Skip to main content


Scaling HPC for 5G, AI, and Whatever’s Next

The past decade has seen one technology cascade into another. The massive numbers of distributed systems in the IoT led to new features in 5G. The big data generated by these connected devices contributed to the rise of AI, which helps automate analytics.

With each successive advancement, computing requirements have increased. As a result, data-driven organizations of all types and sizes are discovering that they need high-performance computing (HPC) platforms.

Let’s take a look at the implications for system engineers.

Cascading Toward Edge Analytics: A 5G Case Study

Consider the requirements of 5G networks. The International Telecommunications Union (ITU) announced in 2017 that 5G mobile cells must support up to 1 million devices per square kilometer. They must also provide a total download capacity of at least 20 Gbps and total upload capacity of 10 Gbps. By contrast, 4G small cells support roughly 100,000 devices per square mile and a peak data rate of about 1 Gbps.

Figure 1 shows the projected data density of 5G networks in major metropolitan areas. To prevent this level of data from clogging backhaul networks, operators are likely to encourage larger-scale data analytics closer to the network edge. Operators who don’t adopt a 5G edge analytics strategy are likely to be penalized with higher costs based on the amount of traffic they generate.

Figure 1. The data density of proposed 5G will require increased processing and storage performance in edge networks for analytics, particularly in urban areas. (Source: Fierce Wireless)

But the vast quantities of data make it difficult to move analytics to the edge. The workloads for 5G edge analytics are comparable to those of big data systems, which typically reside in the cloud or data center. There, workloads can be distributed across vast compute, storage, and networking resources. In contrast, 5G analytics must be more lightly distributed across vast numbers of edge network endpoints.

An AI-Centric Performance Boost

One solution gaining popularity for 5G edge analytics is AI, as it can be used to automate mid- to large-scale data filtering across the 5G infrastructure. Bringing AI-enabled traffic filtering to the rapidly changing demands of these network workloads requires scalable multicore solutions. Here, Intel® Xeon® Scalable processors are one strong option.

Intel Xeon Scalable processors provide HPC platforms a process that can go from four to 28 cores and from one- to eight-socket configurations. In an eight-socket system, for example, Intel Xeon Platinum devices support a maximum of 224 cores.

The processors offer a number of HPC-friendly features to complement their high core counts. For example, they integrate Intel® Advanced Vector Extensions 512 (Intel® AVX-512). These 512-bit vector instructions can execute 32 double precision or 64 single precision floating-point operations per clock cycle. Depending on the specific instruction set, the technology also adds up to two fused multiply-add units to accelerate math-intensive operations.

Intel AVX 512 offers a significant advantage in use cases like principal component analysis (PCA). This machine learning methodology can be used to convert a set of values into linear variables, which can be used by network operators for traffic shaping or enterprises for big data insights.

PCA can be executed using popular machine learning (ML) libraries like MLlib that runs on Apache Spark. But MLlib can be bogged down quickly by larger data sets. Libraries that use Intel AVX 512 can like the Intel® Data Analytics Acceleration Library (Intel® DAAL) can address this problem. As shown in Figure 2, implementing a couple of simple calls to Intel DAAL in MLlib code for PCA can result in significant workload acceleration, which multiplies as the size of the data set increases.

Figure 2. The MLlib machine learning (ML) library that runs on Apache Spark can be leveraged for principal component analysis (PCA) in data analytics, but its performance doesn’t scale well as data sets grow in size. (Source: Intel®)

The processors also feature a reconfigured mesh cache and memory architecture, which reduces latencies and delivers faster access to data sets required by AI and big data analytics applications. In addition, the devices provide a 50 percent increase in memory channels and 20 percent more PCIe lanes than previous-generation processors to support smaller-scale compute clusters.

Keeping 5G Data Flowing

These are just the start of the features that benefit 5G and other HPC applications. For example, the growing amount and diversity of data traffic to be carried over 5G networks necessitates different levels of quality of service (QoS). Intel® Virtualization Technology (Intel® VT) allows cores to be allocated to specific control and data plane networking tasks for various traffic types.

In addition, the larger number of nodes that must be supported by 5G requires switch fabrics that support more devices than at present without losing performance or reliability. Intel® Omni-Path Architecture (Intel® OPA) fabric supports more than 10,000 nodes, while also providing a bit error rate (BER) multiple orders of magnitude lower than InfiniBand enhanced data rate (EDR) at a significantly higher 8-byte message passing interface (MPI) rate (Figure 3).

Figure 3. Intel® Omni-Path Architecture (Intel® OPA) offers a significantly higher 8-byte message passing interface (MPI) rate than enhanced data rate (EDR) InfiniBand. (Source: Intel®)

To assist scaling 5G edge compute capacity as traffic increases, engineers are turning to platforms like the PL-81890 HPC from WIN Enterprises. The PL-81890 HPC is a high-density control server with dual Intel Xeon Platinum or Gold Scalable processors and 12 storage bays integrated into a compact 2U chassis (Figure 4). The system also supports WIN Enterprise’s Trusted Platform Control modules to help maintain the integrity of systems handling sensitive communications.

Figure 4. The WIN Enterprises PL-81890 HPC is a 2U trusted platform for demanding 5G, data analytics, and artificial intelligence tasks. (Source: WIN Enterprises)

Intel OPA host fabric interface adapter cards can be inserted into platforms like the PI-81890 HPC using one of the platform’s PCIe Gen. 3 x16 slots. In addition to individual system performance delivered by Intel Xeon Scalable processors, these expansion options bring 100 Gbps Ethernet bandwidth out of the chassis for connecting large HPC system clusters.

Scaling Further

Although the performance metrics described may be on the high end for many applications, the trajectory of technology over the past 50 years has shown that demand for more processing power and faster networking is omnipresent. For organizations looking to start small, Intel Xeon Gold, Silver, and Bronze processors provide an entry point that allows scalability in the future.

Regardless of the device, the Intel Xeon Scalable processor family offers the compute, storage, virtualization, and networking infrastructure needed for 5G, big data analytics, and AI applications today—and a path to tackling the enabling technologies of tomorrow, whatever they may be.

About the Author

Brandon is a long-time contributor to going back to its days as Embedded Innovator, with more than a decade of high-tech journalism and media experience in previous roles as Editor-in-Chief of electronics engineering publication Embedded Computing Design, co-host of the Embedded Insiders podcast, and co-chair of live and virtual events such as Industrial IoT University at Sensors Expo and the IoT Device Security Conference. Brandon currently serves as marketing officer for electronic hardware standards organization, PICMG, where he helps evangelize the use of open standards-based technology. Brandon’s coverage focuses on artificial intelligence and machine learning, the Internet of Things, cybersecurity, embedded processors, edge computing, prototyping kits, and safety-critical systems, but extends to any topic of interest to the electronic design community. Drop him a line at, DM him on Twitter @techielew, or connect with him on LinkedIn.

Profile Photo of Brandon Lewis