How SoC IP Powers High-Performance Computing Applications

Scott Durrant

Nov 16, 2021 / 7 min read

The notion of high-performance computing (HPC) was, until recently, primarily considered in the context of large-scale industrial, scientific, and institutional uses. Typically, HPC has been associated with massive data-crunching applications where large volumes of raw data need to be analyzed to aid in tasks like oil and gas exploration, climate forecasting, bioscience research, financial markets modeling, and military-grade surveillance. While this may make HPC seem far removed from our everyday lives of checking emails, watching Netflix, taking online lessons, Zoom-ing with grandma, or scrolling through social media, the reality is otherwise.

The value of HPC has expanded well beyond the big company labs to have a critical, albeit largely unnoticed, presence in our living rooms. Access to extraordinary computing power has been democratized in many ways, regardless of our interface to it (e.g., a PC, mobile phone, TV, or smart doorbell). It may not be obvious to the average end user, but powerful processing architectures underly just about everything we do that relies on a combination of connectivity and data. In fact, the lines between “high performance” computing and “garden variety” computing have blurred as we shift closer to an almost entirely cloud-based existence. The bottom line is that everything requires HPC to some degree.

As system demands grow more complex, designers building system-on-chips (SoCs) for HPC need a combination of high-performance and low-latency IP solutions that improve total system throughput, reduce design risk in advanced nodes, and accelerate time-to-market.

Read on to learn more about what makes HPC so ubiquitous, why IP continues to be a top priority for customers designing leading-edge chips, how the quality of your IP can make or break silicon success, and the road ahead.

Video Conferencing

What’s Driving the Growth in HPC?

The main driver behind more ubiquitous HPC is, no surprise, more data. A lot more. As people generate, copy, share, save, and consume more information of all types — photos, movies, financial information, medical records, online shopping transactions, and virtual and augmented reality applications — the world needs faster, more robust, and more secure computing. On top of that, expectations for more intelligent, intuitive, and even autonomous capabilities, enabled by artificial intelligence and machine learning, are driving the need for souped-up computer horsepower at every turn of our lives.

This has been, of course, a natural progression, fueled by the pervasiveness of internet access, faster connectivity, and the relentless pace of Moore’s law which enables more capable and faster computing. At this rate, one only needs to look at the remarkable growth of the data center market to visualize the pace and scale of HPC’s growth trajectory.

But no one could have predicted the arrival of a global pandemic and the impact it would have on our need for even greater levels of performance. Data consumption reached an all-time high in 2020 and most experts point to global usage trends directly impacted by COVID-19. IDC experts who closely monitor these trends say that annual data consumption topped out at 64.2 zettabytes in 2020 and will grow to more than 175 zettabytes in 2025. While that spike may have momentarily taken us off guard, it surely signals a more permanent acceleration.

Silicon Innovation Forms the Foundation of HPC

The foundation supporting the world’s insatiable appetite for data needs to be continuously fortified. For Synopsys, that means arming the companies who operate the infrastructure required to manage and connect all this data with ever more powerful engines – namely faster, larger, and more energy-efficient chips (lest we forget the massive power consumption challenge needed to manage 180 zettabytes of data!).

With increased HPC demand comes the need for new semiconductor innovation — even beyond what Gordon Moore could have imagined. As parallel processing, AI, and quantum techniques proliferate, the silicon architectures needed to enable a new wave of high-value applications become larger, more complex, and demand high performance and energy efficiency.

The world’s leading semiconductor companies driving toward a goal to increase AI compute performance by 1000x underscores the commitment to using innovative design tools and methodologies. Technology innovators like IBM, Intel, NVIDIA, and Samsung, as well as the hyperscalers driving new levels of speed and capacity in data centers, and up-and-coming AI startups are all attacking the scaling and performance challenges from a variety of new angles. HPC involves a wide range of ecosystem players, all connected by their insatiable need for more data throughput.

Designers require high-performance, power-efficient, and low-latency solutions to deliver total system throughput when and where it is needed. Pushing the limits of HPC is driving the next generation of semiconductor SoC designs. When it comes to overall performance, raw compute power is no longer the only need, and the support for new high-speed protocols and optimized data processing, networking, and storage in the cloud is key.

How IP is Helping

Many of the challenges of designing next-generation chips for HPC have been well covered in previous blog posts, particularly as we look at the design tools and methodologies required. This includes our focus on applying AI to design tools to automate and optimize tasks that have scaled beyond human comprehension.

With the need to integrate multi-die and chiplet architectures, HPC silicon is getting more complex. Additionally, the rapid increase in data drives the need for more compute power and for systems to deliver a seamless experience at the endpoint. Systems need faster interfaces to move data from point A to point B, efficient and intelligent storage infrastructure to store and retrieve data, and AI, graphics, and other accelerators to extract meaningful value from all this data. High-performance IP can accelerate the design of chips that address these challenges. And IP verification has expanded from being just a block-level exercise to verifying the IP in a system context (e.g., IP sees real-world stimulus early on).

Another critical element is how silicon-proven IP can be used to cut the development time of next-generation SoCs for HPC. Integrating proven IP reduces integration risk and accelerates time-to-market for the complex chips driving HPC.

Accelerating Silicon Success with Innovations in IP

With a nod to my previous post on how IP is critical to many elements in a complex SoC chip design, let’s look at a few areas where the Synopsys DesignWare® IP portfolio helps address important HPC challenges:

 

Data Transfer Speed

You can’t talk about HPC and cloud without addressing latency and data transfer within the context of SoCs for HPC. Reducing latency will continue to grow in importance as more processing and control systems move online and there is an increasing reliance on the cloud for edge IoT.

In the past, HPC systems would have to copy data from one memory domain to another to allow multiple processors to operate on a single dataset. Such data copies increase application latency as well as power and resource consumption. Today, the trend toward use of cache-coherent technology leaves most of the data in its original location to work on and only transfers the necessary data. This increases the efficiency of these connections while reducing the amount of traffic that’s going across that same channel to improve the overall performance.

Reducing data movement and providing high-bandwidth, low-latency interfaces for moving data when required are key to maximizing performance and minimizing both power consumption and latency. New technologies, such as Compute Express Link (CXL), are addressing the massive amount of data transfer.. Synopsys supports the latest CXL standard, as well as other leading technologies such as DDR5, PCI Express®, and SerDes in the DesignWare IP portfolio.

 

Networking Infrastructure

When it comes to networking, the name of the game is speed. Network infrastructure is a critical component of user experience on most modern applications. Prominent Ethernet switch vendors are already developing 800-Gbps switches based on 112G SerDes. As data volume continues to increase, 1.6-Tbps Ethernet will likely be introduced. This drives the need for reliable IP to improve designer efficiency. To enable faster and more reliable designs, Synopsys offers a complete 200G/400G and 800G Ethernet controller and PHY IP solution.

 

Storage

The latest advances in the storage industry are facilitating the management of growing amounts of data, as well as the use of accelerators to process data. These include the use of computational storage, persistent memory, cache coherent interfaces to persistent storage, and next-generation memory interfaces for higher data transfer speeds. Today, NVMe storage devices are increasingly adopting PCIe 5.0 interfaces to increase SSD throughput to 4 GBps per PCIe lane — a 4x speed increase from PCIe 3.0.

The Synopsys DesignWare Interface IP portfolio supports high-speed protocols such as PCIe, USB, and DDR and is optimized to help designers meet their high-throughput, low-power, and low-latency connectivity for cloud computing storage applications.

 

Security

Data protection is critical for HPC-based cloud computing. To accurately protect the privacy, integrity, and availability of data to authorized users, standards organizations are incorporating security requirements into data interface protocols.

We offer a broad portfolio of highly integrated security IP solutions that use a common set of standards-based building blocks and security concepts to enable efficient silicon design and the highest levels of security in HPC applications. These security IP solutions help prevent a wide range of evolving threats in connected devices such as malware, data breaches, theft, tampering, and side channel attacks.

HPC: Closer to Home Than You Think

Even as industry insiders, we are continuously impressed with the pace and innovation of semiconductor design. It truly is the engine behind every aspect of technical progress in our lives. And the growth feeds upon itself: more horsepower begets more data, which requires more horsepower. This makes high-performance computing a foundation of convenience, safety, automation, and communications that we often take for granted.

Finding the right combination of performance, power, and area in the smallest geometry technologies requires implementing new architectures with silicon-proven semiconductor IP solutions. So, the next time you fire up a Netflix movie or log in to a Zoom call, think about the powerful HPC systems that are delivering a seamless digital experience at your fingertips.

Continue Reading