Cavium is making its ThunderX 64-bit ARM server chips generally available, and is rolling out reference architectures to make it easier for system makers to embrace the multicore systems-on-a-chip.
Cavium first unveiled the ThunderX chips in June at the Computex show, saying the ARM-based SoCs give server OEMs an alternative to the Intel x86-based processors that dominate the data center. In addition, Cavium officials said the chips would have enough performance and power efficiency to run in mainstream servers, and not only low-power microservers aimed at hyperscale cloud and Web environments.
Company officials on Dec. 1 said the ThunderX CN88xx SoCs—for single- and dual-socket systems—are generally available. In addition, Cavium unveiled reference platforms—the ThunderX 1K in an ATX form factor motherboard in a 1U (1.75-inch) chassis for single-socket systems, and the 2K for dual-socket servers in a half SSI form factor sled for a standard 2U (3.5-inch) chassis that can hold up to four sleds.
Cavium is offering the reference platforms not only for hardware makers but also for software developers, according to Rishi Chugh, director of ThunderX product marketing at the chip maker.
“ThunderX based reference systems deliver the industry’s most comprehensive ARMv8 based server class platforms for hyperscale data center, cloud servers, big data and scale-out storage systems,” Chugh said in a statement. “Cavium is also aggressively enabling our ecosystem partners to leverage these platforms to accelerate software validation and optimization on ThunderX.”
Officials with ARM—whose low-power SoC designs can be found in most smartphones and tablets—have for several years been talking about pushing the company’s architecture up into the data center, where businesses increasingly are looking for systems that offer both performance and power efficiency. Chip makers like Cavium, Applied Micro and Advanced Micro Devices are leveraging the company’s 64-bit ARMv8-A architecture to create SoCs for servers.
ARM and its chip-making partners are hoping to chip away at the dominance of Intel, which is answering ARM’s challenge by driving down the power consumption of its Atom and Xeon server chips. Intel already is on its second-generation Atom chip for low-power servers, the C2000 “Avoton” processor, and next year will come out with the next iteration, the 14-nanometer “Denverton” chip. Intel also is sampling Xeon D, the company’s first Xeon SoC, which is aimed at dense, low-power servers as well as networking and storage devices.
ARM officials expect it will take about two years for the ARM-based server market to hit its stride. The company’s server efforts have seen some challenges, from pioneer Calxeda closing its doors a year ago after running out of money to Nvidia and Samsung reportedly pulling back on plans for ARM-based server chips. However, after several years of talk, products are beginning to hit the market. Hewlett-Packard announced in September plans to put Applied Micro’s X-Gene SoC into a Moonshot server model, and Dell has ARM-based servers in its labs running on Cavium and Applied Micro chips. In addition, Qualcomm—the world’s largest mobile chip maker—will be getting into the ARM server chip space, and supercomputer Cray will evaluate using ARM-based SoCs in some of its systems. Cray is working with Cavium to analyze ARM-based systems in supercomputing environments. The two companies will create Cray compute clusters powered by Cavium’s 48-core ThunderX SoCs.
Cavium’s chips come with such features as cache coherency across dual sockets using the company’s Cavium Coherent Processor Interconnect, up to four DDR3/4 72-bit memory controllers, full system virtualization via Cavium’s virtSoC technology and a low-latency Ethernet fabric.
The ThunderX SoCs come in four families that come with custom I/O and accelerators and that are optimized for particular workloads. ThunderX CP is aimed at such tasks as public and private clouds, Web server and social media analytics, while the ST family is for cloud storage, big data and distributed databases. NT is for telecommunications environments, network-functions virtualization (NFV) deployments and embedded networking applications, and SC is for secure computing.