GPU Computing with Jacket in Automated Trader

John MelonakosBenchmarks, Case Studies Leave a Comment

The Q1 2012 issue of Automated Trader contains an excellent “Mashup!” piece reviewing software for algorithmic trading.  The article provides a wonderful glimpse into the 1-2 month adventure of Andy Webb, Automated Trader’s Founder, and Wrecking Crew building a fast trading platform from several technologies.  We heartily recommend that those of you in financial computing go subscribe to get the full story and access to ongoing developments from these Automated Trader thought leaders!

The full trading platform they built was quite extensive.  The part that caught our eye was the core computational component of the pipeline.  That component involved permuting 1,000 potential pairs with cointegration tests for 350 time windows on each potential pair.

The single core MATLAB® version took 70 minutes to run.  Then, using the Parallel Computing Toolbox (™) on 4 CPU cores, the time reduced to 22 minutes.  Next, using Jacket’s GFOR loop on a single GPU, execution time dropped to 5 minutes.  Finally, using Jacket MGL to run the loop iterations on 3 GPUs, the execution time dropped to 1 minute, 52 seconds.

Software Tool Performance Speedup
MATLAB® on a single core 70 minutes n/a
Parallel Computing Toolbox (™) on 4 cores 22 minutes 3.2X
Jacket on a single GPU 5 minutes 14X
Jacket MGL on 3 GPUs 1 minute, 52 seconds 37.5X


The author’s state that this 37.5X speedup was achieved in “less than half a day.”  We’re always impressed by how quickly Jacket programmers are able to achieve success.

You too can get great acceleration with Jacket and GPUs for financial trading problems.  Learn more about Jacket and get started with a trial by taking the Jacket Tour.

Leave a Reply

Your email address will not be published. Required fields are marked *