ARM chip partners Cavium and Marvell are using the Computex 2016 show this week to further their case for the use of low-power ARM-based processors in the data center.
Cavium, one of the first vendors to bring ARM-based data center systems-on-a-chip (SoCs) to market, introduced the next generation of its ThunderX chips that officials said will target data centers and cloud providers with a range of workload-optimized variants. The ThunderX2 will provide two times to three times the performance compared to ThunderX, which the company first launched in December 2014.
For their part, Marvell officials announced the availability of production samples of the Armada 7000 and 8000 families of 64-bit ARM-based SoCs that can address an array of areas, from data centers and IP appliances to small and midsize businesses (SMBs) and small-office and home-office (SOHO) applications. The new chips are the first built on Marvell's MoChi (modular chip) architecture, an approach that centers on taking the various parts of the SoC and making them into modular chips of specific functions that can be interconnected in a way that they can act as though they are integrated onto a single piece of silicon.
Both chip lineups already are being sampled and put into designs by tier-one system OEMs, Marvell officials said. They won't go into volume production until later in 2017.
The announcements are part of a larger push by ARM and its partners to bring ARM's low-power SoC designs—which are found in most smartphones and tablets—into the data center to challenge Intel's dominance in the market. Intel, with its broad range of Xeon chip offerings, owns more than 95 percent of the server chip space and is making aggressive moves in other data center systems, including networking gear and storage appliances.
However, ARM and its partners—as well as IBM and its OpenPower initiative—believe businesses and service providers are looking for alternatives to Intel, or at least a second supplier. Cavium and Applied Micro are among the first chip makers to bring 64-bit ARM-based SoCs to the data center, though a growing number of other vendors—including Qualcomm and Advanced Micro Devices—also are making moves in that direction.
The performance capabilities of the ThunderX and ThunderX2 SoCs enable Cavium to compete favorably against Intel, according to Gopal Hegde, vice president and general manager of the company's Server Processor Group.
"Some of the biggest OEMs and ODMs [original design manufacturers) are shipping systems to end users, but they're afraid of the big guy (Intel) so they don't want to talk publicly about it," Hegde told eWEEK.
Despite the growing number of ARM-based server SoC vendors, the competition is still Intel, he said.
"We don't run into other ARM vendors at all," Hegde said. "We have such a huge advantage over other ARM vendors."
Hegde and other Cavium officials are hoping to press that advantage with ThunderX2, which is built in a 14-nanometer FinFET process and is compliant with the ARMv8.2 architecture and ARM's Server Base System Architecture standard. The chips are optimized for a wide variety of data center workloads, including compute, security, storage, data analytics, distributed databases and network-functions virtualization (NFV), they said.
The processors target one- and two-socket systems and include four workload-optimized chips for disparate workloads. The ThunderX2-CP SoCs are aimed at cloud compute applications, including Web serving, Web caching, search and commercial high-performance computing (HPC) workloads. The ThunderX2-ST chips are for such applications as big data, cloud storage, massively parallel processing databases and data warehousing, while the ThunderX2-SC is for secure Web front-end and security workloads. They support multiple Ethernet interfaces—including 10/25/40/50/100GbE—and PCIe Gen3 interfaces. The ThunderX2-NT targets media servers, scale-out embedded applications and NFV-style workloads.
The chips have speeds of 2.4GHz and 2.8GHz, twice the single-thread performance of ThunderX; up to 54 cores and six DDR4 memory controllers per socket; full system virtualization via Cavium's virtSOC technology; and integrated hardware accelerators, including Octeon and vSwitch offload.
Marvell officials said the MoChi architecture will enable vendor to address a wide array of market needs.
"Areas in compute, scalability and the software ecosystem are increasingly being presented with challenges across many industries," Michael Zimmerman, vice president and general manager of connectivity, storage and infrastructure business units at Marvell, said in a statement. "We believe we have addressed these challenges head on by leveraging our MoChi architecture. Our new SoCs are designed to deliver top performance, efficiency and design integration and support full virtualization."
The new dual- and quad-core Armada 7000 and 8000 families can integrate multiple 10GE ports with packet processor and offload engines for security and storage, officials said.