The Open Euro HPC Project

is dedicated to re-inventing HPC
(High Performance Computing),
making it accessible, lower cost,
ultra-low power and more flexible
with unparalleled scalability.

Join the community of technology leaders working together
to break open proprietary and expensive
High Performance Computing infrastructure to achive
greater choice, customization and cost savings.

Mission

Mission

Open Euro HPC Project is dedicated to re-inventing High Performance Computing [HPC] as an integral part of a new class of facility / service: Low Cost, Ultra-Low Power, Cloud-Based, Converged Data/AI/HPC Centers; hereafter referred to as Universal Computing Centers.

An HPC explosion is already beginning to occur in the cloud. It’s not a matter of “IF”. For the EU, it’s a matter of “HOW” to leverage this unique opportunity to achieve technologic sovereignty, as well as HPC parity with global leaders China and the US. The EU has a window of opportunity – now, to have a profound effect on the coming revolution in data centers, driven in large part by HPC.

The EU is at the beginning of a major sea-state change in data center markets. Will it lead, or will it follow?

Factors motivating the high growth rate of cloud-based HPC include: most on-premises HPC hardware has aged poorly and is clearly inferior to the newer hardware in the public clouds. It’s currently much cheaper to use a public cloud’s HPC service rather than replace old in-house HPC equipment.

Open Euro HPC Project Goals

Open Euro HPC Project goals are to seamlessly integrate a new generation of Data/AI/HPC hardware & software components, optimized across all three types of workloads, with optimized facility & system architectures, to create a new class of Universal Computing Centers (UCC’s) throughout Europe and around the world.

The Open Euro HPC Project intends to attract a broad spectrum of innovators, developers and users to create reference designs for high performance, power efficient, flexible and scalable Data/AI/HPC solutions. Join our community of technology leaders, working together to break open the proprietary and expensive HPC infrastructure, in order to achieve greater availability, choice, customization, and cost savings. Success of the open Euro HPC Project will democratize High Performance Computing, enabling broad, low-cost access to state-of-the-art HPC resources, not only by governments and large corporations, but also by small to medium enterprises and individual consumers.

When Facebook and others launched the Open Compute Project (OCP) in 2011, the goal was to accelerate the commercialization of hardware designs that would enable greater data center efficiency and flexibility. Since then, OCP has gained traction in the market, growing into an influential community focused on open data center hardware & software.

Similarly, the Open Euro HPC Project is focused on supporting and enabling the growing demand for Universal Computing Centers: facilities / services that address the overlap and convergence of Data Center, AI and HPC workloads in a highly energy efficient, low cost, easy to access platform. In addition, innovative UCC business models and user experiences – ALL have to be worked on.

Open Euro HPC project represents a substantial undertaking, but these goals must be pursued if the EU intends to achieve technologic sovereignty and establish itself as a credible player in the HPC marketplace. The Open Euro HPC Project offers an unprecedented Return on Investment.

Open Euro HPC Project Goals

Vision

As High-Performance Computing is moving to the cloud at an increasingly rapid pace, handling more complex tasks and challenges, it must transition to a drastically more efficient, economical, and user-friendly form of computation. Common hardware, and a core set of common software tools must become a commoditized and evolving set of broadly available products, optimized across data center, AI, and HPC workloads. Like the Open Compute Project, we believe open collaboration and standardization are the most effective ways to get there.

Universal Processing Center

Open Euro HPC Project collaboration can bring to market, unprecedented HPC capabilities, in an environmentally friendly, widely available, low cost manner.

Our Goals are

  • to make Data/AI/HPC center hardware & software components, as well as system architectures, open and available to a broad spectrum of innovators and users, which will significantly lower cost, and
  • to create reference designs for more power efficient, more flexible and more scalable converged Data/AI/HPC solutions.

Join our community of technology leaders, working together to break open the proprietary and expensive High-Performance Computing infrastructure in order to achieve greater availability, choice, customization, and cost savings. Success of the open Euro HPC Project will facilitate the democratization of HPC, enabling broad, cost effective access to state-of-the-art HPC resources, not only by governments and corporations, but also by small to medium enterprises and even individual entrepreneurs.

By unleashing the EU’s corporate, SME, and academic prowess, focused on a common goal: the development and demonstration of Europe’s Universal Computing Center model, the Open Euro HPC Project can serve as a focal point and a critical enabler in the resurgence of EU leadership in cloud computing infrastructure and capabilities.

Convergence between HPC, Public and Private Clouds

HPC is without question moving rapidly into the cloud. Unification of HPC with Data Center and AI workload hardware and software tools will provide standardization, dramatic cost reductions, much higher resource utilization and will ultimately provide major economic value.

Infrastructure as a Service [IaaS] is a form of cloud computing that provides virtualized computing resources over the internet, while the user pays only for actual resources consumed during his HPC program run. Despite the huge volume of SaaS and PaaS solutions available within the cloud, the nature of HPC is such that IaaS servers and associated networking are likely to form the bulk of research computing cloud usage for the foreseeable future. The overhead associated with virtualization has previously been cited as the reason not to move into the cloud, but this argument is diminishing as time goes by; because actual overhead is decreasing as cloud vendors shift to custom, external silicon for managing their infrastructure.

Convergence
Relation to Open Compute

Relation
to Open Compute

The Open Euro HPC Project will build on the real-world impact already demonstrated by the Open Compute Project to bring a new generation of low cost, scalable and ultra-low power computation to the cloud, breaking the shackles of expensive legacy HPC systems. A stretch goal of the Open Euro HPC Project is to develop common, optimized hardware solutions and open source tools that address all three categories of computation – data center, AI, and HPC workloads.

Facebook started a project to design the world’s most energy efficient data center, one that could handle unprecedented scale at the lowest possible cost. A small team of engineers spent the next two years designing and building one from the ground up: software, servers, racks, power supplies, and cooling. The result now stands in Prineville, Oregon. It was 38% more energy efficient to build and 24% less expensive to run than the company’s previous facilities—and has led to even greater innovation.

Open Euro HPC is challenging its members to innovate an order of magnitude improvement in performance, energy efficiency, resource utilization, and cost of services and solutions, which will be enabled by a new generation of European Universal Computing Center components: hardware, software, system architectures, and business models.

Challenge Areas

While OCP has made tremendous progress in optimizing component designs for more efficient data centers, it has not purposefully addressed the convergence of Data Center, AI and HPC workloads within cloud-based facilities/services, and the need to optimize UCC building blocks across multiple computational disciplines.

Nevertheless, there are plenty of questions to be answered regarding integrating HPC into the cloud, which range from how to extract the best performance from the underlying hardware platform, to what services are essential to make its usage easier.

Challenge Areas

In the following paragraphs, we discuss the foundation that OCP has already provided, along with the challenge areas that must be addressed, if the vision of breakthrough EU-developed Universal Computing Centers is to be realized.

Servers

CPU & DRAM rails consume over 80% of cloud-provisioned server power at peak load.

CPU & DRAM Source: [Google Open Compute Summit presentation] The Datacenter as a Computer – Luiz Andre Barroso, Jimmy Clidaras, Urs Holzle

Open Compute Project (OCP) Server

  • Based on AMD and Intel processors, with up to 64 and 48 cores per processor
  • Maximum power efficiency, serviceability and cost efficiency from the ground up
  • Available in 1OU and 2OU form factors
  • 2OU form factor: Up to 2TB (2 x 16x DIMMs)

The Open Euro HPC Project must tackle further increases in computational density within HPC servers, as well as addressing power and cooling requirements for increasingly higher performance HPC server reference designs.

OCP efforts have resulted in major progress in HPC server design and deployment.

server

Accelerators

  • nVidia GPU in OCP Big Sur
accelerator
Rack
rack rack
rack rack
Rack Solutions OCP-approved Open Rack Enclosure

Open Euro HPC Project goals for UCC rack designs are centered on accommodating future high-computational-density HPC servers and their associated power, cooling, and top of rack switching requirements.

OCP Open Rack Enclosure Unique Features:

  • Compatible with OCP 21” Equipment
  • Rack top with cable pass through
  • Bus bars NOT included
  • Features & benefits
    • 600 mm of physical width will easily fit in existing data centers
    • Rack Units: 42OU, 48OU
    • Static load 3000 lbs
    • Configurable for up to 3 bus bars
    • Robust steel frame
Power Architecture

Google introduced a 48V rack power architecture at 2016 OCP Summit to replace 12V. Using 48V provides a 16x reduction in power distribution and up to 30% lower conversion losses.

schema

New48VtoPoLDC/DC Voltage Regulators

  • High power CPU & Memory
  • 6 direct 48V conversion VRs

Existing 12V to PoL DC/DC Voltage Regulators

PoL: Point of Load
  • Intermediate 12V voltage bus
  • Lower power rails & storage

Existing 12V to PoL DC/DC Voltage Regulators

  • Up to 30% less conversion losses
  • 16x less power distribution losses
  • Improved Upstream conversion & distribution
  • Improved UPC
  • Power losses reduced
  • Co-located & high density
  • No boost conversion required
  • Supports existing Telecom ecosystem
schema

Open Compute Casestudy

Project Objectives/Solutions

  • Eliminate 480V to 208V transformation >> Used 480/277VAC distribution to IT equipment
  • Remove centralized UPS >> Implemented 48VDC UPS System
Result - a highly efficient electrical system and small failure domain (Figure below)
schema

Total Loss Up to Server: 21 to 27%

schema

Total Loss Up to Server: 7.5%

Cooling

Cooling of the servers, storage and network gear represents a significant share of data center power consumption and its operational cost. With increased power density, a traditional air and mechanical cooling proved to be less efficient. Liquid cooling seems to be a suitable alternative.



Liquid Cooling

High performance computing has used liquid cooling for many years. These solutions have proven to be cost effective and efficient at heat extraction. With the growth of 5G, IoT, VR, CDN, and latent sensitive applications, data centers are being constructed closer to their customers and often in regions of the globe where traditional air and mechanical (e.g. chillers) cooling becomes quite expensive. Increased power density also introduces cooling challenges. In these cases liquid cooling, and specifically warm water cooling, becomes an effective alternative for heat extraction.

Networking

OCP is bringing to networking the guiding principles that it brought to servers & storage, so that end users are given the ability to forgo traditional closed and proprietary network switches - in favor of a fully open network technology stack. Our initial goal is to develop a top-of-rack (leaf) switch, while future plans target spine switches and other hardware and software solutions in the space.


Scope

  • Fully disaggregated and open networking HW & SW
  • Operating System - Linux based operating systems & developer tools, and ReST API’s
  • Fully automated configuration management & bare metal provisioning
  • Universal & Multi-Form Factor Switch motherboard hardware
  • Fully open integration & connectivity
  • Energy efficient power & cooling designs
  • Software Defined Networking (SDN)


Open Euro HPC Project will extend this philosophy to common, optimized networking for Universal Computing Centers that maximizes HPC performance without adversely impacting data center and/or AI workloads.

OCP Networking Suppliers:

Edgecore Networks
ITOCHU Techno-Solutions Corporation
Hyve Solutions


OCP Accepted solution: 100GbE Data Center Switch
- ToR/Spine Switch - PORT TO POWER

Edgecore Networks
Edgecore Networks


Without low-latency interconnects, cloud usage will be effectively impossible for massive MPI (message passing interface) jobs typical of the most ambitious HPC “grand challenge” research.

Storage
storage Wiwynn Open Vault SAS6G JBOD up to 30 hot-pluggable 3.5” HDDs (OCD Approved)
Bryce Canyon sas12g storage server comprising 2 server cards with up-to 36 hot pluggable HDDs each storage

OCP Vendors:

Wiwynn
ITOCHU Techno-Solutions Corporation
Circle B
Penguin Computing




The storage landscape within the cloud is currently one which many HPC users will find challenging. Migrating data directly in and out of instance-attached block storage volumes via SSH (SSH, also known as Secure Shell or Secure Socket Shell, is a network protocol that gives users, particularly system administrators, a secure way to access a computer over an unsecured network. ... Secure Shell provides strong authentication and encrypted data communications) might be the way to go for short, simple tasks – but any practical HPC workflow with data persisting across jobs is going to need to make use of object storage.

Management

At OCP18, significant attention was given to Updating OpenBMC in the following areas:

  • Improved resilience and speed
  • Improved unit test coverage
  • More prerequisite checks
  • Less back-and-forth over SSH
  • Default to HTTP (with checksum validation) instead of SCP
  • Queue of out-of-date systems computed from version data from monitoring OpenBMC Distributions
OpenBMC Distributions

Goal

  • Shared Development
  • Industry standard Interfaces
  • Unified OpenBMC Distribution

Community Project

OCP Power Challenges

Future Power Challenges for IT Equipment

  • Higher power • 150W+
  • Faster transients • 500v/μS
  • Memory
    • - Faster & more DIMMs
    • - Accelerators
  • Similar trends as CPUs, for GPUs, FPGAs, ASICs
Software/Tools Challenge

An open framework of tools is needed to allow the cloud-based HPC ecosystem
to develop more readily, and with less risk to third-party vendors.




Contact us

This field is required
This field is required
This field is required
This field is required
This field is required
This field is required


France, HPE are building
Europe's most powerful
AI supercomputer

France is building a supercomputer that can perform both traditional high-performance computing workloads and AI-specific workloads. The converged design is becoming increasingly common.

AI for HPC and HPC for AI Workflows: The Differences,Gaps and Opportunities with Data Management

The Convergence of Big Data and Extreme-Scale HPC