Cavium makes its ARM for data centers push with new servers

64-bit CPUs are meant for HPC, not simple tasks.

bridgestone data center 1968
Bridgestone

The initial efforts to bring ARM-based processors in the data center were not terribly successful. Calxeda crashed and burned spectacularly after it bet on a 32-bit processor when the rest of the world had moved on to 64-bits. And HPE initially wanted to base its Project Moonshot servers on ARM but now uses Intel Xeon and AMD Opteron.

That’s because the initial uses for ARM processors were low-performance applications, like basic LAMP stacks, file and print, and storage. Instead, one company has been quietly building momentum for high performance ARM processors, and it’s not Qualcomm.

Cavium, a company steeped in MIPS-based embedded processors, is bringing its considerable experience and IP to the ARM processor with its ThunderX server ecosystem. ThunderX is the whole shootin’ match, an ARMv8-A 64-bit SoC plus motherboards, both single and dual socket. In addition to hardware, Cavium offers operating systems, development environments, tools, and applications.

The company has announced its second generation of platforms for data centers, called ThunderX2. ThunderX2 is optimized to drive high computational performance delivering high speed memory bandwidth and optimized memory capacity. The new line of ThunderX2 processors includes multiple workload optimized SKUs for both scale up and scale out applications and is fully compliant with ARMv8-A architecture specifications as well as the ARM Server Base System Architecture and ARM Server Base Boot Requirements standards.

Penguin Computing is the first ODM to announce a ThunderX2 platform. Its Tundra Extreme Scale (ES) server platform is now available for order and will ship in the third quarter. Penguin Computing provides customized build-to-order server solutions for customers with specialized hardware requirements in enterprise, financial, federal government, bioinformatics and Internet segments.

The ThunderX2-based servers will focus on highly-scalable hyperscale and HPC-type workloads, including Big Data, large-scale graph analytics, molecular dynamics, and Ceph / Cloud storage.

It seems a lot of people really underestimated ARM in the early days. They assumed that because it was used in cell phones and tablets, and embedded uses like network fabrics, that it wasn’t a good match for the data center. But a few years of frantic R&D on the part of ARM and third parties like Cavium and Qualcomm are closing the gap with Intel’s Xeon.

Of course it remains to be seen how the benchmarks come out. It has to best Intel in performance per watt, which is one of its chief selling points. Those numbers have not really emerged yet, at least from independent testers.

Join the Network World communities on Facebook and LinkedIn to comment on topics that are top of mind.
Must read: 10 new UI features coming to Windows 10