The Mercury Computer Systems 910-18016 is a 6U VME Fibre Channel module engineered for high-speed data transfer and storage connectivity within RACE++ multicomputer systems. Built around two Xilinx Virtex-II Pro P70 FPGAs (seven-million gates each), the module delivers 267 MB/s connections to the RACE++ crossbar fabric with minimal latency—75 ns per crossbar during connection establishment, 15 ns during active transfer. Memory support includes 8 MB of QDR II SRAM per FPGA with 6.4 GB/s peak access rates and 128 MB RLDRAM II per FPGA for larger dataset staging. The board provides over 6 GB/s of direct I/O capacity through full-duplex fiber-optic and copper serial connections, plus LVDS parallel I/O. Two onboard FPGAs communicate via ten 2.5 Gbps serial links, while the front panel exposes four 2.5 Gbps copper connections per FPGA for mesh networking. Designed for aerospace, defense, and intelligence applications requiring high-bandwidth, low-latency computing with flexible FPGA-to-PowerPC task partitioning.
Technical Specifications
• Form Factor: 6U VME board
• FPGA Devices: Two Xilinx Virtex-II Pro P70 FPGAs, 7 million gates each
• Memory per FPGA: 8 MB QDR II SRAM (6.4 GB/s peak access), 128 MB RLDRAM II
• RACE++ Fabric Bandwidth: 267 MB/s per FPGA to crossbar
• Crossbar Latency: 75 ns (connection setup), 15 ns (active transfer)
• FPGA-to-FPGA Links: Ten 2.5 Gbps serial connections
• Front Panel I/O per FPGA: Six 2.5 Gbps full-duplex fiber-optic, four 2.5 Gbps full-duplex copper serial
• Total I/O Capacity: Over 6 GB/s direct I/O
• Fiber-Optic Connectivity: Full-duplex serial per FPGA
• Copper Connectivity: Full-duplex serial per FPGA, plus four 2.5 Gbps mesh connections
• Parallel I/O: LVDS lines available
– Key Features
• Dual FPGA compute nodes enable parallel processing and application partitioning
• QDR II SRAM provides deterministic, high-speed data access
• RLDRAM II staging capacity supports large working datasets
• Integrated RACE++ crossbar fabric connection eliminates external switching bottlenecks
• Mixed I/O: fiber-optic for long-distance, copper for mesh, LVDS for parallel logic
• Ten 2.5 Gbps inter-FPGA links maximize throughput for tightly coupled algorithms
– Typical Applications
• Real-time signal processing in radar and sonar systems
• High-speed image processing and video analytics
• Cryptographic and data encryption acceleration
• Scientific computing requiring deterministic latency
– Compatibility & Integration
Integrates directly into Mercury RACE++ multicomputer systems. Compatible with PowerPC processor modules for heterogeneous computing configurations. RACE++ switch fabric provides system-wide scalability with minimal added latency.
