Dan's Take

Cray Launches XC50 Compute Blade For ARM Processors

It's more evidence of increasing datacenter fragmentation.

The industry is continuing to see a move away from a uniform architecture in the datacenter to a more fragmented environment. For example, Cray just announced a compute blade addition to its XC50 system, based on the following:

  • Cavium's ThunderX2 ARM processors and a version of Cray Linux (based on SUSE SLES 12)
  • FORTRAN and C/C++ compilers
  • Software libraries designed to support "extreme parallel processing"
  • Storage and processing virtualization technology.

The ARM-based blade is an addition to the XC50 that Cray launched just over a year ago. At the time Cray announced its XC50, it supported a number of different processor architectures, including the NVDIA Tesla P100 GPU accelerator, Intel Xeon and Xeon Phi processors. At the time, the company said that this system was designed to target organizations focused on the uses of "deep learning algorithms."

With the announcement of this new compute blade, the XC50 now has the capability to support a combination of X86, NVDIA and ARM workloads simultaneously, to address organizational requirements for technical and high-performance computing.

ThunderX2 ARM Processors
Cavium's ThunderX2 ARM processor family is a 64-bit custom system-on-a-chip (SOC), based on ARM's server-based architecture standard (ARMv8). It integrates a large number (48) of 64-bit cores, Cavium's Coherent Processor Interconnect (CCPI) to assure a consistent cache for all of the processors, an integrated I/O capacity with hundreds of Gigabits of I/O bandwidth, and multiple DDR4 72-bit memory controllers capable of supporting high-performance server memories with 1+TB of memory in a dual socket configuration.

The goal of this family of SOCs is supporting high-performance workloads while consuming a small amount of power and producing limited heat when compared with SOCs based on other processor architectures.

The Evolution of ARM
ARM Limited, the source of the ARMv8 standards, has developed multiple RISC computing architectures and the architectures of related support devices. It works with more than 1,000 partners to deliver microprocessors and SOCs designed to support powerful/low power consumption applications. The company claims that more than 90 billion processors based on this architecture have been shipped. Suppliers such as Apple, Hitachi, Samsung and many others are building SOCs around this architecture.

If you use a smartphone or a tablet, it's highly likely you're using one of these ARM-based devices.

Dan's Take: the Return of Fragmented Computing
In an article published in October 2016, I commented on the possible return of "fragmented computing" -- computing based upon a number of different processor architectures -- after the industry had seen a steady move toward a uniform, x86-based environment.

Enterprises have been moving back to the position that a single processor architecture really can't address all of their possible computing requirements. This, of course, doesn't mean that they're in the process of abandoning the venerable x86 architecture. It's just that some workloads need features such as higher levels of parallelism, lower cost, lower power consumption or lower heat production.

To address this industry trend, software suppliers, such as Red Hat, SUSE, Canonical and Microsoft, are offering operating systems to support a number of processor architectures that include x86, NVDIA and ARM. They're also offering development tools and application frameworks as well as application, processing, storage and networking virtualization technology.

If we focus on NVDIA and ARM-based solutions for a moment, hardware suppliers including Cray, HPE and Kaleao's KMAX are now offering systems based on these processor architectures.

This also means that mainframe architecture isn't dead. IBM isn't giving up on its Power and Z architectures; I've seen a number of recent surveys showing that enterprises currently using mainframes are planning to increase their use of these systems. Software suppliers such as Compuware are doing their best to entice enterprises to reconsider these systems by offering rapid application development tools to support DevOps-based computing environments.

It's likely that the adoption of these systems will follow past trends. Handheld devices are likely to continue to use ARM. Systems focused on the support of technical, research, academic and high-performance computing workloads are likely to increasingly use non-x86 architectures to address enterprise requirements to reduce power consumption and heat production while still being able to address computational needs.

About the Author

Daniel Kusnetzky, a reformed software engineer and product manager, founded Kusnetzky Group LLC in 2006. He's literally written the book on virtualization and often comments on cloud computing, mobility and systems software. He has been a business unit manager at a hardware company and head of corporate marketing and strategy at a software company.


Subscribe on YouTube