A New Breed of Heterogeneous Computing

By Michael Feldman

April 18, 2012

With the introduction of add-on accelerators like GPUs, Intel’s upcoming MIC chip, and, to a lesser extent, FPGAs, the foundation of high performance computing is undergoing somewhat of a revolution. But an emerging variant of this heterogenous computing approach may upend the current accelerator model in the not-too-distant future. And it’s already begun in the mobile space.

In October 2011, ARM announced their “big.LITTLE” design, a chip architecture than integrates large, performant ARM cores with small, power-efficient ones. The goal of this approach is to minimize power draw in order to extend the battery life of devices like smartphones and tablets.

The way it works is by mapping an application to the optimal cores based on performance demands and power availability. For mobile devices, big cores would be used for performance-demanding tasks like navigation and gaming, and the smaller cores for the OS and simpler tasks like social media apps. But when the battery runs low, the software can shunt everything to the low power cores in order the keep the device operational. ARM is claiming that battery life can be extended by as much as 70 percent by migrating tasks intelligently.

ARM’s first incarnation of big.LITTLE pairs its large Cortex-A15 design with the smaller Cortex-A7, along with glue technology to provide cache and I/O coherency between the two sets of cores. Companies like Samsung, Freescale, and Texas Instruments, among others, are already signing up.

ARM didn’t invent the big core/little core concept though. This model has been kicked around in the research community for nearly a decade. One of the first papers on the subject was written in 2003 by Rakesh Kumar, along with colleagues at UCSD and HP Labs. He proposed a single-ISA heterogenous multicore design, but in this case based on the Alpha microprocessors, a CPU line that, at the time, was being targeted to high-end workstations and servers.

He found that a chip with four different Alpha core microarchitectures had the potential to “increase energy efficiency by a factor of three… without dramatic losses in performance.” He also discovered that most of these gains would be possible with as little as two types of cores.

In a recent conversation with Kumar, he expressed the notion that the time may be ripe for single-ISA heterogeneous chips to find a home in the server arena, even in high performance computing. The driver, once again, is power, or the lack thereof. As server farms and supercomputers expand in size, electricity usage has become a limiting factor. Whether you’re scaling up or scaling out, everyone is now focused on more energy-efficient computers.

“The key insight was that even if you map an application to a little core, it’s not going to perform much worse than running it on a big core,” said Kumar, referring to his earlier research. “But you can save many factors of power.”

The problem with big powerful CPUs like the Xeon, Opteron, and Power is now well known. Although Moore’ Law is still working to expand transistor budgets at a good clip, clock frequencies are stagnant. That means performance and, especially, performance-per-watt are increasing more slowly. For these high-end server chips, essentially you have to spend four units of power to deliver one unit of performance on a per core basis.

That’s a result of the superscalar nature of these big-core microarchitectures, which feature a lot of instruction level parallelism (ILP) and deep pipelines. Such a design reduces execution latency, but at a hefty price in wattage. As Kumar explains it, “It takes a lot of power and a lot of [die] area to squeeze that last 5 to 10 percent of performance.”

The implication is to just switch to smaller, power-efficient cores, with simpler pipelines and less ILP. If you can parallelize an application across many smaller, simpler cores, you get the best of both worlds: better throughput and higher energy efficiency. The problem is that for many applications, decent performance is contingent upon single-threaded performance as well. That has led to the adoption of the types of accelerator-based computing platforms mentioned at the beginning of this article, which pairs a serial CPU chip with a throughput coprocessor.

What the big/little model brings to the table is having both types of cores on the same die. And perhaps more importantly, unlike the CPU-GPU integration that AMD is doing with their Fusion chips and what NVIDIA is planning to do with their “Project Denver” platform, the big/little model consolidates on a homogeneous instruction set.

That has a number of advantages, one of which is easier software development. With a common ISA, there is no need for a complex toolchain with multiple compilers, runtimes, libraries, and debuggers that are needed to deal with two sets of architectures. For supercomputing-type applications though, writing the application is likely to remain challenging, inasmuch as the developer still has to parallelize the code as well as explicitly map the serial work and throughput work to the appropriate cores. Unlike with mobile computing, for HPC, assigning tasks to cores would be more static, since maximizing throughput is the overriding goal.

But where performance has to be compromised because of power or resource constraints, a single ISA chip is a huge advantage. So at run-time, application threads can migrate across the different microarchitectures, as needed, to optimize for throughput, power or both. And since the cores share cache and memory, suspending a thread on one core and resuming it on another is a relatively quick and painless operation.

So, for example, a render server farm equipped with big/little CPUs could shuffle application threads to faster or slower cores depending up the workload mix, available processor resources, and the turnaround time required. If a service level agreement (SLA) was in effect that allowed the rendering job to meet its deadline without maxing out on the big cores, the server farm could save on its electricity bill by utilizing more of the little cores.

It should be noted that power savings can also be achieved by varying a microprocessor’s power supply voltage and clock frequency, otherwise know as voltage/frequency scaling. But as transistor geometries shrink, this technique tends to yield diminishing returns. And as even Intel has concluded, big/little cores — Intel calls them asymmetric cores — seem to deliver the best results.

The most likely architectures to adopt the big/little paradigm over the next few years are x86 and ARM. As mentioned before ARM big.LITTLE implementations are already in the works for mobile computing, but with the unveiling of the 64-bit ARM architecture last year, and with companies like HP delving into ARM-based gear for the datacenter, big/little implementations of ARM servers could appear as early as the middle of this decade.

We may see x86-based big/little server chips even sooner. Intel, in particular, is in prime position to take advantage of this technology. For one thing, the chipmaker is the best in the business at transistor shrinking, which is an important element if you’re interested in populating a die with a useful number of big and little cores. It also has a huge stable of x86 cores designs, from the Atom chip all the way up to the Xeon.

Also, since Intel has little in the way of GPU IP that can be used for computing, the company is most likely to rely on its x86 legacy for throughput cores. For example, it’s not too hard to imagine Intel’s big-core Xeon paired up with its little-core MIC chip in a future SoC geared for HPC duty. The same model, but with a different mix of x86 microarchitectures, could also be used to build more generic enterprise server processors, not to mention its own mobile chips.

Whether Intel intends to go down this path or not remains to be seen. But a recent patent the company filed regarding mixing asymmetric x86 cores in a processor suggests the chipmaker has indeed given serious thought to big/little products. And since both AMD and NVIDIA are pursing their own heterogeneous SoCs, which by the way could also incorporated this technology, Intel is not likely cede any advantage to its competitors.

The big/little approach won’t be a panacea for energy-efficient computing, but it looks like one of the most promising approaches, at least at the level of the CPU. The fact that it incorporates the advantages of a heterogeneous architecture, but with a simpler model, has much to recommend it. And while big/little CPUs may be seen as somewhat of a threat to GPU computing, it can also be viewed as a complementary technology. What is certain is that the days of one-size-fits-all architectures are coming to a close.

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!

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 power it brings to artificial intelligence.  Nvidia's DGX Read more…

Call for Participation in Workshop on Potential NSF CISE Quantum Initiative

March 26, 2024

Editor’s Note: Next month there will be a workshop to discuss what a quantum initiative led by NSF’s Computer, Information Science and Engineering (CISE) directorate could entail. The details are posted below in a Ca Read more…

Waseda U. Researchers Reports New Quantum Algorithm for Speeding Optimization

March 25, 2024

Optimization problems cover a wide range of applications and are often cited as good candidates for quantum computing. However, the execution time for constrained combinatorial optimization applications on quantum device Read more…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at the network layer threatens to make bigger and brawnier pro Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HBM3E memory as well as the the ability to train 1 trillion pa Read more…

Nvidia Appoints Andy Grant as EMEA Director of Supercomputing, Higher Education, and AI

March 22, 2024

Nvidia recently appointed Andy Grant as Director, Supercomputing, Higher Education, and AI for Europe, the Middle East, and Africa (EMEA). With over 25 years of high-performance computing (HPC) experience, Grant brings a 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…

NVLink: Faster Interconnects and Switches to Help Relieve Data Bottlenecks

March 25, 2024

Nvidia’s new Blackwell architecture may have stolen the show this week at the GPU Technology Conference in San Jose, California. But an emerging bottleneck at Read more…

Who is David Blackwell?

March 22, 2024

During GTC24, co-founder and president of NVIDIA Jensen Huang unveiled the Blackwell GPU. This GPU itself is heavily optimized for AI work, boasting 192GB of HB Read more…

Nvidia Looks to Accelerate GenAI Adoption with NIM

March 19, 2024

Today at the GPU Technology Conference, Nvidia launched a new offering aimed at helping customers quickly deploy their generative AI applications in a secure, s Read more…

The Generative AI Future Is Now, Nvidia’s Huang Says

March 19, 2024

We are in the early days of a transformative shift in how business gets done thanks to the advent of generative AI, according to Nvidia CEO and cofounder Jensen 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…

Nvidia Showcases Quantum Cloud, Expanding Quantum Portfolio at GTC24

March 18, 2024

Nvidia’s barrage of quantum news at GTC24 this week includes new products, signature collaborations, and a new Nvidia Quantum Cloud for quantum developers. Wh Read more…

Houston We Have a Solution: Addressing the HPC and Tech Talent Gap

March 15, 2024

Generations of Houstonian teachers, counselors, and parents have either worked in the aerospace industry or know people who do - the prospect of entering the fi Read more…

Alibaba Shuts Down its Quantum Computing Effort

November 30, 2023

In case you missed it, China’s e-commerce giant Alibaba has shut down its quantum computing research effort. It’s not entirely clear what drove the change. 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…

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…

DoD Takes a Long View of Quantum Computing

December 19, 2023

Given the large sums tied to expensive weapon systems – think $100-million-plus per F-35 fighter – it’s easy to forget the U.S. Department of Defense is a 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…

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…

Intel’s Server and PC Chip Development Will Blur After 2025

January 15, 2024

Intel's dealing with much more than chip rivals breathing down its neck; it is simultaneously integrating a bevy of new technologies such as chiplets, artificia Read more…

Baidu Exits Quantum, Closely Following Alibaba’s Earlier Move

January 5, 2024

Reuters reported this week that Baidu, China’s giant e-commerce and services provider, is exiting the quantum computing development arena. Reuters reported � Read more…

Leading Solution Providers

Contributors

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…

Shutterstock 1179408610

Google Addresses the Mysteries of Its Hypercomputer 

December 28, 2023

When Google launched its Hypercomputer earlier this month (December 2023), the first reaction was, "Say what?" It turns out that the Hypercomputer is Google's t Read more…

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…

Shutterstock 1606064203

Meta’s Zuckerberg Puts Its AI Future in the Hands of 600,000 GPUs

January 25, 2024

In under two minutes, Meta's CEO, Mark Zuckerberg, laid out the company's AI plans, which included a plan to build an artificial intelligence system with the eq Read more…

Google Introduces ‘Hypercomputer’ to Its AI Infrastructure

December 11, 2023

Google ran out of monikers to describe its new AI system released on December 7. Supercomputer perhaps wasn't an apt description, so it settled on Hypercomputer Read more…

China Is All In on a RISC-V Future

January 8, 2024

The state of RISC-V in China was discussed in a recent report released by the Jamestown Foundation, a Washington, D.C.-based think tank. The report, entitled "E Read more…

Intel Won’t Have a Xeon Max Chip with New Emerald Rapids CPU

December 14, 2023

As expected, Intel officially announced its 5th generation Xeon server chips codenamed Emerald Rapids at an event in New York City, where the focus was really o Read more…

IBM Quantum Summit: Two New QPUs, Upgraded Qiskit, 10-year Roadmap and More

December 4, 2023

IBM kicks off its annual Quantum Summit today and will announce a broad range of advances including its much-anticipated 1121-qubit Condor QPU, a smaller 133-qu Read more…

  • arrow
  • Click Here for More Headlines
  • arrow
HPCwire