A recent report from the Securities Technology Analysis Center (STAC) shows for the first time that GPUs boost computing speed such that risk management benchmarks, like STAC-A2, can be operated in real-time, enabling them to be used pre-trade. The testing was demonstrated using STAC-A2, the “Greeks” benchmark for risk analysis. Used in pricing and risk management, the benchmark suite was developed by quants and technologists from some of the world’s leading financial institutions.

“Risk management is a big deal in banking these days,” said Peter Lankford, founder and director of STAC, in a recent interview with Go Parallel. “Banks need to run simulations in order to quantify their risk. STAC-A2 is based on market risk, which is what’s likely to happen to the value a portfolio if the market moves one direction or another. Running these simulations takes a lot of computing power, consequently there is a lot of attention on how much you can do in what unit of time and with what amount of energy. STAC-A2 is a suite of benchmarks based on options Greeks, which tell you the sensitivity of the value of an option based on other kinds of changes, such as changes in interest rates.”

According to results published by STAC, STAC-A2 benchmarks running on NVIDIA graphics processing units (GPUs) returned nearly an order of magnitude speed up compared to traditional x86 CPUs.

For a test machine, STAC used an IBM System X iDataPlex server with two 8-core Intel Xeon E5-2660 @ 2.20GHz (SandyBridge) processors and two NVIDIA K20Xm GPUs. The software stack was coded by NVIDIA using the CUDA 5.5 toolkit.

The GPUs fulfilled their role as accelerators nicely, as STAC reports: “In the mean end-to-end Greeks benchmark (STAC-A2.beta2.GREEKS.TIME), this system was over 9x the average speed of a system with the same class of CPUs but no GPUs and over 6x the average speed of the fastest publicly benchmarked system without GPUs. This system was also the first to handle the baseline problem size in ‘real time’ (less than one second).”

General-purpose GPUs (GPGPUs) are among the computing technologies, along with CPUs and FPGAs, poised to boost the speed, capacity, and accuracy of financial workloads and/or to reduce the cost of computation. This is key in a field like financial analytics that is mostly concerned with speed. Real-time risk management is an undeniable trend and any enabling technology will be put to the test, but ultimately it is for users to make the final decision.