Exascale Computing Project FAQ - 2 pp.
from
https:// www.exascaleproject.org/wp-content/uploads/2018/01/ECP_Factsheet_v1.0_20180103.pdf
Some supercomputer terms
A petaflop (PF) is the ability of a computer to do one quadrillion floating point operations per second (FLOPS). Additionally, a petaflop can be measured as one thousand teraflops.
A petaflop computer requires a massive number of computers working in parallel on the same problem. Applications might include real-time nuclear magnetic resonance imaging during surgery or even astrophysical simulation.
Today's fastest parallel computing operations are capable of petaflop speeds. The world's fastest supercomputer today, Titan, is capable of 20 petaflops.
Exascale computing … refers to computing systems capable of at least one exaFLOPS, or a billion billion calculations per second. Such capacity represents a thousandfold increase over the first petascale computer that came into operation in 2008. At a supercomputing conference in 2009, Computerworld projected exascale implementation by 2018.
A qubit is 1 and 0 at the same time UNTIL the data in the qubit is measured, i.e. transferred from multidimensional hyperspace into our timeframe.
It is 1 and 0 simultaneously because, from our perspective, it represents a probability. It may be useful to think of the probability as representing a range of timestreams that could potentially manifest. When the qubit is measured as a 1 or a 0, ONE of those potential timestreams actually does manifest. Or rather, we connect the timestream where the measurement takes place (ours) with that potential timestream, effectively guiding our timestream to anneal with that potential timestream and become it.
Read this short article to understand the meaning of measurement in this context.
https:// phys.org/news/2012-10-gently-cubit-superposition.html
Deep Dig on Exascale Computing Project (ECP)
https:// www.exascaleproject.org/ has current articles (today’s date 2/16/18).
Anon’s notes on http:// procurement.ornl.gov/rfp/Exascale-rfi/RFI-VendorMtg-v2.pdf
Exascale Systems RFI Vendor Meeting
Al Geist, Head of RFI Dev Team – 2/28/17 – National Nuclear Security Admin., US Dept of Energy | Office of Science.
Acronyms: RFI = Request for Information;
RFP = Request for Proposals
HPC = High-Performance Computing
PF = petaflop, a measure of
• Market survey to determine scope of forthcoming RFP.
• Asking about systems that vendors could build for 2021, 2022, 2024.
• RFI responses to be shared only with 6 DoE labs and HQ:
–Argonne Nat Lab (ANL)
–Lawrence Berkeley Nat Lab (LBNL)
–Lawrence Livermore National Laboratory (LLNL)
–Los Alamos NL (LANL)
–Oak Ridge NL (ORL)
–Sandia Nat Laboratory (SNL)
–DoE HQ
• Goal of exascale project: US leadership in high-performance computing (HPC)
• Seek to increase capability 5x over current HPC roadmap (which had projected to achieve exascale by 2027)
• For 2022: seek exascale supercomputer that can solve science probs 50x faster (or more complex) than 20 PF (petaflop) systems can today.
• Needs software stack
• Power envelope 20-30 MW
• Requirements might be relaxed for 2021 exascale (cooling, budget, space, S/W) but will nonetheless be a DoE production system upgradable to vendor’s product roadmap for exascale. Site for the 2021 exascale is TBD. (Implying that only ONE 2021 exascale system is planned.)
• Responses to RFI (<50 pp.) were due 3/31/17.
PathForward-II RFP
• Refers to advanced hardware (“component, node, and system”) that may enable dramatic performance improvement, e.g. 100x in 2021, on “some classes of applications”.
• Want it to intersect with 2021 exascale proj. Seeking innovations in power consumptions, performance, programmabiltiy, reliability, data science, machine learning, and/or portability.
• Reduce TCO.
Software Gap Analysis
• Compare current SW portfolio vs. emerging data analytics and machine learning needs vs. vendors’ projected exascale SW stacks.
CORAL RFP
• CORAL = Collaboration of ORNL, ANL & LLNL.
• 2018 CORAL RFP will seek 3 exascale systems by 2022-2023.
• Goal: solve emerging data science + machine learning probs, in addition to trad modelling + simulation apps.
• Seeking diverse architectures (my take: DoE labs don’t want to be dependent on a sole approach or sole vendor)
• DoE wants diversity for price competition, reduce risks, promote HPC ecosystem for national competitiveness.
• LANL, SNL & LBNL plan systems acquisitions in 2024.