Intel may be best known for its x86 server chips, but its growing silicon portfolio targets cloud and edge applications, as well as 5G.
Its data center dominance, however, traces back to its x86 chips. In 1998 the company launched its Xeon processors — a brand of x86 chips — and it celebrated the processors’ 20th anniversary this year. The chip giant has sold more than 220 million of these processors over the past two decades, generating $130 billion in revenues.
The Xeon processor “is where we started bifurcating, investing more heavily in the acceleration and the set instruction, specifically for data-center workloads,” said Jennifer Huffstetler, VP and GM of data center product management.
This includes investing in the chips’ capabilities related to SDN and NFV. “Every year we make continual enhancements and improvements in VM performance, VM migration, which becomes really important for cloud architectures,” Huffstetler said.
Intel launched its Xeon Scalable processors in July 2017. At the launch, company execs said early-access customers using the integrated AVX-512 instructions saw more than 100 percent performance improvement compared to the earlier chips. Some of the early access customers using the new chip include AT&T, Google, Cisco, and Amazon Web Services (AWS).
“Here we are a year out and all 10 of the top 10 communication service providers around the globe have deployed Xeon Scalable — AT&T, Verizon, Telefonica, China Mobile, every major player,” Huffstetler said.
The Xeon Scalable processors also added a new encryption instruction set that accelerates the encryption of data as well as security features such as the Intel Key Protection technology. These new capabilities aim to improve security without hurting performance. “As security becomes more important, we’re removing the performance overhead that was previously required to run those algorithms,” Huffstetler said.
Then, in January, came the security vulnerabilities disclosed by Google’s Project Zero team and heard round the world: Spectre and Meltdown, which affect Intel x86 and other processors. As late as last month Intel disclosed new Spectre-like vulnerabilities that could allow hackers to access data that is supposed to be secured in protected areas on its Core chips or in the cloud.
The most recent security flaw disclosures come as Intel is planning to ship the next generation of its Xeon Scalable processors by the end of this year. “It will include hardware mitigations for Spectre and Meltdown,” Huffstetler said, referring to the next-gen processors. When asked if Intel will have to sacrifice some chip capabilities to avoid future security threats, Huffstetler, in short, said no: “Security is always top of mind, and we’re always architecting to deliver greater performance and greater security at the same time.”
Intel also makes modem chips for 5G devices and has been working with telecommunications equipment partners such as Nokia, Ericsson, and Huawei, as well as communications service providers like AT&T, Verizon, Telia, Telstra, Korea Telecom, SK Telecom, NTT DoCoMo, China Mobile, and China Unicom on 5G trials. These trials use its 5G Mobile Trial Platform (MTP)— a small, portable platform that allows for the development and testing of proto-5G features and mmWave technologies, devices, and network capabilities.
“Intel has participated in trials worldwide, achieving a number of ‘firsts,’” said Asha Keddy, Intel VP and general manager of next generation and standards. “These include enabling 5G use cases with the first 5G 4K golf broadcast at the 2018 US Open, the first smart stadium in China, and the first industrial 5G trial in EMEA.”
Huawei and Deutsche Telekom used Intel technology to complete the first 5G non-standalone (NSA) new radio (NR) interoperability trial in January as did Ericsson and China Mobile on the first 5G standalone (SA) NR interoperability trial in June. That same month, Intel participated in the first multi-vendor 5G NSA NR interoperability and development testing based on the 3GPP Release 15 March specifications.
In recent years, Intel has expanded its silicon portfolio to include specialized field programmable gate arrays (FPGAs).
An FPGA is an integrated circuit that can be programmed anytime, even after the device has been shipped to customers. Because of this programmability, combined with high throughput and very low latency, FPGAs are ideal for many cloud and edge applications, and they will be critical for 5G networks. Intel says its Programmable Solutions Group revenue has grown double digits as customers use FPGAs to accelerate AI, among other applications.
In July, Intel announced plans to buy eASIC, a privately held programmable chip company based in Santa Clara, California. eASIC makes structured ASICs, which is an intermediary technology between FPGAs and ASICs.
“Specifically, having a structured ASICs offering will help us better address high-performance and power-constrained applications that we see many of our customers challenged with in market segments like 4G and 5G wireless, networking, and IoT,” said Dan McNamara, SVP and general manager of the programmable solutions group at Intel. “We can also provide a low-cost, automated conversion process from FPGAs (including competing FPGAs) to structured ASICs. Longer term, we see an opportunity to architect a new class of programmable chip that takes advantage of Intel’s Embedded Multi-Die Interconnect Bridge (EMIB) technology to combine Intel FPGAs with structured ASICs in a system in package solution.”
Earlier this month Intel bought NetSpeed Systems, a company that provides system-on-chip (SoC) design tools and interconnect fabric intellectual property. NetSpeed’s network-on-chip (NoC) tool automates SoC front-end design and generates programmable interconnect fabrics. This becomes more important as SoCs — like Intel’s FPGAs— grow more complex and specialized.
Intel’s IoT group makes hardware platforms with components including the microprocessor, chipset, stand-alone system-on-chip (SoC), or multi-chip package. This makes up the bulk of the group’s momentum, the company says.
Segments including retail, industrial, and video are early adopters. But Intel says imaging and video use cases — especially those utilizing AI — span nearly every segment. To this end the company recently launched a toolkit designed to fast-track development of high-performance computer vision and deep learning inference applications at the edge.
The toolkit is compatible with open source frameworks like Caffe and TensorFlow, and it works with Intel’s traditional CPUs, CPUs with integrated graphics, or chips specially made for AI inference — like the FPGA — chips and the Intel Movidius vision processing unit (VPU) that entered production earlier this year.