SciFinance: the advantage for derivatives pricing model development

 

Automatic Parallel Code Generation

NVIDIA CUDA

Given their inherent parallelism, many financial engineering problems are tailor-made for parallel computing technologies. Other implementation approaches include hand-coding of individual pricing models or enabling select pricing model libraries for parallel computing. SciFinance provides a unique, elegant and highly efficient alternative to this labor-intensive approach.

Other parallel computing approaches

Hand-coding pricing models

Hand programming a complex pricing model for a parallel environment requires significant parallel computing programming expertise and nearly always a complete redesign of the implementation. Parallel programming is time consuming, typically requiring trial and error, detailed profiling (when such tools exist) and is notoriously difficult to debug.

Enabling select pricing model libraries for parallel computing

Organizations with existing pricing model libraries may consider reprogramming select pricing model library components, for example, a Monte Carlo simulation engine. Part of the appeal of such a solution is that it appears to be purely an IT department undertaking with minimal or no involvement of quantitative development groups. Unfortunately this is seldom the case, as existing implementations must be redesigned and interface changes ripple through other components and existing libraries.

 

The SciFinance advantage: parallel code without programming

SciFinance builds derivative pricing models as source code from high level specifications. SciGPU, the parallel computing component of SciFinance, automatically generates CUDA source code for any Monte Carlo-based pricing model (runs on NVIDIA GPUs). SciComp Consulting, our highly-skilled and professional quantitative development team provides expert, cost-effective GPU programming and porting services for PDE (partial differential equation) pricing models.

In order to create a GPU-enabled Monte Carlo pricing model you simply add the keyword “CUDA” to any existing serial model specification. SciFinance does the rest. Absolutely no CUDA or parallel programming expertise is required. The resulting GPU-enabled code is restructured to take advantage of available GPU hardware and software features for astounding accelerations. In addition, all Monte Carlo pricing models generated by SciFinance are OpenMP-compliant so that you can take advantage of near linear speedups in multi-processor environments.

SciComp NVIDIA CUDA Tech Brief

>Download the SciComp / NVIDIA Tech Brief to learn more

Typical SciFinance parallel code speedups:

  • CUDA Monte Carlo pricing models:
    • 30X-50X faster than serial code (single GPU, double precision)
  • OpenMP-compliant code executes in the multi-processor environment with nearly linear speed-up
    • 3.9X faster than serial code on a quad-core PC, 22X on a 24 CPU workstation



Video

SciFinance GPU Computing Movie

 

 

Parallel computing case study

 

SciComp / Reval Case Study

>Download the Reval/SciComp Case Study

In order to quickly develop new structured products and vastly speed up Monte Carlo-based derivatives for its Software-as-a-Service (SaaS) customers, Reval uses SciFinance.


 

Need more information on SciFinance and parallel computing? Contact us >>