Full Speed and Capacity
Speed Only Hardware Can Deliver
Compared to the best off-the-shelves software solutions, NovaSparks feed-handlers can process market data feeds 2 to 5 times faster on average and 10 to 50 times during market peaks. NovaSparks expects those acceleration factors to be maintained, and even increased in the future, given the power wall preventing CPUs to be used with higher operational frequencies. By comparison, FPGAs are still using moderate operational frequencies and thus will keep outperforming CPUs. In addition, NovaSparks constantly works at lowering the latency of its solutions as demonstrated the diagram below, showing the average historical wire-to-wire latency for the processing of Nasdaq TotalView ITCH feed, including book building.
Capacity Only Software Could Deliver
Processing speed is not very useful if a feed handler can only process a limited number of instruments (symbols) or publish a limited depth in the order books. Historically, FPGA have suffered from heavy restrictions in term of instrument count and order book depth because of the inherent memory limitations imposed by FPGA card architecture. NovaSparks is different. NovaSparks does not rely on off-the shelves FPGA cards. Instead, it designs its own FPGA modules and PCIe cards with both speed and scale in mind. The diagram below shows the historical instrument scaling capability for one FPGA module.
