High Performance Computing is no longer the premise of large government deployments; the compute power is now highly relevant for the enterprise segment as well. ARM, the semiconductor IP company that is now working with various formats – embedded and open platform – to bring HPC technology to more and more devices and to enterprise solution. Darren Cepulis, Data Centre Architect and Evangelist, ARM Marketing and Business Development tells HPC Asia how the company is constantly collaborating with solution providers like HPE, TSMC to bring products and services that improve compute power in data centres and network infrastructures. An accomplished technologist with over 25 years of experience leading firmware architecture, strategy, standards, and design, Cepulis focuses primarily on HPC, enterprise and cloud-oriented servers, storage, and software ecosystem. At ARM he is driving the company’s HPC strategy and ecosystem with partner companies worldwide. He shares with us on how the HPC market is moving and what ARM is doing in it.
How would you define HPC today? At what levels are the enterprises today in terms of adopting HPC within their organisations?
HPC is about maximizing the amount of precise compute that can be applied to a specific problem, with technologies focused on parallelism, interconnects, and data management. Large HPC clusters can run well above 1 Peta-FLOP of compute power and HPC technology is prevalent throughout the world today. Originally seen as a technology that was only affordable by large government labs or universities, we now see more than half the market for HPC supporting enterprise business use cases such as manufacturing, biotech, weather, engineering, and finance, just to name a few. There really isn’t an industry or vertical where HPC isn’t being applied to solve the science of the day.
What is the need that is driving the enterprises to adopt cloud and look beyond to HPC?
We are seeing HPC workloads leverage cloud-based infrastructure for many of the same economic reasons that enterprise datacenter workloads leverage a cloud solution. An organization can rent an HPC cluster for a short period of time as needed, rather than making a multi-million dollar capital investment in an HPC cluster.
What is your roadmap for HPC? What kind of solutions are you working on to help partners develop HPC solutions?
ARM is the world’s leading semiconductor IP company, and we deliver IP that has been optimized in terms of performance, power-consumption, and design area across a wide variety of markets, including HPC. We enable both embedded and open platform standard designs that target compute, storage, and IO devices for the datacenter. Without going into specifics, our infrastructure roadmap is enabling a greater range of core counts, clock speeds, and interconnects at cutting edge silicon manufacturing nodes. For example, we recently announced collaboration with TSMC targeting 7nm FinFET technology for high-performance compute. Additionally, ARM is investing in the analysis of workloads that server chips will need to undertake. This will help our partners to build more optimized chips and enable ARM to design more advanced processors for future servers.
Cloud computing is fast becoming the norm and this is disrupting the existing IT infrastructure within an enterprise. How can the organisations prepare for this disruption?
I don’t view cloud computing as a disruptor, but rather as an enabler of better, more optimized choices and an augmenter of existing IT infrastructure. Cloud computing is valuable to organizations due to its flexibility, nimbleness, and ease-of-management. IT managers can more quickly apply or remove workload-optimized services to their businesses. IT organizations that best understand their businesses’ varying IT demands and use will be able to optimally allocate workloads to cloud resources versus internal infrastructure.
How is ARM developing technologies/solutions that enable businesses to adopt cloud computing easily?
Being a provider of system on chip (SoC) IP and an enabler of dense, energy-efficient chip designs, ARM is working to reduce the overall TCO for IT providers in both cloud and traditional data-centers. Furthermore, ARM invests in the optimization of cloud-related software technologies such as Xen, KVM, and Openstack running on ARM-based platforms.
How are you looking at standardizing the cloud-computing platform for easier and faster deployment?
Cloud service providers standardize within their organizations on specific server designs and software platforms. ARM is very cognizant of the need for platform standards in order to streamline the deployment and management of servers in the data-center. We publish our own set of server hardware and firmware requirements specifications for our infrastructure SoC partners in order to drive a consistent baseline implementation for the overall ecosystem.
How will this help organisations take a step towards adopting HPC?
By providing standards based platforms to the industry, our partners enable a wide variety of OS and application choices for the end-customers, be they cloud, HPC, or hyperscale focused. All of today server markets, including HPC, are commodity-driven, where TCO criteria such as cost, power-efficiency, and density are most important. These markets should benefit from the variety of optimized infrastructure solutions that ARM enables.
Can you give us a technology/solution roadmap that organisations should look at or plan for to be future ready?
At present, there are server and HPC SoC products available from partners such as Applied Micro, AMD, and Cavium. These are all innovative 64-bit SoC’s that target a variety of prevalent data-center and HPC workloads. Looking ahead, we expect to see an even greater variety of SoC solutions as partners such as Qualcomm, Broadcom, and HiSilicon reach production.
Darren Cepulis is a data center architect and evangelist within the ARM Marketing and Business Development segment, driving the company’s HPC strategy and ecosystem with partner companies worldwide. An accomplished technologist with over 25 years of experience leading firmware architecture, strategy, standards, and design, he focuses primarily on HPC, enterprise and cloud-oriented servers, storage, and software ecosystem. Before joining ARM, he worked at Compaq and HP leading design efforts for industry changing servers such as the Systempro, ProLiant, and Moonshot servers. Most recently, he was Lead Software Architect for HP’s Extreme Scale-out Server Products Division, where he drove new modular hardware and processor agnostic architecture that supported both Intel and ARM compute modules, as well as new rack and enclosure level management architecture. He was recognized by HP as a Distinguished Technologist. The holder of 47 patents, he has a Bachelor’s Degree in Computer Engineering from the University of Illinois.