Revving up HPC Workloads with FPGAs
In enterprise data centers, big data just keeps getting even bigger — and that’s a good problem to have. There is a tremendous amount of potential business value in the flood of data flowing from the Internet of Things, social media, customer information systems and other pipelines filled with incoming bits and bytes. The trick is to capture value from all of that data — and that takes ever-faster computing systems.
This is where field-programmable gate arrays (FPGAs) take to the IT playing field. FPGAs are a type of processor that can be programmed after manufacturing and deployed alongside general-purpose CPUs to accelerate throughput for targeted functions in compute- and data-intensive workloads. FPGAs allow developers to offload repetitive processing functions in workloads to speed up application performance.
This capability makes FPGAs particularly beneficial in HPC applications, according to Addison Snell, CEO of Intersect360 Research, a firm that focuses on the HPC market.
“A big trend that we are seeing in high-performance computing worldwide is the diversification of processor technologies corresponding with the diversification of high-performance workloads,” Snell says. “End users, particularly in commercial markets, are dealing not only with their traditional HPC applications but they’ve recently gone through a wave of big data and analytics types of applications, and now also are being asked to address machine learning, deep learning and AI types of applications. And they need to do all of this within their existing infrastructure, and usually without a major corresponding increase in budgets.”
These challenges have led organizations to explore the types of processing elements that can best be deployed against a widening set of needs in the application space, Snell notes. And one of those technologies is FPGAs.
“While FPGAs have been part of the HPC market for decades, they are now seeing increased deployment,” he says. “FPGAs can be tailored to specific workloads that are often repeated. We see them deployed especially for certain highly scalable applications — such as genomics, financial modeling and analytics — as well as hyperscale applications like Internet search.” FPGAs are also valuable tools for accelerating data movement in storage systems, he notes.
Looking ahead, Snell expects FPGAs to see broader adoption in the HPC space as users adapt their workloads to take advantage of diverse processor technologies. And even better, the economics of FPGAs can be great for HPC users. FPGAs can accelerate workloads while using far less power than general-purpose CPUs.
Hitting the accelerator with FPGAs
Intel sees great potential in using FPGAs to accelerate genomics. FPGAs are part of the integrated hardware-software BIGstack for genomics that Intel developed in partnership with the Broad Institute of MIT and Harvard, the leader in genomics research.
BIGstack 2.0, announced in November 2017, incorporates Intel® Xeon® Scalable processors, Intel® 3D NAND SSD and Intel FPGAs, along with the latest genomics tools from the Broad Institute, to provide a 3.34x speed-up in whole genome analysis, according to an internal Intel analysis. This kind of speed will allow researchers “to analyze more genomes, more quickly and at lower cost, enabling new discoveries, new treatment options, and faster diagnosis of disease,” according to Intel.
Elsewhere in the HPC world, Microsoft is using FPGAs to accelerate processing and networking. In a paper exploring the topic, a team from Microsoft proposed a new cloud architecture that uses reconfigurable logic, in the form of FPGAs, to accelerate both network plane functions and applications. The paper explains how this architecture can be used for both service acceleration (Web search ranking) and network acceleration (encryption of data in transit at high speeds).
Moving toward the mainstream
With the rising importance of compute accelerators in HPC systems, FPGAs are moving into the mainstream of enterprise computing. Dell EMC, for example, now makes Intel FPGA Accelerators available in Dell EMC PowerEdge™ R640, R740, R740xd, R7425, R840 and R940xa servers.3 These systems build on Dell EMC’s leadership in the combined HPC server and storage market.4
Amazon Web Services, meanwhile, has built an end-to-end solution that lets developers build FPGA-powered applications and services and list them in the AWS Marketplace. Many developers are seizing the opportunity, according to a blog post from Jeff Barr, chief evangelist for AWS — and for good reason. “Properly programmed, an FPGA has the potential to provide a 30x speedup to many types of genomics, seismic analysis, financial risk analysis, big data search, and encryption algorithms and applications,” Barr writes.5
So, what’s the bottom line? If your organization uses HPC systems, FPGAs are probably in your future — if they are not there already.
 Intel article, “Intel and Broad Institute Deliver For Genomics Research, Again,” Nov. 13, 2017.
4 Intersect360 Research, 2018.
5 Jeff Barr, AWS chief evangelist, “EC2 F1 Instances with FPGAs – Now Generally Available,” April 19, 2017.