JP Morgan Buys Into FPGA Supercomputing

By Michael Feldman

July 13, 2011

One of the largest financial institutions in the world is using FPGA-based supercomputing for analyzing some of its largest and most complex credit derivative portfolios. JP Morgan, along with Maxeler Technologies, has built and deployed a state-of-the art HPC system capable of number-crunching the company’s collateralized debt obligation (CDO) portfolio in near real-time.

CDOs are instruments in which the credit assets are divided into different bundles or tranches, according to their relative risk of default. During the credit crisis of 2007-2008, CDO valuation tanked as the value of the underlying assets, mostly mortgages, fell off a cliff. Part of the problem was that many of the computer models didn’t assess the risk parameters of the various mortgages correctly. The less obvious aspect was that these instruments were so complex that it was difficult for the models using traditional computer technology to analyze these portfolios effectively.

With the credit crisis in full swing in 2008, Stephen Weston joined JP Morgan’s London office, heading up a team devoted to making the company’s financial algorithms and models run more effectively. In what started out as a blue-sky technology project almost three years ago, Weston’s group has implemented a production-ready solution that speeds up the company’s CDO risk models by a factor of more than 130. “This, to us, is a step change,” said Weston, talking about the project during a presentation at Stanford University in May.

Execution time was the critical factor. Prior to the FPGA solution, JP Morgan’s main risk model for analyzing their CDO portfolio took 8 to 12 hours to complete — an overnight run requiring a cluster of thousands of x86 cores. If the model failed to execute correctly, there was no time to resubmit the application for that day. Worse yet, the credit risks and valuation are in constant flux. That snapshot of the previous day may no longer be useful. “It was a bit like driving your car on the freeway at 90 miles per hour by looking in the rear view mirror,” said Weston. “It could be fun, but there’s a high probability it could be a destructive activity.”

With the speedup, the same risk model took four minutes, with the FPGA processing eating up just 12 seconds of that. It’s not just that they could run the models faster though. The better performance allowed them to run multiple trading/risk scenarios throughout the day. So traders can evaluate more scenarios using different combinations of default criteria. In a nutshell, the time compression allowed JP Morgan to get a better handle on the risk profile of their CDO assets.

In general, porting legacy applications like these financial risk models to FPGAs is no small task. Programming them with low-level VHDL, the traditional programming language of FPGAs, is time-consuming, tedious, and generally unsuited for application developers. Weston knew that it would be a tough sell to convince the quants and management types at the company that this could be a viable solution for a production environment.

In fact, initially JP Morgan looked at GPUs for acceleration. They ported one of their models to the graphics architecture and were able to get a 14- to 15-fold performance boost. But they thought they could do even better with FPGAs. The problem was that it was going to take about 6 months for an initial port. That’s when they went to Maxeler and initiated a proof-of-concept engagement with them.

Maxeler is a London-based technology vendor specializing in FPGA acceleration for high performance computing applications. Unlike most FPGA vendors though, Maxeler offers a vertically integrated solution: hardware, high-level compilers (Java), runtime support, development tools, and FPGA porting expertise. As such, the company is able to meet application programmers on their own turf and help them navigate the eccentricities of FPGA software development. At least, that’s Maxeler’s pitch.

With JP Morgan, it all seemed to work. With Maxeler’s help, Weston’s group was able to port the time-critical, compute-intensive pieces of their C++ risk model (the Copula and Convoluter kernels, in particular) to the FPGA platform in about 3 months. The end result was something Weston felt was sustainable for their production environment.

Part of the effort to port to risk model involved redesigning the original C++ code, which was chock full of templates and objects. Those languages structures are great for application abstraction, said Weston, but they effectively kill parallelism, and thus performance. So the first phase of the code migrations was to remove all uses of classes, templates, and other C++ abstractions that got in the way of parallelization.

With the lower level code exposed, it became much simpler to tease out the parallelism that could be exploited by the FPGAs.  In this case, the flattened C++ source was ported to Java, which the Maxeler compiler is able to convert to VHDL.

Hardware-wise, the final target system is a 40-node hybrid HPC cluster from Maxeler. Each node houses eight Xeon cores hooked up to two Xilinx Virtex-5 (SX240T) FPGAs via PCIe links. Memory is split between the CPU (24GB) and the two FPGAs (12 GB each). Two terabytes of hard disk storage are hung off an Ethernet connection.

The advantage of the FPGA is that it is built for parallelism and allow the application to be intimately mapped onto the hardware. The devices are especially suited to applications that can exploit fine-grained parallelism and very deep pipelines. Unlike linear computations on fast CPUs (~2.6 GHz), parallel computation on slower FPGAs (~200 MHz) can yield many more calculations per watt. As Weston put it, “We went from computing in time to computing in space.”

Right now the company is in the final stages of the project to integrate it with the rest of their production infrastructure. They are also looking to move the technology into other areas of their business like FX trading and high frequency trading, and in some cases are seeing even better performance improvements. Their Monte Carlo model, for example, was able to realize a 260- to 280-fold speedup using FPGA acceleration.

Apparently JP Morgan feels bullish enough about the technology to warrant a direct investment. In March, they acquired a 20 percent stake in Maxeler for an undisclosed amount. Although the investment is probably just a rounding error for the financial giant, it signals company’s interest in making sure Maxeler’s intellectual assets are intact.

There is certainly plenty of room to expand the Maxeler footprint at JP Morgan. To run all aspects of their financial business, the company currently has 14 thousand applications running on 50 thousand servers spread across more than 42 datacenters worldwide. Only a fraction of those applications will be amenable to acceleration, but each one has the potential to raise the company’s bottom line.

“If we can compress the space, the time and the energy required to do these calculations, then it has hard business value for us,” noted Weston. “It gives us, ultimately, a competitive edge.”

Subscribe to HPCwire's Weekly Update!

Be the most informed person in the room! Stay ahead of the tech trends with industry updates delivered to you every week!

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently king of accelerated computing) wins again, sweeping all nine Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research computing centers, national labs, federal agencies, and univ Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst firm TechInsights. Nvidia's GPU shipments in 2023 grew by more Read more…

Weekly Wire Roundup: June 2-June 7, 2024

June 8, 2024

Computex (and Jensen Huang) gave us an extra day of news this week, compensating for last week's shorter, holiday-driven news cycle. On Sunday ahead of the official start of Computex, Nvidia's CEO Jensen Huang deliver Read more…

ASC24 Expert Perspective: Dongarra, Hoefler, Yong Lin

June 7, 2024

One of the great things about being at an ASC (Asia Supercomputer Community) cluster competition is getting the chance to interview various industry experts and learning more about the various challenges the students are Read more…

HPC and Climate: Coastal Hurricanes Around the World Are Intensifying Faster

June 6, 2024

Hurricanes are among the world's most destructive natural hazards. Their environment shapes their ability to deliver damage; conditions like warm ocean waters, guiding winds, and atmospheric moisture can all dictate stor Read more…

MLPerf Training 4.0 – Nvidia Still King; Power and LLM Fine Tuning Added

June 12, 2024

There are really two stories packaged in the most recent MLPerf  Training 4.0 results, released today. The first, of course, is the results. Nvidia (currently Read more…

Highlights from GlobusWorld 2024: The Conference for Reimagining Research IT

June 11, 2024

The Globus user conference, now in its 22nd year, brought together over 180 researchers, system administrators, developers, and IT leaders from 55 top research Read more…

Nvidia Shipped 3.76 Million Data-center GPUs in 2023, According to Study

June 10, 2024

Nvidia had an explosive 2023 in data-center GPU shipments, which totaled roughly 3.76 million units, according to a study conducted by semiconductor analyst fir Read more…

ASC24 Expert Perspective: Dongarra, Hoefler, Yong Lin

June 7, 2024

One of the great things about being at an ASC (Asia Supercomputer Community) cluster competition is getting the chance to interview various industry experts and Read more…

HPC and Climate: Coastal Hurricanes Around the World Are Intensifying Faster

June 6, 2024

Hurricanes are among the world's most destructive natural hazards. Their environment shapes their ability to deliver damage; conditions like warm ocean waters, Read more…

ASC24: The Battle, The Apps, and The Competitors

June 5, 2024

The ASC24 (Asia Supercomputer Community) Student Cluster Competition was one for the ages. More than 350 university teams worked for months in the preliminary competition to earn one of the 25 final competition slots. The winning teams... Read more…

Computex 2024: Nvidia, AMD Push GPUs; Intel Revs Up x86 Power Efficiency

June 5, 2024

"The days of millions of GPU data centers are coming," said Nvidia CEO Jensen Huang during a keynote at Computex. Huang's predictions are becoming bolder and bo Read more…

Using AI and Robots to Advance Science

June 4, 2024

Even though we invented it, humans can be pretty bad at science. We need to eat and sleep, we sometimes let our emotions regulate our behavior, and our bodies a Read more…

Atos Outlines Plans to Get Acquired, and a Path Forward

May 21, 2024

Atos – via its subsidiary Eviden – is the second major supercomputer maker outside of HPE, while others have largely dropped out. The lack of integrators and Atos' financial turmoil have the HPC market worried. If Atos goes under, HPE will be the only major option for building large-scale systems. Read more…

Comparing NVIDIA A100 and NVIDIA L40S: Which GPU is Ideal for AI and Graphics-Intensive Workloads?

October 30, 2023

With long lead times for the NVIDIA H100 and A100 GPUs, many organizations are looking at the new NVIDIA L40S GPU, which it’s a new GPU optimized for AI and g Read more…

Nvidia H100: Are 550,000 GPUs Enough for This Year?

August 17, 2023

The GPU Squeeze continues to place a premium on Nvidia H100 GPUs. In a recent Financial Times article, Nvidia reports that it expects to ship 550,000 of its lat Read more…

Choosing the Right GPU for LLM Inference and Training

December 11, 2023

Accelerating the training and inference processes of deep learning models is crucial for unleashing their true potential and NVIDIA GPUs have emerged as a game- Read more…

Everyone Except Nvidia Forms Ultra Accelerator Link (UALink) Consortium

May 30, 2024

Consider the GPU. An island of SIMD greatness that makes light work of matrix math. Originally designed to rapidly paint dots on a computer monitor, it was then Read more…

Nvidia’s New Blackwell GPU Can Train AI Models with Trillions of Parameters

March 18, 2024

Nvidia's latest and fastest GPU, codenamed Blackwell, is here and will underpin the company's AI plans this year. The chip offers performance improvements from Read more…

Synopsys Eats Ansys: Does HPC Get Indigestion?

February 8, 2024

Recently, it was announced that Synopsys is buying HPC tool developer Ansys. Started in Pittsburgh, Pa., in 1970 as Swanson Analysis Systems, Inc. (SASI) by John Swanson (and eventually renamed), Ansys serves the CAE (Computer Aided Engineering)/multiphysics engineering simulation market. Read more…

Some Reasons Why Aurora Didn’t Take First Place in the Top500 List

May 15, 2024

The makers of the Aurora supercomputer, which is housed at the Argonne National Laboratory, gave some reasons why the system didn't make the top spot on the Top Read more…

Leading Solution Providers

Contributors

AMD MI3000A

How AMD May Get Across the CUDA Moat

October 5, 2023

When discussing GenAI, the term "GPU" almost always enters the conversation and the topic often moves toward performance and access. Interestingly, the word "GPU" is assumed to mean "Nvidia" products. (As an aside, the popular Nvidia hardware used in GenAI are not technically... Read more…

The NASA Black Hole Plunge

May 7, 2024

We have all thought about it. No one has done it, but now, thanks to HPC, we see what it looks like. Hold on to your feet because NASA has released videos of wh Read more…

GenAI Having Major Impact on Data Culture, Survey Says

February 21, 2024

While 2023 was the year of GenAI, the adoption rates for GenAI did not match expectations. Most organizations are continuing to invest in GenAI but are yet to Read more…

Google Announces Sixth-generation AI Chip, a TPU Called Trillium

May 17, 2024

On Tuesday May 14th, Google announced its sixth-generation TPU (tensor processing unit) called Trillium.  The chip, essentially a TPU v6, is the company's l Read more…

Q&A with Nvidia’s Chief of DGX Systems on the DGX-GB200 Rack-scale System

March 27, 2024

Pictures of Nvidia's new flagship mega-server, the DGX GB200, on the GTC show floor got favorable reactions on social media for the sheer amount of computing po Read more…

Intel’s Next-gen Falcon Shores Coming Out in Late 2025 

April 30, 2024

It's a long wait for customers hanging on for Intel's next-generation GPU, Falcon Shores, which will be released in late 2025.  "Then we have a rich, a very Read more…

Intel Plans Falcon Shores 2 GPU Supercomputing Chip for 2026  

August 8, 2023

Intel is planning to onboard a new version of the Falcon Shores chip in 2026, which is code-named Falcon Shores 2. The new product was announced by CEO Pat Gel Read more…

Shutterstock 1285747942

AMD’s Horsepower-packed MI300X GPU Beats Nvidia’s Upcoming H200

December 7, 2023

AMD and Nvidia are locked in an AI performance battle – much like the gaming GPU performance clash the companies have waged for decades. AMD has claimed it Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire