“The system can solve for smaller numbers in under 6 minutes — to use an unquantified but deeply meaningful measure of time, “time to get a coffee”.
NVIDIA has blown away a previous financial trading benchmark used to evaluate trading algorithms and trading strategies, coming in with performance 6,000 times faster in a system stack that was running its NVIDIA DGX-2 server technology.
Financial trading in the modern world is heavily reliant on high speed trading algorithms. According to trading market research compiled by the industry, financial trading algorithms account for nearly 90 percent of all public trading.
The test involved running 20 million simulations on a basket of 50 instruments in 60 minutes. The previous benchmark was set at 3,200 simulations within the allotted time. This represents and achieved acceleration of over 6000 times. Another test running 10,000 simulations on a basket of 48 instruments was completed in six minutes.
The test results were confirmed by the Securities Technology Analysis Center (STAC), which runs benchmark tests on emerging technology using realistic volumes of data and simplified trading algorithms. The platform tested is configured to be used by hedge funds undertaken high speed trading.
NVIDIA commented in the official STAC-A3 report: “NVIDIA has demonstrated that for the baseline backtesting problem in the benchmark, we have essentially “solved” the problem of delivering significantly more instrument simulations than ever before. Our scaling results demonstrate that we can attack larger problems as well while still delivering record breaking performance.”
“This performance is delivered with standard (annotated) Python code leveraging publically available libraries. The scaling results (Table 3 [Below]) show that in addition to delivering total throughput for large numbers of simulations, the system can solve for smaller numbers in under 6 minutes — to use an unquantified but deeply meaningful measure of time, “time to get a coffee”. As our RAPIDS-based library functions expand and increase their acceleration over today’s features, we anticipate improvements in all aspects of scaling and performance.”
NVIDIA DGX-2 Tested by STAC Benchmark Council
STAC and its hardware testing division, the STAC Benchmark Council, consists of over 300 financial institutions and more than 50 technology vendors.
Working together they are establishing standards for technologies used in financial organisations, particularly where algorithms are used for high-speed trading. The tests measures the pace and scalability in which a stack can complete certain jobs.
The STAC-A3 benchmark test aims to give a common basis for quantifying how useful emerging technologies will be for the industry. The STAC-A3 Benchmark test uses repeatable workloads and relevant metrics to test new innovations in memory, processors, analytical libraries, programming tools and storage systems.
At its core the STAC-A3 test is a set of simplified trading algorithms designed to evaluate a stack as if it was been used in real-world trading conditions.
The STAC-A3 report notes that: “The algorithms involved would probably be disastrous if deployed in the market, but the important thing is that they exercise a backtesting architecture in the same ways that production-worthy algorithms would.”
The 6,250 times acceleration achieved in the benchmark test was in the A3.β1.SWEEP.MAX60 benchmark in which STAC ran the test on a stack consisting of a “STAC-A3 Pack for Python with RAPIDS (Rev A)” integrated with Dask, Numba, and cuDF, running on an NVIDIA DGX-2 server containing 16 x V100 (Volta) GPUs, 2 x Intel Xeon Platinum 8168 processors, and 30TB of NVME SSDs for application data.
Michel Debiche, a former Wall Street quantitative analyst and current director of STAC’s analytics research said: “The ability to run many simulations on a given set of historical data is often important to trading and investment firms. Exploring more combinations of parameters in an algorithm can lead to more optimized models and thus more profitable strategies.”