For example, there are more than 100 exchange-traded options linked to Microsoft's stock; each time the stock is traded, those values need to be recalculated. "That is a hugely intensive computation," relates Daniel Marques, CTO at Chicago-based Ballista Securities, which launched Ballista ATS, an alternative trading system for block options trades, in October. Marques explains that each option is tied to a complex formula that measures the option's sensitivity to change in the underlying stock. These options analytics, often referred to as Greeks, are critical in calculating options values.
Currently there are approximately 350,000 options contracts that trade on about 3,500 stocks. And the Options Price Reporting Authority (OPRA) has told the industry to prepare for 1 million messages per second, which is roughly 10 times as much message traffic as in the equity markets, according to Gerald Hanweck, founder and principal partner at Hanweck Associates, a quantitative financial consulting firm in lower Manhattan.
Because of the skyrocketing amount of data and the complexity of options calculations, derivatives pricing is hitting a wall in terms of processing power, says Hanweck, who is a former chief equity derivatives strategist at J.P. Morgan. In addition to latency concerns, the required power consumption and server rack prices are straining Wall Street data centers, he adds.
To boost their computational power, Wall Street firms and third-party software developers in the options and derivatives space are turning to graphics processing units, or GPUs, from NVIDIA, a Santa Clara, Calif.-based chip maker that gained prominence for its support of 3-D graphics in video games. While NVIDIA's GeForce graphics cards are ubiquitous in the consumer and entertainment computing markets, the company's Tesla professional workstations, which feature hundreds of processor cores to support parallel processing, have gained momentum in the high-performance computing market, finding their way into molecular dynamics, neuron simulation, MRI processing and atmospheric cloud simulation, as well as derivatives pricing and Monte Carlo simulations.
GPUs: Game On
"The natural evolution of that chip has turned it into a highly programmable and very powerful processor [able] to handle large, data-intensive number-crunching problems," says John Milner, director of business development for NVIDIA's GPU Computing Group, who notes that the computer industry is going through a transition. "There isn't enough speed coming from the CPU," he continues, adding that chip makers are moving to multicore servers that support parallel processing in order to improve performance.