Cluster One - Building the World's largest SuperComputer

Discussion in 'Distributed Computing' started by Cluster One, Sep 14, 2017.

  1. Cluster One macrumors newbie

    Cluster One

    Joined:
    Sep 14, 2017
    #1
    Hi everyone,

    I’m the founder of Good Ai Lab and we have just launched our new product Cluster One. It’s a very big project that heavily depends on a community of people being involved and I would love to get your feedback on it. At Cluster One we are trying to help advance science by building the world's largest AI supercomputer.

    We understand how much computing power is wasted every day (around 10 billions hours!) and we feel that with our expertise, and if we all join together, we could really make a difference in advancing scientific research.

    The product has just launched this week and so I would love your feedback on the site to understand if everything makes sense and would it be something you would want to try, and if not what would stop you?
     
  2. ChristianVirtual, Sep 14, 2017
    Last edited: Sep 15, 2017

    ChristianVirtual macrumors 601

    ChristianVirtual

    Joined:
    May 10, 2010
    Location:
    日本
    #2
    What differentiate you from existing projects like F@H or BOINC ?

    Who has access to the computing power the community provide ?

    Can a community member decide which projects he want to support ?

    Will the science result in public domain ?

    Do the scientists have to pay to use the cluster ?

    Windows ? Linux ? Mac ?

    Any compensation for the community ? Coins ? Credits ? Points ? Cookies ?


    I think, as you come to a computing sub forum, there is not much wasted computing power. My system run 24/7 and are loaded ...
     
  3. ChristianVirtual macrumors 601

    ChristianVirtual

    Joined:
    May 10, 2010
    Location:
    日本
    #3
    I signed up ; but there was not question for hardware I could offer (and influence my position on the invite). Only referral link to push the position ...
     
  4. malomarrec, Sep 16, 2017
    Last edited: Sep 16, 2017

    malomarrec macrumors newbie

    malomarrec

    Joined:
    Sep 16, 2017
    #4
    --- Post Merged, Sep 16, 2017 ---
    Hi ChristianJapan,
    Those are great questions!

    What differentiate you from existing projects like F@H or BOINC?

    Our movement aims at bringint the power of AI to the research community. We are bringing the industry via our TensorPort platform together with the research community and individual computing providers. We are building the largest AI supercomputing, so we will be able to support any research initiative that needs computing power for AI.

    We are more general purpose than F@H in terms of application, as you can decide to donate your compute to a range of projects, and we will deal with assigning it to whoever you chose requires compute at the moment.

    Who has access to the computing power the community provide?
    The user choses how their compute is attributed. We will gradually bring projects to the platform based on your feedback, and ask you what you want to contributed to :)

    Can a community member decide with projects he want to support?
    Absolutely. We are in alpha right now so we will assign compute to a few projects we will partner with, accoring to our first users preferences.
    IOn the final release, you will be able to select specifically what project you are donating too.

    Will the science result in public domain?
    It will be the choice of whoever uses the compute. However we will make that very clear to the you so that you can be sure what your compute is used for.

    Do the scientists have to pay to use the cluster?
    No. We will support a paying version in the future: in that case, people that join the cluster will be retributed for that compute.

    Windows? Linux? Mac?
    All of them. We will be starting off with what is most requested. What do YOU have?

    Any compensation for the community ? Coins ? Credits ? Points ? Cookies ?
    As stated above, there will be too models:
    - donate your compute to advance science. In that case, maybe cookies ;)
    - sell your compute. In that case, you will be payed in a format that we will define together!

    Cheers,
    https://clusterone.com?kid=H6H0Z
     
  5. ChristianVirtual macrumors 601

    ChristianVirtual

    Joined:
    May 10, 2010
    Location:
    日本
    #5
    I’m fine with cookies, get nothing else since years and even have to pay for those.

    Else I’m primary on Linux (CentOS 7) with nV card like 1080ti, 1080 and 980Ti ...and few CPU cores.
     
  6. Cluster One thread starter macrumors newbie

    Cluster One

    Joined:
    Sep 14, 2017
    #6
    Hi everyone,

    Just a quick update - we've launched Cluster One! The announcement article has been published on Medium explaining why it has been created, feel free to read it here: https://medium.com/@mhejrati/announcing-cluster-one-the-largest-ai-supercomputer-3abff76a0bb2

    To clarify, here's the problem Cluster One is aiming to resolve and how others can contribute to it.

    A year ago, top cancer researchers reported to President Obama about the state of cancer research, and most of their recommendations mentioned large scale computing as a way to move the industry forward.


    AI is not affordable for every company and is exclusive because of:

    • algorithms and science is still limited
    • knowledge and expertise to use it is rare
    • infrastructure is hard to build and costly to operate

    We see ourselves as solving a part of that problem. Organizations such as the Allen institute are advancing and spreading algorithms. Companies such as Andrew Ng’s deeplearning.ai are trying to spread deep learning skills.

    What’s missing in that picture is spreading affordable infrastructure and tools. That’s why we are launching Cluster One.

    It is important to understand what size we need to reach before being able to do something meaningful.

    For example, take Diabetic Retinopathy, a disease that affects people with diabetes, and can ultimately cause blindness. It affects nearly 100 Million people in the world.

    For the sake of understanding what it would take to offer a screening solution through AI, let us assume the following.
    • explore 100 ideas
    • run 50 experiments per idea
    • run each of them for a week of computation, on 50 machines
    That’s a total of 42MM compute hours.

    That would cost around $10MM on the public cloud (eg: on AWS’s c4.2xlarge), or several dozens of millions of upfront investment for a private infrastructure.

    Or it could be provided by 15,000 contributors who provide 8 hours of compute a day for a year, on recent computers.

    That’s why we believe in the power of distributed computing and we’re on a mission to scale AI to enable researchers push science further. Feel free to reach out to me if you have any further questions or would like to know more about the movement. Or if you want to help, sign up to join Cluster One community.
     
  7. ChristianVirtual macrumors 601

    ChristianVirtual

    Joined:
    May 10, 2010
    Location:
    日本
    #7
    Why are you not just using BOINC ? There is the infrastructure and community established ? 15.000 users will most likely not a problem. Switching projects is quite common practice.

    Source code and skilled people are around, too.
     
  8. ChristianVirtual macrumors 601

    ChristianVirtual

    Joined:
    May 10, 2010
    Location:
    日本

Share This Page

7 September 14, 2017