I'm going to a meeting about using GPUs as this is something in which I've been interested for some time now: http://events.thalesians.com/ . I'd be interested in hearing other concerns people have about the general applicability of GPUs to programming problems.
Here's the abstract of the talk to be given: This talk will discuss the application of Monte Carlo methods to a GPU implementation of the LIBOR Market/BGM Model. In the process we will discuss random number generation and inverse normal distribution functions designed for execution on the GPU. We will demonstrate how some level of abstraction can be implemented to obtain hardware agnostic code. We will also briefly discuss the limitations of the GPU for the BGM model and point at opportunities for the use of GPUs in the credit analytics space. Resources, including open source code focused on BGM implementation in CUDA, will be identified and used for illustration purposes to help jumpstart the GPU development work. SPEAKER Peter heads the Rates Group at Quantifi. As Director, Peter is responsible for managing the product development process of all Rates and Hybrid Solutions within the Quantifi product suite. Peter started in Research and Technology at Bear Stearns and Deutsche Bank. He traded fixed income derivatives, government bonds and agencies for Lehman Brothers and Salomon Brothers. He was responsible for fixed income derivatives trading desk for a number of European banks. Most recently he refocused on technology and specifically concentrated on machine learning and high frequency trading on parallel systems prior to joining Quantifi in 2009. On Tue, Mar 1, 2011 at 4:52 PM, Raul Miller <[email protected]> wrote: > On Tue, Mar 1, 2011 at 4:36 PM, Alex Gian <[email protected]> > wrote: > > Thing is, with these modern GPU cards, usually on a PCI-Express bus, is > > that you can have as many as you like... > > > > Also it's not OpenGL that you'll be using to implement anything, more > > likely OpenCL (looking good currently, very C-like) > > http://www.khronos.org/opencl/ > > or NVIDIA's CUDA. > > The thing that bothers me about CUDA is that not only is it locked > into one vendor, but that you apparently have to know which card you > are programming against to get anything meaningful done (though that > might have changed since I last tried to figure it out, or maybe my > initial impression was wrong). > > But OpenCL looks promising. > > -- > Raul > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm > -- Devon McCormick, CFA ^me^ at acm. org is my preferred e-mail ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
