JPMorgan lifts the lid on GPU-led performance breakthrough

JPMorgan lifts the lid on GPU-led performance breakthrough

JPMorgan is taking advantage of graphic processors from Nvidia to deliver a 40% increase in the speed of its risk calculations, as part of a three-year plan to cut computation data costs by 75%.

So far, more than half the equity derivative-focused risk computations run by the bank have been moved from running on just CPU-based systems, to running on hybrid GPU/CPU-based systems.

Utilising Nvidia's Tesla GPUs as companion processors has accelerated application performance by 40X compared to running them on CPUs alone, says the bank, and delivered over 80% savings by providing higher performance for the same power.

The roll-out is part of a five-year plan instituted by the bank in 2009 to streamline the number of trading platforms it deploys from ten to two with the aim of generating $300 million in cost savings per year. The lower costs and improved efficiency generated by the effort are expected to reap benefits of $1 billion.

With the new Nvidia processors deployed across multiple data centres and shared between applications across the bank's global network, JPMorgan is now able to calculate risk across a range of products in a matter of minutes rather than overnight.

In a statement, Nvidia says: "Even the longest-running exotics instruments can be calculated on an as-needed basis, faster than ever before, enabling more frequent runs and more complex scenario calculations."

JPMorgan and Nvidia are not alone in pushing the technology, with other top tier banks and tech outfits such as Advanced Micro Devices exploring the advances enabled by GPUs and other accelerators such as field programmable gate arrays (FPGA).

Comments: (0)