# Reference request: Survey article on GPU in Finance

I would like to get and idea of how people use GPUs in finance.

I can find some specific papers or books on the subject.
GPUs in binomial model, finite difference, monte carlo,...

But I couldn't find any recent survey papers.

Do you know some recent survey paper?
Or any kind of reference which try to gives a birdview on GPUs in finance?

Labart, Lelong,2011,A Parallel Algorithm for solving BSDEs - Application to the pricing and hedging of American options

Bradley, 2012, State of GPU computing in computational finance
The paper digs deep in current GPU method in QF, but the general survey is very short.

-
I remember reading a similar question in Wilmott forums about a year ago. The answers seemed to say that GPGPU is rarely if ever used in QF. – Oleg2718281828 Jul 23 '12 at 18:50
Nicolas, why don't you add some specific references to the things you have already found? – Ryogi Jul 23 '12 at 20:58
GPU is used for problems that can be solved in parallel. A genetic algo, GRRN network, and some optimization problems can benefit from a GPU (1000x speed-up). – Ram Ahluwalia Jul 24 '12 at 1:38
I don't know of any survey articles but even a cursory look online shows several case studies that could be interesting and show the breadth of uses. Maybe start by looking at nVidia's SDK and case studies, where there are several (simple) examples. – Mark Gray Jul 24 '12 at 10:26
1000x speed-up is marketing hype and typical academic exaggeration (comparing to poor 1-core C code). If you measure the memory bandwidth of RAM and VRAM, you'll see that 1000x is not going to happen for the vast majority of algorithms, even if they are perfectly suitable for GPU. – Oleg2718281828 Jul 25 '12 at 2:47

The Pagès-Wilbertz paper is a very good one. To answer more directly to you underlying question that is: "in which quant finance area to use hardware acceleration?"; the points to take into account are:

• GPU is very good for parallel computations (already underlined in remarks)
• but bad for memory sharing between the master software and the GPU-hosted library
• FPGA is good for parallel computations too (but harder to use compared to GPU)
• FPGA works well for working on data flows

Consequently:

• if you need to solve numerically a PDE with few parameters (coefficients of the derivative terms): GPU is nice;
• if you need to compute on the flight analytics for high frequency trading: FPGA is nice.
-

There are few surveys atm as people are still relatively secretive about it because of the various challenges a production system poses. Actually a major bank even backstepped after some initial efforts. So there is now quite some activity in the field but not so much as the initial hype suggested. You can also try asking in the dedicated Linkedin group.

Edit: I could give you various additional references on specific applications if you want to get an idea of the field anyway.

-
I think your answer is pretty interesting but it lacks a lot of references. What bank are you talking about? What surveys? What is the LinkedIn group? We need links. Then, additional reference are always welcome for latter users and to enhance the overall quality of the site. – SRKX Nov 12 '12 at 23:47
I havent found "official" statements of that bank's decision, so it's just a rumour that would be unsafe to spread further. But it can be asked on the LinkedIn group (which is "GPUs in Finance"), many guys from such bank are reading so I leave them the choice of wether to share their experience on electronic ink. And I know of no good survey, only specific applications (which would be OT here). – Quartz Nov 26 '12 at 15:42