Mizuho–DL Financial Technology Co., Ltd.Mizuho–DL Financial Technology Co., Ltd.

One MIZUHO

GPU Computing

We work on the GPU Computing based on the financial technologies and computation skills which we have developed through many years of research and development. We earned patents to generate random numbers on a GPU both in Japan and the US. (JP:5059928, US:8786617)

GPU Computing

GPU (Graphics Processing Unit) is a special processor dedicated to displaying computer graphics on a pc. Recently, because of high computational performance, researchers started using GPUs for running general purpose applications. This is called GPU computing or GPGPU (General Purpose Computation on GPUs).
Lately, there are some programming languages and hardwares suitable for GPU computing, it is easy to get the computational performance of GPU.

Advantage of GPU Computing

Calculation runs faster

Some applications greatly run faster in GPU compared with CPU. In the case of Monte Carlo Method, we obtained 100 times or more speed–up, and can bring a decrease in the computing time or the improvement of the accuracy of calculation.

Reduction in hardware cost and operation cost

The GPU computing can greatly reduce the hardware cost and operation cost compared with the case to obtain an equal computation performance with a lot of CPUs.

Our Approach

Application of financial engineerings to GPU computing

We work on the application of valuation and risk analysis of derivatives, VaR calculations or simulations on Risk Management to GPU computing, and achieved great speed–up at the computing time.

Example to valuation of derivatives

Underlying Derivatives Model Computation Method Ratio of Calc Time
(CPU/GPU)
Equity American option Local Volatility (Dupire) Finite Difference About 70 times
Currency Barrier option Stochastic Volatility(Heston) Monte Carlo About 300 times
Interest Exotic product including TARN SABR–LMM Monte Carlo About 150 times

Development GPU–based framework for Monte Carlo Method

To apply the GPU computing to Monte Carlo Method, we develop the GPU–based framework.

Image:Development GPU-based framework for Monte Carlo Method

In the framework, it separates to the part of peculiar processing to GPU, and the part of not peculiar processing to GPU that named modeling and simulations. We become possible an efficient, prompt development of various Monte Carlo Methods by this separation. We support your financial business by only our solution based on the financial technologies which we have developed through many years of research and development, and on computation skills including GPU computing.

Top of Page