JPMorgan is using a Maxeler supercomputer to measure risk in its fixed income operations by assessing tens of thousands of possible market scenarios, constantly examining the time path and structure of the associated risk. With the supercomputer, which went live in December, the investment bank can calculate complex scenarios which used to take hours in just a few minutes. The investment bank was awarded the ‘Most Cutting Edge IT Initiative’ at the prestigious American Financial Technology Awards (AFTA).
Oskar Mencer, CEO and founder of the UK-based Maxeler, said other banks are using supercomputers, but only JPMorgan would talk about its operations.
With the new Maxeler technology, JPMorgan’s trading businesses can now compute orders of magnitude more quickly, making it possible to improve our understanding and control of the profile of our complex trading risk,” said Peter Cherasia, head of markets strategies at JPMorgan. The bank has ordered a second supercomputer which will be equivalent to over 12,000 conventional x86 cores, providing 128 Teraflops of performance.
John Barr, research director for financial markets at the 451 Group said Maxeler’s dataflow technology supports researchers who want to focus on the data and not just the computation.
Image via CrunchBase
“The Maxeler Dataflow Supercomputer enables JPMorgan to minimize its risk and respond more effectively than its competitors to rapidly changing market conditions, particularly the financial turmoil in Europe.” Dataflow is the term Maxeler uses to describe its racks made up of 1U (pizza box) MaxNodes. Each MaxNode is a computing powerhouse with 12 Intel Xeon CPU cores, up to 192GB of RAM for the CPUs and up to 4 MAX3 dataflow compute cards, each with the latest Xilinx Virtex-6 FPGAs.
“The true value of accelerating applications through using alternative computing approaches like data parallel or dataflow comes from enabling new levels of performance, changing the user’s competitive dynamics or unlocking new market opportunities,” said Carl Claunch, vice president and distinguished analyst at Gartner.
Maxeler uses Field-Programmable Gate Array (FPGA) accelerators and maps compute-intensive algorithms directly into parallel FPGA hardware and couples that with a conventional CPU like an Intel processor over a high-speed I/O (Input/Output) bus.
FPGA’s are equally well known for their high speed and difficulty in programming; Maxeler’s deep skills in working with the devices is at the core of its supercomputing success.
Mencer said the company’s skills with FPGAs were developed in research projects at Stanford beginning in 1995 with Michael J Flynn, Maxeler’s chairman and a professor at Stanford University. The company moved to London because the world’s best pool of FPGA experts were coming out of Imperial College London. More than a third of the company’s 80 employees are from Imperial, he added.
Maxeler programs FPGAs in about the time it takes to write software, Mencer said.
“FPGAs are like paper. You can say paper makes it really hard to write a Pulitzer winning article, but it is just a surface. The FPGA doesn’t have anything in it, it’s all blank infrastructure. For over 16 years we have focused on what to put on there and to make it easy to write and as fast as possible to compute. You have FPGAs in your car, in your TV, in your monitors.”
Definitions of supercomputers are unsettled. Mencer said that a cluster doing transactions like Google would be a farm, a cluster or distributed computing.
Comments
Great article, offering insights into the two approaches to accelerating compute speeds, whether via FPGA or GPU.
Curious why the image is an Altera FPGA, when Maxeler uses Xilinx Virtex-6 FPGAs. Disclosure, Xilinx is a client.
Got it — the old FPGA I saw the first time round was the Altera. Stand corrected — thanks. t
Great article! Way to take the complex topic of FPGAs and make examples & analogies that the intelligent layperson can understand.
At Wall Street FPGA (www.WallStreetFPGA.com) we have seen that this is just the beginning. The more areas of financial services one explores, the more uses for FPGAs become apparent. And while the larger banks such as JPMC are leading the charge, there is evidence of exchanges, broker-dealers, buy-side, and other groups researching and using this technology to save energy and remain competitive.
Sometimes FPGAs get lumped into the polarizing category of High Frequency Trading but when considered (as above) for risk checks, saving energy, and really protecting pensions and retirements funds from abrupt market movements, the importance hits a lot closer to home.
The tech side of financial services industry is currently dominated software people who live primarily in the Intel x86 (CPU) world. Their background and training is in a way counter-intuitive to the way FPGAs work not to mention there are territorial issues to deal with. Once these *minor* issues are surpassed, hopefully this technology will have wider acceptance.
-Terry