Nvidia Corp. is showing off the credentials of its artificial intelligence platform in the financial services industry, claiming a 6,000-times performance boost for running an algorithm used by hedge funds to benchmark the testing of trading strategies.

“Backtesting,” as it’s called, is important for financial traders because it allows them to evaluate how new trading algorithms perform on historical data and gauge their effectiveness. Algorithms themselves are critical since they perform around 90% of all public trades, according to the Global Algorithmic Trading Market 2016-2020 report.

So Nvidia’s claims of a massive 6,000-times performance boost for these tasks, such as monitoring automated trading, could have big implications in trading, allowing them to come up with more sophisticated trading models, at a lower cost, and get them up to speed faster than was previously possible.

Nvidia’s benchmark tests were validated by the Securities Technology Analysis Center, which is an organization made up more than 390 leading banks, hedge funds and financial services tech companies.

For the tests, Nvidia compiled a platform made up of 16 V100 graphics processing units housed in a DGX-2 server, running Python libraries made up of the company’s CUDA-X AI software that combines its RAPIDS and Numba machine learning software. RAPIDS is a set of libraries used to simplify GPU acceleration, while Numba allows data scientists to write Python that’s compiled in CUDA, expanding the capabilities of RAPIDS. The entire setup was designed to run multiple tests in parallel with one another.

Nvidia’s tests used the STAC-A3 benchmark suite for backtesting trading algorithms to demonstrate its performance, shattering all previous results by running 20 million simulations on a basket of 50 instruments over a 60-minute period. The previous record was just 3,200 simulations over a 60-minute period.

The ability to use Python is key because it allows data scientists who aren’t experts in CUDA to do the simulations, John Ashley, director of global financial services strategy at Nvidia, said in a press briefing. “This proves how well it works even in Python,” he said.

Nvidia declined to name early adopters of the technology, which often want to keep their methods secret to keep an edge over rivals. But he added, “We believe many of our financial services clients have been doing this for years with CUDA.”

Michel Debich, STAC’s director of research, said the ability to run many simulations on a set of historical data is key for trading and investment firms. “Exploring more combinations of parameters in an algorithm can lead to more optimized models and thus more profitable strategies,” he said in a statement.

With reporting from Robert Hof

Image: Prashanth Raj/Flickr