GPU Computing

CT Parallel Random Number Generator [PRNG]

CT Financial Library

The CLUSTERTECH Parallel Random Number Generator (CT-PRNG) is based on Mersenne Twister which has a period of 2¹⁹⁹³⁷-1. It generates multiple independent streams simultaneously across a cluster of CPUs and GPUs. and uses a jump-ahead feature to guarantee the quality of the output.

CT Financial Library
The CLUSTERTECH Financial Library facilitates the pricing of derivatives using Monte Carlo simulation and tree based methods on GPUs. It is easy to use and offers the highest performance in the industry .

¹ Intel® high end Core™ i7 Processor specification ²NVIDIA Tesla technical specification

CT Parallel Random
Number Generator [PRNG]

CT Financial Library in GPU

Sales and Services Support

Hong Kong:	+852 2655 6100
China:	+86 010 8260 8832
(Tollfree)	400 8108832

CLUSTERTECH offers high performance for computationally intensive applications by utilising massively parallel graphics processing units (GPUs). While an Intel i7 processor has a peak performance of approximately 100 GFlops¹, the latest NVIDIA Tesla GPU can achieve over 900 GFlops² in single precision performance. GPUs also have benefits in terms of cost and power consumption compared with CPUs.

CLUSTERTECH offers packaged libraries and consulting using GPUs. We have helped our customers gain a technical advantage over their competitors by implementing GPU solutions in the following industries.

	Banking and Financial Sector
	Interest rate derivative products are highly popular in banks and other financial institutions. However since their maturity can extend over several decades, pricing them can be computational exhaustive. A widely used model is the Brace-Gaterak-Musiela (BGM) model and we have implemented a Monte Carlo simulation-based BGM interest rate paths generator which offers a 18x speedup over a multi-threaded CPU implementation. On the other hand, our tree-based model for pricing American options achieves a speedup of 30x over an optimized CPU implementation.

	Numerical Analytics
	A bottleneck on Monte Carlo simulation is often the problem of generating normally distributed random numbers. Clustertech’s random number generator, CTPRNG, uses the popular Mersenne Twister algorithm for uniform random number generation and can generate 100 million normally distributed random numbers in 0.3s. This constitutes a 30x speed up over a CPU implementation. Click here to learn more about our implementation

	Weather Forecasting
	CLUSTERTECH has developed GPU-optimised rainfall prediction models. An existing model was able to reduce processing time from 20 minutes to less than 6 minutes, an average speedup of 4x, enabling our clients to move their model from an expensive cluster to a single GPU-equipped machine.

	Oil & Gas
	Computational-intensive tasks are common in the oil and gas industry. Analysing a large set of geophysical data to achieve a reliable underground oil estimate involves seismic data processing, noise reduction, reservoir modelling, and solving linear equations. GPUs have allowed geophysicists to apply advanced processing and visualisation techniques on multi-terabyte datasets. We are working with our partners to apply GPU and field programmable gate array (FPGA) technologies to improve the speed, yield and accuracy in seismic processing.
In summary, GPU computing offers Improvements in program execution speed allowing jobs to complete in less time or previously intractable problems to be solved. Lower development cost and shorter implementation time compared to other parallel processing techniques. A reduction in total cost of ownership due to lower power requirements and greater computational density. To explore the application of GPUs to your specific processing needs,please contact us.