About a-team Marketing Services
The knowledge platform for the financial technology industry
The knowledge platform for the financial technology industry

A-Team Insight Blogs

NovaSparks Adds Data Feeds; Outlines FPGA Matrix Architecture; Details Performance

Subscribe to our newsletter

NovaSparks has added support for an additional five data feeds to its Gen2 appliance, while at the same time outlining the FPGA matrix architecture that the appliance implements in order to handle multiple feeds and application functionality, and detailing its latency profile.

The new data feed support comprises feeds from Bats (BYX and BZX), Direct Edge (NG-A and NG-X), NYSE Arca, the London Stock Exchange (Level 2) and Turquoise (Level 2). Support for Nasdaq, CME and Eurex was already available. NovaSparks says the choice of the added data feeds was driven by customers, though it is currently focusing on providing support for equities and futures exchanges in Europe and North America.

Gen2 is a 2U rackable appliance based on FPGAs from Altera – up to three per appliance. Multiple appliances can be inter-connected to scale up the matrix to support multiple data feeds and trading functions.

The company has released data detailing the latency profile of its matrix architecture. “Our FPGA Market Data Matrix Feed Handlers process the cash equity feeds in less than 700 nanoseconds even during micro bursts and message volumes peaks,” says NovaSparks CEO Yves Charles.

The chart below details the latency observed when a full day of ITCH message data – some 280 million packets – is replayed in just eight minutes.

 

NovaSparks attributes the consistency of latency under load to the matrix architecture’s  implementation of a full set of functions in multiple FPGAs, as opposed to other approaches where FPGAs are used simply for basic acceleration, such as network protocol handling and data transformation, while most of the processing is performed on a traditional CPU.

In the NovaSparks architecture, FPGAs are used for UDP, IP and network handling, message parsing, symbol filtering, book building and fan out to downstream applications. Yves says that the company will this month roll out a proof of concept implementation running trading algorithms on FPGAs within the matrix.

Subscribe to our newsletter

Related content

WEBINAR

Upcoming Webinar: Data platform modernisation: Best practice approaches for unifying data, real time data and automated processing

Date: 17 March 2026 Time: 10:00am ET / 3:00pm London / 4:00pm CET Duration: 50 minutes Financial institutions are evolving their data platform modernisation programmes, moving beyond data-for-cloud capabilities and increasingly towards artificial intelligence-readiness. This has shifted the data management focus in the direction of data unification, real-time delivery and automated governance. The drivers of...

BLOG

Bloomberg BQuant Wins A-Team AICM Best AI Solution for Historical Data Analysis Award

When global markets were roiled by the announcement of massive US trade tariffs, Bloomberg saw the amount of financial and other data that runs through its systems surge to 600 billion data points, almost double the 400 billion it manages on an average day. “These were just mind-blowingly large volumes of data,” says James Jarvis,...

EVENT

AI in Capital Markets Summit London

Now in its 3rd year, the AI in Capital Markets Summit returns with a focus on the practicalities of onboarding AI enterprise wide for business value creation. Whilst AI offers huge potential to revolutionise capital markets operations many are struggling to move beyond pilot phase to generate substantial value from AI.

GUIDE

Pricing and Valuations

This special report accompanies a webinar we held a webinar on the popular topic of Pricing and Valuations, discussing issues such as transparency of pricing and how to ensure data quality. You can register here to get immediate access to the Special Report.