® V FPGA driving a 100G optical link, (2) Micron Hybrid-Memory Cube interoperability, (3) OpenCL applications, and a biggie (4) QPI linking to an Intel Xeon Processor. This looks like an interesting platform for real-time OpenCL/FPGA development. The latter highlights the potential for larger FPGA power/compute optimized applications beyond that of the Intel FPGA/Xeon combo chip. CPU-FPGA combinations are in use in search engines to reduce power consumption over either CPUs or GPUs.
100G Optical Solutions Running on High-Performance FPGAs
This demonstration shows high-performance Stratix® V FPGAs driving a 100G CFP2 optical link across 10 kilometers of optical fiber. This technology shows a solution for flexible 100G data centers applications using FPGAs.
FPGA Interoperability with Micron’s Hybrid-Memory Cube
This demonstration shows the interoperability between Stratix V FPGAs and Micron’s Hybrid Memory Cube (HMC). HMC addresses the limitations of conventional memory technologies by providing up to 15 times the bandwidth while using 70 percent less energy and 90 percent less space than existing technologies.
Financial Model Acceleration with FPGAs
This demonstration uses OpenCL to efficiently implement multiple kernels on an FPGA performing a Monte Carlo simulation to price options. Utilizing the OpenCL standard on a massively parallel FPGA architecture provides an effective solution for system acceleration, offering significantly higher performance at much lower power versus GPUs.
QuickPath Interconnect (QPI) Linking an Intel Xeon® Processor to an FPGA
Located in Pactron’s booth in the general exhibit area, this demonstration coherently connects an Intel Xeon processor to a high-performance Stratix V FPGA using the QPI protocol. The FPGA is configured as the home agent and extends co-processing for the Intel processor. FPGAs provide a platform to significantly increase compute performance while reducing system costs and power in high-performance computing and data center applications.