RESEARCH REPORT

In brief

In brief

  • This report explains why specialized compute and heterogenous hardware will become key to enterprise infrastructure.
  • The "jack of all trades" CPU will be supplemented with purpose-built accelerators like GPUs, ASICs and FPGAs to solve specific business problems.
  • Quantum and bio-inspired computing will take computational power to a new level, enabling enterprises to rethink intractable problems and explore new solutions.
  • The report provides recommendations for enterprises as they prepare to make the jump to a far more heterogeneous computing future.


Move over Moore’s Law

For decades, general computing power has been increasing exponentially. But demand keeps growing. Compute-hungry machine learning algorithms, big data analytics, and mobile and edge devices are just the latest developments to drive up enterprise requirements for computing power.

The problem? General purpose compute just can’t keep increasing performance to meet this growing demand. Moore’s Law—the observation that computing power doubles every two years or so—has held true for years. But the point at which enhancing CPU performance at the same rate becomes simply uneconomical is fast approaching. Some think it might happen as early as 2021.

So what’s next? As CPU architecture hits its limits, enterprises will increasingly turn to specialized infrastructure to meet their needs for more computational power. Enterprise infrastructure is about to get far more varied.

GPU

Graphics Processing Units (GPUs) outperform general-purpose CPUs when large blocks of data are processed in parallel.

ASIC

Application-Specific Integrated Circuits (ASICs) are hard-wired for a particular purpose offering greater speed and power-efficiency.

FPGA

Field-Programmable Gate Arrays (FPGAs) can be reconfigured after manufacturing, excelling in parallel applications and power-constrained environments.

Quantum

The emerging field of quantum computing promises to vastly outperform classical computing in certain domains such as simulating quantum systems (including molecules).

Bio-inspired

Bio-inspired computing draws insights from biological systems to pursue performance gains in fields like perception and multisensory integration.

Optical

Harnessing light to perform calculations, optical computers could potentially solve some problems faster than silicon chips could.

Everything in its right place: Planning for a heterogeneous future

How will enterprises blend general and special-purpose compute in this next-generation heterogeneous infrastructure environment? By understanding the strengths and weaknesses of each processor type–and deploying them where they can add most value:

  • CPUs for innovation. The low cost, flexibility, and accessibility of CPUs will ensure they remain a foundation of enterprise architectures for years to come, making them the best choice for core computing requirements and early-stage enterprise innovation.
  • Accelerators for optimization. As concepts, devices, and methods are refined and improved, optimization becomes more important than flexibility. This is where purpose-built accelerators like GPUs, ASICs and FPGAs become key, trading flexibility to focus on optimizing specific problems, from throughput to latency to power consumption.
  • Emerging compute for revolution. Emerging accelerators like quantum and bio-inspired computing will fundamentally change computation, enabling enterprises to solve previously intractable problems such as truly in-silico drug design or real-time adaptive logistic routing.
The key is to understand the strengths and weaknesses of specialized hardware options, ensuring each processor type is applied to the task it’s best suited to.

Making the jump to computational variety

It’s time to prepare for the coming era of heterogenous computing infrastructure. Here are some recommendations to guide enterprises’ thinking as they plan for the addition of ever more specialized hardware:

  1. Revisit existing algorithms. Enterprises should be reassessing existing assumptions about the speed and memory cost of communicating with specialized hardware and should be looking to rethink siloed processing decisions that under-utilize accelerators.
  2. Accelerate frictionless computation. Early movers should identify specific enterprise problems which could benefit from using accelerators, creating reusable application-specific implementations or leveraging accelerator services provided by vendors.
  3. Architect for difference. It’s now time to be thinking about the unique requirements of heterogeneous computation when creating or refreshing architectures, including co-locating data-intensive workloads with infrastructure where necessary.
  4. Redefine the edge. Data-rich and real-time workloads will often also need to be located closer to the data—that is on devices at the network edge. That means a heterogeneous architecture that can operate with low latencies and low power will be key.

Time to embrace heterogeneity

In recent years, the homogeneity, commoditization, and cost-effectiveness offered by virtual machines and cloud computing have enabled enterprises to solve problems once thought too expensive to address, helping established companies evolve and grow and sparking a host of new digital native organizations.

Now the infrastructure landscape is changing once again. A more heterogeneous computing infrastructure will offer enterprises the opportunity to find revolutionary solutions to some of their most computationally challenging problems.

It’s time to embrace the heterogeneous future and start planning now. There are challenges and obstacles to overcome. But the early movers have an opportunity to capture an outsized share of the business innovation, optimization and even revolution on offer.

Subscription Center
Stay in the Know with Our Newsletter Stay in the Know with Our Newsletter