GPU Demo Applications

NAG has produced a suite of GPU demo applications. The intention with the demos is to show not only how to use the NAG GPU routines, but also how to create parameterised GPU kernels that use the NAG GPU components effectively. Parameterising the kernels allows them to be auto-tuned, and for complex GPU kernels, auto-tuning can potentially give large performance gains.

The demos use a simple financial model: a Black-Scholes call option with deterministic term structures of interest and volatility. The model is straightforward enough that it is immediately understandable without being trivial, and does not detract from the main point of the demos, which is to show how to structure efficient, parameterised GPU kernels. The model can be replaced without much alteration to the structure of the kernels.

There are three main demo programs:

1.) A classic pseudorandom Monte Carlo pricer using both the device function and the pre-written kernel random number generators.
2.) A scrambled quasi Monte Carlo pricer using Sobol points and a Brownian bridge. This also produces error estimates for the price.
3.) A quasi Monte Carlo routine which prices several similar call options simultaneously using Sobol points and a Brownian bridge.

All the GPU kernels are auto-tuned at runtime, calculations can be performed in single or double precision, and CPU equivalents of all the GPU kernels are provided so that identical prices can be obtained from the GPU and the CPU.

Sample Output

Sample output from the demos run on two different systems is given below.

1) Machine Specification:
CPU: Intel Xeon E5410 at 2.33GHz
OS: Windows 7 64bit

Demo Sample Output

2) Machine Specification:
CPU: Intel Core i7 860 running at 2.8GHz
OS: Windows 7 64bit

Demo Sample Output

In order to obtain a copy of the NAG GPU demos, please complete the online registration of interest form here or contact us directly for more information.