Skip to content

Latest commit

 

History

History
113 lines (102 loc) · 7.18 KB

computers.md

File metadata and controls

113 lines (102 loc) · 7.18 KB
Raw

Computers

  • GT PACE Phoenix

    • User guide here
    • Login via ssh <GTusername>@login-phoenix-rh9.pace.gatech.edu to get the RHEL9 nodes
    • Purpose: All-purpose campus resource of CPU and GPU jobs with a variety of hardware.
    • "Rules": Use the embers queue type to use idle nodes at zero cost.
    • Get access by ltting Spencer know, and he will fill out this form to get you access.

  • GT ICE

    • Resources/User guide (click Available Resources, e.g.)
      • This looks like ~40 V100s, 8 A100s, 4 A40s, 20 RTX6000s, and 4 MI210s.
    • May need to contact Spencer for access.
    • Most GPU nodes sit idle
      • On those nodes: MaxNodes=UNLIMITED MaxTime=18:00:00

  • GT Rogues Gallery

    • User guide here
    • Purpose: Use of brand-new, forward-looking, or weird hardware. At the time of writing, including an NV H100 server, AMD MI210 GPU server, Bluefield-2/3 SmartNICs, RISC-V and ARM CPUs, etc.
    • "Rules": Few rules, just follow guidelines in documentation. No limitations on hardware access/node hours.
    • Get access via this link

  • GT Wingtip-gpu3

    • User guide here
    • Purpose: Small (but possibly very long) GPU jobs, hosts 5x NV A100-80GB at the moment
    • "Rules": There is no scheduler on this machine, so be mindful of others' use of it.
    • Get access by emailing Will Powell, cc me.

  • ACCESS-CI computers

    • These are a set of university supercomputers, listed here. Each has its own user guide. We have access to NCSA Delta (A100 GPUs), PSC Bridges2 (V100 GPUs), Purdue Anvil, and Texas A&M ACES (H100 GPUs) at the time of writing, but can change to others as needed.
    • Purpose: All-purpose resources for CPU and GPU simulation.
    • "Rules": Be mindful of the node hours that are available. Queue times might be long.
    • Our account number: PHY210084
    • Get access by
      • Creating an account here
      • Then, message Spencer on Slack with your username
    • On NCSA Delta
      • The account name is bbsc-delta-gpu for GPU resources
      • The account name is bbsc-delta-cpu for CPU resources

  • Oak Ridge National Lab OLCF: Frontier/Wombat/Andes/etc.

    • Purpose
      • Frontier: Very large-scale GPU simulation on AMD MI250X GPUs.
      • Wombat: Testbed for next-gen HPC platforms, including ARM nodes and soon next-generation NVIDIA nodes (GraceHopper).
      • Andes: For postprocessing
    • Our account number: CFD154
    • "Rules": Ask Spencer before running any jobs that use a very large number of node hours
    • Get access by
      • Create an account by following these instructions
      • The account/allocation number is CFD154.

  • Sandia National Lab (SNL)

    • Purpose: Resources for DOE-sponsored/funded research projects are only available to those students working on these projects. You will only have access to non-restricted resources.
    • "Rules": Usually, there are not many rules aside from the very many that they will impute onto you as you acquire access to these machines.
    • Login process (Sandia National Lab-specific)
      • Onto the DaaS
        • VMware Horizon (download online)
        • URL: daas.sandia.gov
        • Passcode: [PIN] + [your yubikey1timepassword]
        • Password: [kerberos pw]
        • 3 options
          • badge update
          • conference room
          • daas <- open this one
        • Can complete training and do other things here, like look at your WebCARS to get WC_ID (which you need to submit jobs)
      • Onto a computer remotely
        • https://hpc.sandia.gov/access/ssh/
        • Can do the below with DaaS (using my example username, [usrname])
          • ssh [usrname]@srngate.sandia.gov
          • Passcode: [PIN] + [yubikey one time pw]
          • Choose a computer: e.g., Skybridge, Attaway, Weaver, etc.
          • Press 1 - ssh session
          • Default user name ([usrname])
          • Password is (usually) the Kerberos one
          • If it asks for token OTP (e.g., on Weaver) then this is [PIN] + [yubikey1timepassword]

  • LLNL Livermore Computing: Lassen, Tioga, etc.

    • Anyone working on a specific LLNL project can use LLNL CZ (non-restricted) resources
    • Talk to Spencer about getting access to CZ (collaboration zone) if you are working on a LLNL project
    • "Rules": Usually not many rules aside from the very many that they will impute onto you as you acquire access to these machines.
    • Login process (Lawrence Livermore National Lab-specific)
      • Onto LC-idm
        • URL: ic-idm.llnl.gov
        • Passcode: [PIN] + [rsa one time password]
        • Can use to view user profile and request roles (ask for resources on specific machines)
      • Onto the LC
        • URL: lc.llnl.gov
        • Passcode: [PIN] + [rsa one time password]
        • Requires three logins to fully log in
        • Can use to access collaboration tools such as Confluence and Gitlab, user documentation, and MyLC for alerts, machine status, and job status
      • Onto a computer remotely
        • Can do the below with ssh (using my example username, [usrname], for a specific llnl machine, [llnlmachine])
          • ssh [usrname]@[llnlmachine].llnl.gov
          • Passcode: [PIN] + [rsa one time password]

  • Department of Defense

    • Anyone working on a DOD project can use DOD HPCMP (non-restricted) resources
    • The process of getting permissions to the non-restricted systems is a bit tedious but usually worth it
    • See here for information on the available supercomputers
      • In particular, it's useful to keep an eye on upcoming systems
      • Current unclassified systems are here
    • Talk to Spencer about getting access to a DOD machine if you are working on a DOD project
    • Subproject: ONRDC51242690, Group: 5124D690, Project: 5124
      • Site: NAVY
        • nautilus, nautilus.navydsrc.hpc.mi
        • st-vsm1, st-vsm1.navydsrc.hpc.mil
        • stportal
      • Site: ERDC
        • gold, gold.erdc.hpc.mil
        • viportal
    • Docs available here