FPGA-based Market Data Processing

What is nxFeed ?

nxFeed is a feed handler which streamlines market data application development by processing data feeds on FPGA and making them available to applications via a simple, normalized API.


Designed to complement trading applications or in-house ticker plant development, nxFeed will arbitrate, decode, normalize and build order books on a plug-n-play, FPGA-based NIC — significantly reducing application processing resources and overall latency. This allows developers to focus on what they do best – build core business logic.


nxFeed reduces market data processing latency and server footprint compared to software-based solutions while also providing the flexibility for deployment as an in-application architecture using PCIe or distributed via UDP multicast over Ethernet.


Who should use nxFeed?

Frequently used by electronic trading applications or normalized data processing systems — nxFeed has the flexibility and power to process raw market data from even the most volatile exchange feeds.


Designed to be a tool in any market data developers toolbox nxFeed is easy to integrate with new or existing applications. nxFeed provides data processing with plug-and-play, FPGA-enabled market data handling and a simple C++ API.


When used for algorithmic trading, integrating nxFeed is often a first step towards end-to-end FPGA-based trading strategies using nxAccess, the Enyx execution engine with an embedded FPGA algorithm sandbox and accelerated execution capabilities.


How is nxFeed unique?


With constant, deterministic, and jitter-free performance, nxFeed filters unnecessary data and increases overall performance by reducing the network and server load. Whether publishing updates to a local application via PCIe or over the network to remote applications — nxFeed guarantees consistent and transparent performance.


Check out our performance reports for more information.




With software simulators available to aid the development cycle — the path to integrating with the nxFeed API is simple and can be achieved in as little as a week. There is no requirement to install dedicated FPGA hardware, allowing firms to maximize time-to-production for their applications.


With an easy-to-use API, nxFeed offers features that were typically reserved only for software solutions:


  • TCP-based remote application resynchronization
  • Centralized dictionary management
  • Centralized application administration & management
  • Live monitoring counter & latency statistics



By taking advantage of the portability of FPGA-based NIC, nxFeed can be deployed alongside your application on the same server, or as a feed handler serving data to many remote applications.


nxFeed is designed to easily scale from single server to multi-site trading infrastructures by including two types of access via a single API:


  • PCIe local delivery for an application running on the host server
  • UDP Multicast over Ethernet distribution for up to 8 different normalized feeds

Performance reports


<1.2 µs*


<8 µs*

consult our latency reports

*SOP-to-SOP latency measured on the switch

nxFeed: Use cases

» In-house tickerplant


  • Internal data distribution without having to normalize and build books for each exchange format
  • Single, simple, normalized API delivered to existing software application stacks


» ULL trading application


  • Delivery of market data updates over PCIe to application via C++ API
  • Lowest latency configuration
  • Initial step towards nxAccess — an FPGA-based trading engine



» Market data vendor distribution


  • Leverage nxFeed to normalize data and distribute only what is required over network locations
  • Simple, consistent client API for receipt of data



» Service provider: value added data feed


  • Use nxFeed to generate proprietary data feeds from market data sources
  • One consistent model across all exchanges
  • Typical deployments include consolidated order books or SIP generation


Supported Markets

view all venues


– Average Latency < 1.2 μs
– Maximum Latency < 8 μs
– Between 16-250k symbols per card (See: Global Market Coverage)
– By-Message A/B feed arbitration
– Book-based & order-based normalized market data
– Symbols / message type / book depth filtering
– Exchange resynchronization
– Normalized protocol across venues
– Comprehensive APIs for configuration & symbology management
– Up to 8 UDP multi-cast outputs with different publication configurations
– TCP-based consumer resynchronization
– Live monitoring counter & latency statistics