AMD Virtex UltraScale+ HBM high performance FPGA®

Based on the UltraScale architecture, the latest Virtex® UltraScale+ devices provide the highest performance, including the highest signal processing bandwidth at more than 20 TeraMACs of DSP compute performance.

They also deliver the highest on-chip memory density with up to 378 Mb of total on-chip integrated memory, plus up to 8 GB of HBM Gen2 integrated in-package for 460 GB/s of memory bandwidth. Virtex UltraScale+ devices deliver significant capabilities with integrated IP for PCI Express, Interlaken, 100 G Ethernet with FEC, and Cache Coherent Interconnect for Accelerators (CCIX).

Features

• Internal SRAM Bandwidth = 30 TB/s
• HBM2 Memory Bandwidth = 460 GB/s and DDR Memory Bandwidth = 38 GB/s
• 24.5 Tera OP/s DSP compute performance
• Look-Up Tables = 1,079 K PCI Express Gen4 × 8 with CCIX

Additional features

• Integrated 100 G Ethernet MAC with KR4-FEC and 150G Interlaken cores
• Over 2× system-level performance per watt over Virtex-7 FPGAs
• Up to four speed-grade improvement with high utilization
• Up to 128-33 G transceivers deliver 8.4 Tb of serial bandwidth
• 58 G PAM4 transceivers with KP4-FEC enable data transmission at 50 G+ line rates
• 460 GB/s HBM bandwidth, and 2,666 Mb/s DDR4 in a mid-speed grade
• A 5:1 card reduction for 1 Tb MuxSAR transponder
• UltraRAM for on-chip memory integration
• VCXO and fractional PLL integration reduces clocking component cost High Performance FPGA with on-board High Bandwidth Memory
• Internal SRAM Bandwidth = 30 TB/s
• HBM2 Memory Bandwidth = 460 GB/s and DDR Memory Bandwidth = 38 GB/s
• 24.5 Tera OP/s DSP compute performance
• Look-Up Tables = 1,079 K PCI Express Gen4 × 8 with CCIX
• Up to 60 % lower power vs. 7 series FPGAs
• Voltage scaling options for performance and power
• Tighter logic cell packing reduces dynamic power
• Seamless footprint migration from 20 nm planar to 16 nm FinFET+
• Co-optimized with Vivado Design Suite for rapid design closure
• SmartConnect technology for intelligent IP integration
• Of very high value are the IPs:
• TCAM SmartCoreTM , a ternary, content-addressable memory
• Random Forest, provides acceleration, e.g. for implementing machine learning algorithms

Applications

• AI
• Wired Comms HBM as a packet buffer (400G+)
• Data Center Storage Acceleration
• Smart NIC
• Machine Learning
• A&D Any number of radar and electronic warfare applications
• Algorithmic TCAM Random Forrest