eWEEK content and product recommendations are editorially independent. We may make money when you click on links to our partners. Learn More.
ARM officials at the company’s TechCon 2015 show last week said that even though there hadn’t been many high-profile announcements around servers running on the company’s low-power architecture, there is a lot of work being done behind the scenes.
However, between that show and the SC 15 supercomputing show this week in Austin, chip manufacturers and server makers are making their cases heard.
Five system makers at the supercomputing show unveiled new servers that are powered by Cavium’s 64-bit ARMv8p-A chip, ThunderX. That came a week after Applied Micro President and CEO Paramesh Gopi at the TechCon show unveiled the next generation of its ARM-based 64-bit system-on-a-chip (SoC)—X-Gene 3—which he said will compete for higher-end data center workloads. At SC 15, Applied Micro announced the X-Tend integrated interconnect technology that will appear in the new chip when it starts sampling later next year.
ARM officials for several years have been talking about extending the company’s low-power chip designs—found in most smartphones and tablets—into the data center, challenging Intel’s dominance in servers. Company executives have said over the past few months that they expect to capture 25 percent of the market by 2020, but as yet, there are few ARM-based systems on the market, with Hewlett Packard Enterprises’ (HPE) Moonshot systems among the best known. There hasn’t been much in the way of news regarding ARM servers in recent months.
However, that has changed over the course of seven days. On the same day that Applied Micro introduced X-Gene 3, officials with EMC, HPE and Morgan Stanley talked about testing they’re doing with ARM-based systems. End-user interest comes not only from a desire for low-power systems in the data center, but also for a second source of silicon for competition and supply chain reasons.
At SC 15, the momentum continued. Penguin Computing, Gigabyte, E4 Computer Engineering, Inventec and Wistron all announced systems based on Cavium’s 48-coreThunderX SoC. The system makers—most of them original design manufacturers (ODMs), which sell directly to end users—are offering a mix of one- and two-socket systems that come in 1U (1.75-inch) and 2U (3.5-inch) form factors. Officials with most of the vendors said they were responding to customer demand for ARM-based systems.
“Penguin has invested in our new ThunderX-based Valkre server families based on the competitive advantage it can bring to our customers,” Jussi Kukkonen, director of product management at Penguin, said in a statement. “Customers have already begun placing orders for these systems, and we are deploying in the high-performance computing, financial services and other market sectors.”
Chen Lee, sales director for Gigabyte USA, said in a statement that the company began working with Cavium in late 2014, and it already has shipped an array of products to customers.
“Our entire portfolio of production systems is now available for order and we are shipping production platforms to customers,” Lee said. “We are seeing strong demand for these platforms and we expect the demand to further accelerate in 2016.”
Steve Cumings, director of market development at Cavium, said in a statement that “customers have been testing 48-core Cavium ARMv8 ThunderX-based servers for months, and have reported significant benefits in workload performance, reduced power consumption, and overall TCO [total cost of ownership].”
For its part, Applied Micro has its first X-Gene SoC on the market and is currently sampling the second generation. However, with X-Gene 3, the company expects to challenge Intel in the higher end of the server market. When ARM officials first began talking about moving up into the data center, the initial target were smaller micro servers for environments that demand density and power efficiency.
System Vendors Unveil ARM-Based Server Offerings
However, executives with both Applied Micro and Cavium said they intended to power systems for mainstream workloads, and Applied’s Gopi said at the TechCon show that X-Gene 3 will compete with Intel’s mid-range Xeon E5 and high-end E7 processors. X-Gene 3 will house 32 cores running up to 3GHz, eight DDR4 memory channels and 42 PCIe Gen 3 lanes. It will be manufactured via Taiwan Semiconductor Manufacturing’s 16-nanometer FinFET process and will offer four to six times the performance of the current X-Gene chips, according to Gopi.
At the TechCon show, the CEO teased the integrated connectivity technology that will come with X-Gene 3, but said more details would come at SC 15. The company on Nov. 16 introduced X-Tend, which officials said will enable multiple generations of X-Gene SoCs to be connected with large, elastic memory pools. Emerging workloads like data mining, real-time analytics and machine learning demand minimal latency and flexible, immediate provisioning of compute and memory resources, they said.
X-Tend is made up of on-chip and software components that will not only connect multiple X-Gene nodes to each other, but also show multiple terabytes of shared DRAM as a single cluster. Evaluation platforms for X-Tend are shipping to Applied Micro customers, officials said.
“X-Tend utilizes open standards, including PCI Express, to provide multi-socket configurations with a single symmetric multiprocessor OS image, accessing huge memory pools with very low latency,” Gopi said in a statement. “On-demand scale-up capability is particularly applicable to compute-intensive workloads like high-performance computing and advanced search algorithms.”
He said last week that X-Gene 3 will be a key product for driving ARM’s push to gain 25 percent of the server market. ARM officials at TechCon said they were confident in their ability to hit that goal, both because of the chips being developed and the growth of the surrounding software ecosystem. However, they will have to overcome significant challenges. Intel is continuing to drive down the power consumption of its chips, and is adding accelerators like GPUs and field-programmable gate arrays (FPGAs). At the same time, the company is driving development of its upcoming “Skylake”-based “Purley” Xeons, which reportedly will come with improved performance, a new memory architecture, an integrated network fabric and an integrated accelerator option.
In addition, ARM is facing competition from IBM and its OpenPower Foundation—which was created to expand the use of the Power architecture—for the position as the data center alternative to Intel. At SC 15, IBM announced a partnership to expand the use of FPGAs in Power systems.