exascale systems request for information vendor meeting › rfp › exascale-rfi ›...

www.ExascaleProject.org

Exascale Systems Request for Information Vendor Meeting

Al Geist Head of RFI Development Team February 28, 2017 Washington, DC

http://www.exascaleproject.org/

2 Exascale Computing Project, www.exascaleproject.org

Agenda

Time Topic 8:30 – 9:00 Overview of the RFI: purpose, scope, and timeline

9:00 – 9:45 Overview of what information each RFP Writing Team is looking for: - ECP PathForward-II RFP - ECP software gap analysis - CORAL RFP (with potential APEX long-range R&D)

9:45 – 10:00 Q&A with vendors 10:00 – 10:15 Break 10:15 – 10:45 Detailed information about each RFI question

10:45 – 11:30 Q&A with vendors 11:30 – noon Break 12:00 – 5:00 pm Private 25 minute vendor Q&A with the RFP representatives

No presentations by vendors or reps – just answering vendor questions


What is a Request for Information (RFI)?

Before DOE releases a large Request for Proposals (RFP), such as for a system acquisition, it does a market survey by sending out a Request for Information (RFI). The information returned by the vendors provides scope to what makes sense to ask for in the RFP.

An RFI is for informational purposes only and may not necessarily result in a solicitation. Solicitation or awards of contracts are subject to available DOE funding and the priorities of the new administration.

Typically a vendor meeting is not held with an RFI, but because of the complexity of this Exascale RFI, we are holding one.


Complexity of this Exascale RFI

• Combines a system acquisition RFI and two ECP RFIs that were planned to be released at the same time asking for similar information.

• Asking at high-level for all the different systems you could build

• Asking for this for three different time frames – 2021, 2022, 2024

• Given these three sets of possible systems what are your priorities for PathForward-II R&D and NRE?

• What is your planned exascale software stack and software roadmap

• Are there any ECP requirements which if relaxed in RFP would provide substantial innovation or acceleration to your long-term roadmaps?

• Estimated costs – This is NOT an RFP – nothing is binding, but we expect you will respond with values in the ball park.


‘

Who sees your Exascale RFI Responses?

Exascale Computing Project (ECP) released the Exascale RFI Feb 7th The responses will be shared only with the 6 DOE laboratories in ECP:

• Argonne National Laboratory • Lawrence Berkeley National Laboratory • Lawrence Livermore National Laboratory • Los Alamos National Laboratory • Oak Ridge National Laboratory • Sandia National Laboratories • And DOE Headquarters

Unlike an RFP, an RFI has no silent period, so vendors are free to talk with labs and ECP about the RFI and the future RFPs as much as they want.


ECP Goals

Time (CY)

Capability

2017 2021 2022 2023 2024 2025 2026 2027

10X

5X

• Foster a rich exascale ecosystem • Help ensure continued U.S. HPC Leadership • Enable delivery of an exascale system in 2021

and capable exascale computers in 2022


2022 Capable Exascale Computers

ECP defines a capable exascale system as a supercomputer that • Can solve science problems 50x faster (or more complex, for example, more physics,

higher fidelity) than the 20 PF systems of today can solve comparable problems.

• Must use a software stack that meets the needs of a broad spectrum of applications and workloads

• Have a power envelope of 20-30 MW

• Must be sufficiently resilient such that user intervention due to hardware or system faults is required on the order of a week.


2021 Exascale Computer

The site of the 2021 system is TBD and will be informed by the power, space, cooling, and budget requirements from this Request for Information. Based on the results of the RFI, the 2021 system might have more relaxed requirements for power consumption, resilience, and programmability.

The 2021 system is expected to be a production system for DOE mission needs, and either to be upgradable to capable exascale, or be on the vendors product roadmap to capable exascale.


‘

Radical, Novel, Advanced solutions are not a Requirement

We want the vendors to propose what they see as being needed to meet performance, reliability, programmability, data science convergence, and power requirements. • If that involves radical new concepts, we are interested in hearing

about these solutions.

• If vendors can meet the requirements without needing new radical solutions that is fine and likely preferred.

We want to encourage vendors to propose new ideas where they provide a path for addressing our requirements but we don’t need novelty or “advancedness” just so we can claim things are “advanced”.


An RFI is valuable to both DOE and the Vendors (even though no money is involved)

A system RFP can be over a hundred pages of very detailed requests similar in depth to a system build statement of work. The RFI responses provide breadth and depth to what goes into the RFP.

The RFI responses provide the scope of potential diversity

The RFI responses provide vendor priorities for R&D and Non-recurring Engineering (NRE) The RFI responses describe why certain R&D activities could have high impact on future systems (perhaps at higher risk). This is valuable to the PathForward-II RFP writing team.


‘

RFI Response Due Date

Responses to this RFI are due no later than 5:00 PM EDT on or before March 31, 2017.

One response per vendor is preferred, but two responses per vendor are allowed in case you want to answer the questions with respect to the 2021 systems in one response and answer the questions with respect to the 2022 systems in a separate response.

Responses shall not exceed 50 pages (each), including all attachments.


Session 2 • Second session starts by showing the timeline and how the timing of the three

potential RFPs are fed by this RFI response.

• Then gives information about each of the RFPs and why we are asking for what we are in this RFI


NRE Early System

Notional Timeline for 2021-2023 DOE System Acquisitions

2016 2017 2018 2019 2020 2021 2022 2023 2025 2024 2026

Early Exascale System

NRE 2022-2024 Systems Capable Systems

Upgrade?

RFI release

System RFP release Proposals evaluated

Draft system RFP

2024 Systems

PathForward-II

Exascale Computing Project PathForward

RFP

Calendar Years

RFI responses due back end of 1Q 2017 provides information for: • 2Q 2017 PathForward-II RFP released. Selections by 3Q 2017 • 2Q 2017 ECP Software Gap Analysis. 3Q 2017 SW PathForward RFP released (if needed) • 1Q 2018 Draft CORAL RFP. 2Q 2018 Final CORAL RFP released. Selections 4Q 2018

Presenter

Presentation Notes


PathForward-II RFP (from the ECP Hardware Technology Team)

• PathForward-II has a goal to support high payoff, innovative hardware technologies and systems technologies that may have higher risk. It is focused on component, node, and system architecture designs that will intersect with the 2021 exascale system.

• Also of interest to the PathForward-II RFP team: – Innovations that may enable dramatic acceleration of certain applications,

for example, delivering a 100x increase in 2021 on some classes of applications while still being able to solve the full range of DOE applications

– Developments that promote wider diversity in the exascale ecosystem – Innovations in power consumption, performance, programmability, reliability,

data science, machine learning, or portability – Reducing total cost of ownership


PathForward-II Awards

• Number and size of the PathForward-II awards will depend on – The DOE budget under the new administration and – The information pertaining to PathForward-II returned in

your RFI responses

• Duration of PathForward-II in the notional timeline extends into early 2021. This is due to two reasons: – We want to include the possibility that some of our

PathForward-II interest areas may intersect your systems that come after the 2021 system

– This duration includes potential future PathForward-II activities that can’t be awarded until after ECP CD-2/3 approval in late 2019.


ECP Software Gap Analysis (by the ECP Software Technology Team)

During a recent ECP project review one of the recommendations was to perform a software gap analysis. In 2017 ECP will perform a software gap analysis based on the ECP applications and DOE HPC facilities’ software needs overlaid with:

• The current portfolio of ECP Software Technology projects

• The emerging data analytics and machine learning needs • The vendors’ projected exascale software stacks

• The vendors’ software roadmaps through 2024 to see if discovered gaps may be closed by the evolution of the vendor’s software

Overlay the apps and facilities SW needs

HPC Facilities SW needs

Look for gaps

What vendors and others supply


“Potential” ECP Software PathForward RFP

This RFI seeks to collect the vendor software stacks and software roadmaps for overlaying in the Software Gap Analysis. The software stack information is one of the items requested in the 2021 system description and also requested in the 2022 system description.

The software roadmap information is one of the items requested in the 2024 system description.

The results of the gap analysis will determine if a Software PathForward RFP to vendors is warranted. So we encourage vendors to provide their software information in their responses.


CORAL RFP (CORAL stands for Collaboration of ORNL, ANL, LLNL)

• In 2018 a single CORAL RFP will be released for delivery of three capable exascale systems by the 2022-2023 timeframe. The RFP will also include NRE for the systems.

• These systems will have to be designed to solve emerging data science and machine learning problems in addition to the traditional modeling and simulations applications.

• The DOE Leadership Computing Facility has a requirement that the ANL and ORNL systems must have diverse architectures. Given the ECP goal of fostering a rich exascale ecosystem, LLNL has the option to choose a system that is diverse from both the ANL and ORNL systems.

• Diversity is very important to DOE. A main goal of this RFI is to understand the breadth of potential diversity for exascale systems.


There are Many Types of System Diversity

• Systems can vary from one another in many different dimensions – System (architecture, interconnect, IO subsystem, density, resilience, etc.) – Node (heterogeneous, homogeneous, memory and processor architectures, etc.) – Software (HPC stack, OS, IO, file system, prog environment, admin tools, etc.) – Hardware e.g.

• Ways Systems can be diverse – Few big differences – Many little differences – Different technologies – Different ecosystems, i.e., vendors involved

technology

scale type

DDR NV PIM

size DIMM

Memory

on die stacked

Fat thin accel

#cores homo

Processor

hetero

topologies

perf optical

Network copper


How Diverse is Enough?

How diverse is enough? There is no hard metric, we will evaluate diversity by how much it will benefit the exascale ecosystem

Having system diversity provides many advantages. • It promotes price competition, which increases the value to DOE.

• It promotes a competition of ideas and technologies, which helps provide more capable systems for DOE’s mission needs.

• It reduces risk that may be caused by delays or failure of a particular technology or shifts in vendor business focus, staff or financial health.

• It helps promote a rich and healthy high performance computing ecosystem, which is important for national competitiveness and DOE’s strategic plan.


What about the 2021 System?

• The site of the 2021 system is TBD and will be informed by the power, space, cooling, and budget requirements from this RFI. – A decision is expected around June 2017

• If the site is one of the CORAL labs, then the CORAL RFP will state: “Within the goal of having three capable exascale systems by 2022-2023, if an early exascale system can be delivered in 2021 and upgraded to a capable exascale system by 2023, then provide the upgrade as an option.”

• If the site of the 2021 system is outside the CORAL labs then (in addition to the CORAL RFP for three 2022 systems) a separate RFP for a single 2021 system will be released in 2018 by the chosen lab.


What is Non-Recurring Engineering (NRE)?

The purpose of Non-Recurring Engineering (NRE) is to fund the development of technology (both hardware and software) that

– Is required to deliver the proposed HPC system(s), and

– Would not be developed (or developed in time) by the vendor if the system(s) were not purchased

NRE is proposed by the vendors in an RFP response – Some propose a system and say it can be enhanced if the following NRE is funded – Some propose a system and say NRE is required to build as proposed – Both are equivalent as long as what happens if NRE (is/is not) funded is described

The CORAL RFP evaluation will consider System + NRE with respect to (total cost, reduced risk, improved capability)


How NRE is going to be Different in CORAL RFP

The system acquisition funds come through the DOE facilities as always, but the NRE funds come through the Exascale Computing Project (ECP).

The ECP plan has a much higher NRE funding amount than in past DOE HPC acquisitions (caveat: subject to the future DOE budgets)

Each major acquisition in the ECP timeframe is expected to have an associated NRE investment and possibly more than one to mitigate technology risks and ensure a healthy ecosystem

Your RFI responses will help inform us on the number, timing, and approximate size and of NRE investments.


CORAL RFP and Benefits to 2024 APEX systems

• The ECP rich ecosystem goal also extends to the LANL/SNL and LBNL systems acquisitions in 2024. – ECP NRE supplies the opportunity – The CORAL RFP supplies the vehicle

• The CORAL RFP will include language to the effect: “If long-range R&D (or NRE) done in the ECP timeframe (2018-2022) could enhance production systems in 2024, your response to the CORAL RFP should describe it.”

• This RFI has a question about 2024 systems specifically so we can – Know possibilities of what to write into the CORAL RFP about enhancing

production systems in 2024 – Understand the long-term benefits of ECP to the exascale ecosystem


Session 3

• Vendor Questions on Initial Material • Followed by 15 minute break


Session 4

• Fourth session is deep dive on the RFI questions. • Each question listed in the last section of the RFI attachment is the

title of a slide followed by bullets of some key points of what is being asked for.

• RFI is for the collection of information not a “give us a list of things we may fund” – that is what an RFP is for. Thus, your responses are not graded on how many questions you answer or how well you answer.

• If a question is not applicable to your company or you have no answer at this time then simply say so in your response.


Describe your possible 2021 early exascale hardware/software solutions (“none” is a valid ans.)

• Provide a high-level description of the possible 2021 systems – much of this could be done in a table or matrix with associated text to describe key characteristics and differences between the systems

• Software environment(s) – your different systems may have or require different software stacks. We want to understand the breadth of software environments that may be exist in a rich exascale ecosystem

• If not initially capable exascale, what options (if any) for upgrading to capable exascale

• Specify which of the 2021 solutions you consider the best (can be more than one) and explain why. Important for us to know your preferences and the why is your opportunity to give technological as well as business reasons.


Describe your possible 2022 capable exascale hardware/software solutions

• Provide a high-level description of the possible 2022 systems – much of this could be done in a table or matrix with associated text to describe key characteristics and differences between the systems

• Notional system architecture sketches with characteristics of the nodes, memory hierarchy/hierarchies, interconnect(s), and system(s)

• Software environment(s) – your different systems may have or require different software stacks. We want to understand the breadth of software environments that may be exist in a rich exascale ecosystem

• Specify which of the 2022 solutions you consider the best (can be more than one) and explain why. Important for us to know your preferences and the why is your opportunity to give technological as well as business reasons.


Describe your possible 2024 exascale hardware/software solutions

• Please provide a high-level description of 2024-2025 system(s) • This far out, the 2024 descriptions may only be notional. But we are

interested in how possible ECP long-term R&D (or NRE) investments will impact the out-year vendor roadmaps.

• CORAL RFP will include text to this effect. – Could be CORAL system NRE that will have high impact on systems after CORAL – Could be long-term R&D key to shifting your future roadmap (but funding short-term)

• A separate RFI will come out in a few years for the 2024 acquisitions – so vendors will have the opportunity to give details then.

• Software roadmap – We want to understand the breadth of software environments that may be exist in the out-years and where gaps may arise in the exascale software ecosystem


How is your preferred system(s) for 2021 and/or 2022 well suited for the DOE Mission needs?

• The answer to this question falls naturally in the section where you specify which of your solution(s) you consider best.

• But focuses not on technology and business reasons but rather on how your solution meets science needs for performance, programmability, portability, reliability, data science, and machine learning.

• To help us understand your system’s ability to do data science, machine learning, and simulation science, estimate the system’s peak FLOPS (double precision, single precision, and half precision), memory bandwidth, injection bandwidth, and bisection bandwidth.

• Also describe any system features that may provide substantial acceleration for certain applications.


Address the Potential Impact of NRE Funding on your proposed system(s)

• Discuss NRE activities that you would propose and how they impact the systems that you describe in your RFI response. – This can include any systems you describe not just your preferred ones and a

single NRE activity may have benefit for several of your proposed systems – Indicate your priorities for your proposed NRE efforts

• What are the primary areas in which you feel NRE funding is needed to deliver a capable exascale system? – Specifically asking about meeting the ECP goal

• How would NRE funding alter the system that might otherwise be bid – For example, compressed schedule, improved technology, reduced cost?

• If applicable, how would the NRE work complement your related R&D funding (internal and external such as PathForward)?


How do the proposed systems represent an advanced technology solution? • This question is specifically collecting information for writing the

PathForward-II RFP. • Responses to this question can include innovations in any or all of your

proposed solutions, but particularly the 2021 solutions and solutions that would promote wide ecosystem diversity. – Also of interest, innovations that may enable dramatic performance increase on certain

applications but may not be broadly applicable across all DOE science areas.

• PathForward II has a goal to support high payoff, innovative hardware technologies and systems technologies that may have higher risk. It is focused on component, node, and system architecture designs that will intersect with the 2021 exascale system.

• Clearly label and prioritize the potential PathForward-II efforts in your response.


Representative areas in which DOE has interest for NRE and PathForward-II:

• This is not a question in the RFI. It is a list provided for your information as you respond to the RFI questions about NRE and PathForward-II.

• You are not expected to respond to all items in the list.

• And the last item “Any other innovations” is intended to give you the opportunity to describe high-impact innovations for PathForward-II (or NRE) that may be specific to your proposed systems.


Tradeoffs and Risks

• Are there requirements which, if relaxed, would enable you to offer a more compelling exascale solution in the 2021-2022 timeframe?

• If so, discuss the impact of relaxing any of the ECP requirements on your proposed 2021 and 2022 systems. particularly those relaxations that: – would provide substantial innovation or acceleration to your long-term roadmaps – would substantially reduce any aspect of total cost of ownership.

• Identify what you perceive as the top risks and uncertainties for meeting all the ECP requirements – especially the 2021 schedule. – Suggest potential tradeoffs or mitigations in your RFI response.


Cost Estimates • This is an RFI. The cost estimates are non-binding. They are just your

best estimates at this time. • We are gathering information that will help us scope future RFPs. It is in

your response to these future RFPs where your cost estimates will be more formal.

• For this RFI we are asking you to provide cost estimates for: – All your described systems, if possible, but particularly your preferred systems – If any of your 2021 systems are upgradable and not already capable exascale,

provide a cost estimate for upgrading to capable exascale. – Your highest priority NRE tasks, and primary capable exascale NRE tasks, etc. – Include any cost considerations that you feel would be important in our preparation of

a 2017 PathForward-II RFP. – Include any other cost considerations that you feel would be important in preparation

of a 2021-2023 system acquisition RFP.


Session 5

• Questions from vendors on anything presented during the morning or something they saw in the RFI that was not covered. Vendor Questions on preparing Responses for the Exascale Request for Information

• Afternoon: Individual Vendor Sessions: Room Studio F – Starting at noon. – 25 minutes with 5 minute breaks between

exascale systems request for information vendor meeting › rfp › exascale-rfi ›...

Documents