In-Socket Accelerators — When to Use Them

By Kevin Urban

June 5, 2008

Today’s system architects have a tough enough job solving difficult architectural problems for applications like 40G line cards, HD Video Transcoding Systems, and next generation RADAR applications. However the most demanding part of their job is they also need to draw the difficult line between hardware and software; keeping manpower requirements in the equation, costs in check, and architecting a solution that can be built inside the market required timeframe. This is all in a day’s work for some, but an almost impossible task for many.

Since software is generally cheaper to develop and the number of software engineers outnumber hardware engineers by almost an order of magnitude, the explosion of software-based solutions on x86 platforms has arguably become the de facto standard for many platforms. Many of these platforms were (and still are) built from the previous generations of CPUs with reusable software, and thus leveraged faster next-generation CPUs or multicores for their roadmap. That is, until now.

For 2008, the industry buzzwords are ”hardware acceleration.” CPU vendors are integrating custom integrated IP into their chips. AMD and Intel are creating an ecosystem for third party accelerators, named Torenzza and QuickAssist. GPU vendors are setting their sights on general purpose functionality. Meanwhile, a host of other chip companies, too many to mention, are developing new products that target this high performance computing (HPC) market.

A little known fact in the HPC community is that the embedded computing market has always solved their problems with a combination of CPUs and accelerators. Due to different space, weight, power, and environmental requirements for high performance embedded computing (HPEC), a large portion of those accelerators are implemented with field programmable gate arrays (FPGAs) from companies like Altera, Xilinx, and others.

As these two markets collide, the emergence of some of the world’s largest companies from HPC and HPEC are now working together (Intel, AMD, Altera, and Xilinx) to make x86 CPUs with their own internal accelerators and easy to access external FPGA accelerators the solution of choice for both HPC and HPEC. As CPU-bound, software-only solutions require the benefits of hardware acceleration to remain competitive, architects must figure out what to accelerate and how to make the price, performance, and power trade-offs that meet market requirements.

What do I accelerate?

In these examples, we will concentrate on FPGAs, but many of the concepts can be applied to other types of accelerators. A simplistic and very common place for designers to start is to profile your C/C++ code, find the routines that take most of your clock cycles, and start your efforts to increase performance or remove bottlenecks there. Hardware designers have done this for years and found the following types of applications that make sense to run on FPGA hardware. Here are some typical applications that can be parallelized beyond a factor of 10x improvement:

1)  Filters – FIR, IIR, Poly-Phase.
2)  Fast-Fourier Transforms (FFT).
3)  Encryption – AES, TDES, DES, etc.
4)  Video Transcoding – MPEG2, H.264, VC-1, and others.
5)  Compression – ZLIB, GZIP, etc.
6)  Bioinformatics – Smith Waterman, BLAST, ClustalW.
7)  Random Number Generation (RNG) – SOBOL and Mersenne Twister for Monte-Carlo.
8)  Medical Imagining – CT Back Projection.
9)  Packet and Network Processing (IPv6, Deep Packet Inspection).
10) Market Data – FIX, FAST FIX, OPRA, etc.

Another way to describe or find these algorithms in your code is to think of them in one of two ways:

1)  Bit-level processing with deep instruction pipelines.
2)  Vector-based processing of large amounts of data.
 
Both of these are trying to analyze data in ways that are not the standard 32 bit or 64 bit instruction, which creates overheard for the CPU or GPU because it has a fixed data and instruction set size. Additionally, many of these functions can have very deep instruction pipelines, which in an FPGA can be defined as deep as necessary. FPGAs can be programmed to be 3 bit machines in the example of BLAST that can be parallelized hundreds of times with instruction pipelines over one hundred deep. This creates a single resulting chip which can contain a 1000-core “machine” running at 300MHz or 100x faster than a single 3GHz CPU core.
 
Another example is encryption, which contains many XOR functions at the bit level, which again can be created easily in a FPGA, parallelized thousands of times, to create a machine that can encrypt data at 4GB/s (Bytes, not bits) using less than 50 percent of a large 65nm FPGA.
 
A secondary effect of FPGA technology is substantial power savings over today’s high-end CPUs or GPUs. The largest FPGA built on 65nm technology can consume 25 to 30 watts of power versus a x86 CPU, which can run over 100 watts, or worse, a GPGPU, which can easily run over 200 watts. Combine this power savings with any of the examples above and you can have two orders of magnitude improvement on a performance/watt metric. This can be hard to ignore if you are in a design environment that has a limited power budget, like a UAV (Unmanned Aerial Vehicle), or a financial datacenter with major power/cooling concerns.

How do I design my FPGA accelerator?

In order to design an accelerator, you need to decide if you need acceleration in the first place. If someone can achieve what they need with just moving from dual to quad cores or with a C-code rewrite, then that is usually the easier. FPGA acceleration and other exotic technology come into play for those designers who are really pushing the envelope.

The recent push towards industry-standard APIs for accelerators allows architects to build their own accelerated systems with CPU+FPGA rapidly, and avoid “single vendor” lock-in and end-of-life (EOL) issues. Additionally, open standard organizations like OpenFPGA (www.openfpga.org) are making a concerted effort to help take FPGAs into the mainstream of accelerated computing. By taking an open standards approach for API’s and benchmarks, and by bringing the community together, they are bringing the combination of x86 CPU+FPGA closer to mainstream HPC and even the desktop.

As an alternative to the standard programming API approach, many designers would prefer to just purchase a working acceleration technology and use vendor-supplied building blocks to implement their algorithms. Products like those from Cavium and LSI/Tarari are attractive here. Similarly, out-of-the box FPGA IP solutions like XtremeData’s Floating-Point Vector Math Library, Fast-Fourier Transform, and Random Number Generation offer rapid development options. These solutions are pre-programmed hardware solutions that fit into the design flow very easily. No hardware expertise is needed to use them, which is why they are gaining a lot of attention in the marketplace. Plug them into your system, compile a library, or make a function call in your C/C++ routine, and you are off to the next task.

Finally, a third solution is to hire a hardware expert to help build your own secret-sauce algorithm. FPGA and ASIC consulting companies help customers build exactly what they want, make HW/SW trade-offs, and in some cases help train their customers to maintain and enhance the hardware design. If you need acceleration to be part of your differentiation, then this can be a very profitable and easy way to approach new technologies.

It’s faster… now what? (How do you measure your performance, price, power gains?)

Typically, designers who are looking at acceleration techniques are looking for about a 10x performance improvement of their total application to make it worth the effort. Depending on the goal of the system, the units of measure for success can vary wildly.

Power

For example, every watt that can be saved in a UAV will yield many different advantages to the system: Power and cooling savings, which equals less weight, which means that weight can be replaced by fuel, which means the UAV can fly further or faster, which might be critical to a new mission. Thus a solution that leverages an accelerator might be of equal or “only 2x the performance” to the existing system, but if that yields solutions that are half the size or power, then that “only 2x” savings becomes quite significant.

Performance

Another example of how applying acceleration techniques can pay off is raw performance. For example, Investment banks, hedge funds, and boutique trading firms know that increased performance of Monte-Carlo simulations for derivatives pricing and risk assessments can have big money implications. The key requirement in this space is maximum performance of double-precision floating point calculations. This is one of the main reasons why these institutions are turning toward multicore CPUs, and when maximum performance is required, increasingly towards FPGA-based solutions. For example, today’s largest 65nm Stratix III FPGAs from Altera can support up to 40 gigaflops of sustained double-precision floating point throughput at around 30 watts. Importantly, from the end-user poin of view, the ability to finish these calculations faster and more accurately than your competitor can yield tens of millions of dollars in increased profits from exploiting arbitrage opportunities.

Human Benefit

Other examples include FPGAs used for algorithms such as BLAST in the bioinformatics space, which if done faster, can help unravel the codification and functionality of genes and proteins, with major implications for the drug-discovery process. In medical imaging, back-projection algorithms implemented in FPGAs allow for smaller, cheaper, and faster x-ray CT scanners, which can put life-saving information in the hands of more doctors that need it. Increased precision in a next generation radar system can lead to better missile defense or earlier warning of possible threats, which can save lives.

Other

In general, there are dozens of ways to measure acceleration success or lack of it. It will take some effort to evaluate, implement, and then to maintain them. However, many companies are having great success “however they measure it,” using some of the techniques mentioned so far in this article. To help you find other terms for success, you can address these questions: Does 100x performance break a paradigm? Does this create an entirely new market? Does it provide value that cannot be defined by dollars saved, but by lives saved, breakthroughs made, or other similar metrics? These questions really measure the intangible value of acceleration, and the business-savvy people are the ones who can extract tangible value from these type of differentiators.
 
Think Differently: Re-look at the algorithm

Lastly, the industry is beginning to realize that many of the implementations that we use today are the way they are because of 40 years of von Neumann programming and thinking. Since college (or even earlier) many of today’s programmers have been taught, trained, and lead into thinking that a serial approach is the best and only approach to every problem. The majority of today’s engineers tackle problems the way they were taught or the ways that have worked in the past. Now that the “core wars” are virtually over that 40 years of legacy is breaking. GPUs now have 128 cores and growing. FPGAs can be designed to have thousands of cores, and the word “accelerator” has become a mainstream term once again. Questions like “Can we use direct linear filtering rather than transforming to the frequency domain?”; “Can my accelerator put compression/decompression closer to the CPU so bottlenecks are removed?”; “Do I use double-precision floating point because it is easy or because I need it?”; and many others need to be asked and answered at the architectural level, before our preprogrammed way of thinking gets in the way.

Regardless of what questions your company needs to ask itself or how you decide to solve your problem, programmers and architects are realizing that having 8, 16, or 32 cores isn’t going to solve all their problems anymore. Additionally, as new generations of FPGAs are available, they will have an increased performance per watt advantage over future generations of GPGPUs and manycore CPUs. An example is the recently announced 40nm Stratix IV family from Altera, which is again 2x larger and reasonably faster than previous generations. So step back, think of the algorithm, look at it through a parallel set of eyes, and enjoy. The world is parallel once again, shouldn’t your implementation be?

About the Author

Kevin Urban, director of business development, XtremeData, Inc., has been actively involved in the creation and marketing of advanced communication systems for over 15 years. He holds undergraduate and graduate degrees in electrical engineering from the University of Michigan and a graduate degree in business from the University of Chicago. Additionally, he owns patents for his work in wire-speed latency sensitive scheduling algorithms.

Subscribe to HPCwire's Weekly Update!

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

SIA Recognizes Robert Dennard with 2019 Noyce Award

November 12, 2019

If you don’t know what Dennard Scaling is, the chances are strong you don’t labor in electronics. Robert Dennard, longtime IBM researcher, inventor of the DRAM and the fellow for whom Dennard Scaling was named, is th Read more…

By John Russell

Leveraging Exaflops Performance to Remediate Nuclear Waste

November 12, 2019

Nuclear waste storage sites are a subject of intense controversy and debate; nobody wants the radioactive remnants in their backyard. Now, a collaboration between Berkeley Lab, Pacific Northwest National University (PNNL Read more…

By Oliver Peckham

Using HPC and Machine Learning to Predict Traffic Congestion

November 12, 2019

Traffic congestion is a never-ending logic puzzle, dictated by commute patterns, but also by more stochastic accidents and similar disruptions. Traffic engineers struggle to model the traffic flow that occurs after accid Read more…

By Oliver Peckham

Mira Supercomputer Enables Cancer Research Breakthrough

November 11, 2019

Dynamic partial-wave spectroscopic (PWS) microscopy allows researchers to observe intracellular structures as small as 20 nanometers – smaller than those visible by optical microscopes – in three dimensions at a mill Read more…

By Staff report

IBM Adds Support for Ion Trap Quantum Technology to Qiskit

November 11, 2019

After years of percolating in the shadow of quantum computing research based on superconducting semiconductors – think IBM, Rigetti, Google, and D-Wave (quantum annealing) – ion trap technology is edging into the QC Read more…

By John Russell

AWS Solution Channel

Making High Performance Computing Affordable and Accessible for Small and Medium Businesses with HPC on AWS

High performance computing (HPC) brings a powerful set of tools to a broad range of industries, helping to drive innovation and boost revenue in finance, genomics, oil and gas extraction, and other fields. Read more…

IBM Accelerated Insights

Tackling HPC’s Memory and I/O Bottlenecks with On-Node, Non-Volatile RAM

November 8, 2019

On-node, non-volatile memory (NVRAM) is a game-changing technology that can remove many I/O and memory bottlenecks and provide a key enabler for exascale. That’s the conclusion drawn by the scientists and researcher Read more…

By Jan Rowell

IBM Adds Support for Ion Trap Quantum Technology to Qiskit

November 11, 2019

After years of percolating in the shadow of quantum computing research based on superconducting semiconductors – think IBM, Rigetti, Google, and D-Wave (quant Read more…

By John Russell

Tackling HPC’s Memory and I/O Bottlenecks with On-Node, Non-Volatile RAM

November 8, 2019

On-node, non-volatile memory (NVRAM) is a game-changing technology that can remove many I/O and memory bottlenecks and provide a key enabler for exascale. Th Read more…

By Jan Rowell

MLPerf Releases First Inference Benchmark Results; Nvidia Touts its Showing

November 6, 2019

MLPerf.org, the young AI-benchmarking consortium, today issued the first round of results for its inference test suite. Among organizations with submissions wer Read more…

By John Russell

Azure Cloud First with AMD Epyc Rome Processors

November 6, 2019

At Ignite 2019 this week, Microsoft's Azure cloud team and AMD announced an expansion of their partnership that began in 2017 when Azure debuted Epyc-backed ins Read more…

By Tiffany Trader

Nvidia Launches Credit Card-Sized 21 TOPS Jetson System for Edge Devices

November 6, 2019

Nvidia has launched a new addition to its Jetson product line: a credit card-sized (70x45mm) form factor delivering up to 21 trillion operations/second (TOPS) o Read more…

By Doug Black

In Memoriam: Steve Tuecke, Globus Co-founder

November 4, 2019

HPCwire is deeply saddened to report that Steve Tuecke, longtime scientist at Argonne National Lab and University of Chicago, has passed away at age 52. Tuecke Read more…

By Tiffany Trader

Spending Spree: Hyperscalers Bought $57B of IT in 2018, $10B+ by Google – But Is Cloud on Horizon?

October 31, 2019

Hyperscalers are the masters of the IT universe, gravitational centers of increasing pull in the emerging age of data-driven compute and AI.  In the high-stake Read more…

By Doug Black

Cray Debuts ClusterStor E1000 Finishing Remake of Portfolio for ‘Exascale Era’

October 30, 2019

Cray, now owned by HPE, today introduced the ClusterStor E1000 storage platform, which leverages Cray software and mixes hard disk drives (HDD) and flash memory Read more…

By John Russell

Supercomputer-Powered AI Tackles a Key Fusion Energy Challenge

August 7, 2019

Fusion energy is the Holy Grail of the energy world: low-radioactivity, low-waste, zero-carbon, high-output nuclear power that can run on hydrogen or lithium. T Read more…

By Oliver Peckham

Using AI to Solve One of the Most Prevailing Problems in CFD

October 17, 2019

How can artificial intelligence (AI) and high-performance computing (HPC) solve mesh generation, one of the most commonly referenced problems in computational engineering? A new study has set out to answer this question and create an industry-first AI-mesh application... Read more…

By James Sharpe

Cray Wins NNSA-Livermore ‘El Capitan’ Exascale Contract

August 13, 2019

Cray has won the bid to build the first exascale supercomputer for the National Nuclear Security Administration (NNSA) and Lawrence Livermore National Laborator Read more…

By Tiffany Trader

DARPA Looks to Propel Parallelism

September 4, 2019

As Moore’s law runs out of steam, new programming approaches are being pursued with the goal of greater hardware performance with less coding. The Defense Advanced Projects Research Agency is launching a new programming effort aimed at leveraging the benefits of massive distributed parallelism with less sweat. Read more…

By George Leopold

AMD Launches Epyc Rome, First 7nm CPU

August 8, 2019

From a gala event at the Palace of Fine Arts in San Francisco yesterday (Aug. 7), AMD launched its second-generation Epyc Rome x86 chips, based on its 7nm proce Read more…

By Tiffany Trader

D-Wave’s Path to 5000 Qubits; Google’s Quantum Supremacy Claim

September 24, 2019

On the heels of IBM’s quantum news last week come two more quantum items. D-Wave Systems today announced the name of its forthcoming 5000-qubit system, Advantage (yes the name choice isn’t serendipity), at its user conference being held this week in Newport, RI. Read more…

By John Russell

Ayar Labs to Demo Photonics Chiplet in FPGA Package at Hot Chips

August 19, 2019

Silicon startup Ayar Labs continues to gain momentum with its DARPA-backed optical chiplet technology that puts advanced electronics and optics on the same chip Read more…

By Tiffany Trader

Crystal Ball Gazing: IBM’s Vision for the Future of Computing

October 14, 2019

Dario Gil, IBM’s relatively new director of research, painted a intriguing portrait of the future of computing along with a rough idea of how IBM thinks we’ Read more…

By John Russell

Leading Solution Providers

ISC 2019 Virtual Booth Video Tour

CRAY
CRAY
DDN
DDN
DELL EMC
DELL EMC
GOOGLE
GOOGLE
ONE STOP SYSTEMS
ONE STOP SYSTEMS
PANASAS
PANASAS
VERNE GLOBAL
VERNE GLOBAL

Intel Confirms Retreat on Omni-Path

August 1, 2019

Intel Corp.’s plans to make a big splash in the network fabric market for linking HPC and other workloads has apparently belly-flopped. The chipmaker confirmed to us the outlines of an earlier report by the website CRN that it has jettisoned plans for a second-generation version of its Omni-Path interconnect... Read more…

By Staff report

Kubernetes, Containers and HPC

September 19, 2019

Software containers and Kubernetes are important tools for building, deploying, running and managing modern enterprise applications at scale and delivering enterprise software faster and more reliably to the end user — while using resources more efficiently and reducing costs. Read more…

By Daniel Gruber, Burak Yenier and Wolfgang Gentzsch, UberCloud

Dell Ramps Up HPC Testing of AMD Rome Processors

October 21, 2019

Dell Technologies is wading deeper into the AMD-based systems market with a growing evaluation program for the latest Epyc (Rome) microprocessors from AMD. In a Read more…

By John Russell

Intel Debuts Pohoiki Beach, Its 8M Neuron Neuromorphic Development System

July 17, 2019

Neuromorphic computing has received less fanfare of late than quantum computing whose mystery has captured public attention and which seems to have generated mo Read more…

By John Russell

Rise of NIH’s Biowulf Mirrors the Rise of Computational Biology

July 29, 2019

The story of NIH’s supercomputer Biowulf is fascinating, important, and in many ways representative of the transformation of life sciences and biomedical res Read more…

By John Russell

Xilinx vs. Intel: FPGA Market Leaders Launch Server Accelerator Cards

August 6, 2019

The two FPGA market leaders, Intel and Xilinx, both announced new accelerator cards this week designed to handle specialized, compute-intensive workloads and un Read more…

By Doug Black

When Dense Matrix Representations Beat Sparse

September 9, 2019

In our world filled with unintended consequences, it turns out that saving memory space to help deal with GPU limitations, knowing it introduces performance pen Read more…

By James Reinders

With the Help of HPC, Astronomers Prepare to Deflect a Real Asteroid

September 26, 2019

For years, NASA has been running simulations of asteroid impacts to understand the risks (and likelihoods) of asteroids colliding with Earth. Now, NASA and the European Space Agency (ESA) are preparing for the next, crucial step in planetary defense against asteroid impacts: physically deflecting a real asteroid. Read more…

By Oliver Peckham

  • arrow
  • Click Here for More Headlines
  • arrow
Do NOT follow this link or you will be banned from the site!
Share This