The new SmartNIC fb4CGg3 series is optimized for financial trading applications to utilize the performance, scalability and power efficiency of the new Ultrascale+ FPGA
COPENHAGEN, Denmark – September 14, 2016 – Silicom Fiberblaze, a leader in FPGA acceleration of high-speed networks and applications, today announced the SmartNIC fb4CGg3, the first FPGA-based adapters optimized for financial trading applications to use the Xilinx Ultrascale+.
The fb4CGg3 series adapters is supported by the SmartNIC development platform. This combination provides trading firms with the highest performing, most scalable application platform available in the financial services industry. Utilizing the new Ultrascale+ VU9P from Xilinx, this combination provides substantial advantages, including:
- 2.5 million logic cells, nearly 2x larger than Fiberblaze’s previous generation, more easily handling complex trading algorithms
- 382Mb of on-chip memory, nearly 4x larger than the previous generation, allowing applications to support more symbols and run more efficiently
- Up to 8GB of external DDR4 memory
- 4 x QSFP28 to support up to 4 ports of 40G or 100G, and up to 16 ports of 10G or 25G
- 20% More power efficient compared to earlier generations
- Easy porting of existing firmware modules for e.g. trading logic
- Plug and play exchange of time critical modules such as MAC, DMA and memory controllers
Armed with the Ultrascale+ FPGA and the SmartNIC application developers can handle more of the complex trading stack directly on the adapter without requiring data to be transferred across the PCI bus, saving valuable latency and CPU cycles. The fb4CGg3 with SmartNIC can easily handle critical processes of electronic trading, including: symbol decode, filtering, and normalization; building order books; the trading algorithm; risk management; and, order execution.
The fb4CGg3 with SmartNIC includes complete NIC functionality, including TCP Offload Engine (TOE), UDP Offload Engine, and drivers. Customers can leverage the Fiberblaze API to add their own custom logic to the data path, providing a complete high performance platform for trading applications. Alternatively, customers can utilize the entire FPGA for their own custom logic and benefit from the high performance hardware building blocks that Fiberblaze provides.
“The fb4CGg3 with SmartNIC the Ultrascale+ FPGA processes applications faster, with greater scalability thanks to the substantial on-chip memory, and with lower power consumption,” said Henrik Lilja, general manager of Silicom Fiberblaze. “It also enables the application stack to be run entirely on the card without crossing the PCI bus, providing substantial latency and overall performance benefits.”
Fiberblaze also provides SmartNIC compatible optimized IP building blocks that save customers considerable time and investment during development, including:
- The ultra-fast FastMAC, for industry leading latency performance.
- ETLM TCP Offload Engine, which implements a complete protocol stack, providing ultra-low latency, zero latency jitter performance.
- UDP Offload Engine, which implements a complete protocol stack, providing ultra-low latency, zero jitter, line-rate performance for market data feeds.
- Protocol Parser provides ultra-fast jitter-free decoding, filtering and normalization of market data feeds for most U.S. exchanges.
The first fb4CGg3 are now available with a number of different Ultrascale options. The first engineering samples with Ultrascale+ are available as samples now, and will be available in quantities in June 2017.