Sunday, July 07, 2013

Parallel Programming for Quantitative Finance

Investment banks like calculations that require powerful computational resources. In many cases Monte-Carlo simulations are run on huge GRID systems that cost a lot.
Such systems are usually home grown and really look expensive to replicate elsewhere.

Few monthes ago I read an article in RISK magazine that outlined the same problem and as an option a  chipper and more flexible approach was mentioned that is based on multi-core CPU and GPU.

GPU programming probably is not too complex although it definitely requires some background desk devs or quants might not have. A company Xcelerit made an attempt to ease the parallel programming. They provide SDK that allows quants develop and execute their C++ programs on a high-performance environment.

Furthermore they have a library that comes with base statistics functions, market data adapters and a number of interfaces for commonly used software packages, e.g. MATLAB, Excel.

They also outlined some case studies of inefficiencies of sequential program execution on a GRID compared to multi-core CPU architectures.

Examples of SDK usage

HSBC usage

Let's see whether this attempt of Xcelerit will have a successful continuation.

No comments: