Calls for LUMI compute time 


The resources from the Dutch LUMI share are made available through the Continuous Call for Computing Time on the National Computer Facilities by the Dutch Research Council (NWO). Affiliates of one of the Dutch Universities or Grand Technology Institutes can apply for LUMI resources through this call.
The following two grants are available:

  • LUMI pilot projects that requires limited amounts of computing time. 
  • LUMI regular projects that requires large amounts of computing time.

Both types of projects have a maximum duration of 1 year. 

More info

More information about:

  • The scope of each grant type 
  • Eligibility criteria
  • How to request compute time

Can be found on the SURF website Access to compute services and the NWO website Computing Time on National Computer Facilities.




Which projects can apply for LUMI compute time?

LUMI serves as and extension of our national compute facilities and serves the growing demand for compute resources. It contains a large GPU partition with 2978 GPU nodes and is especially suitable for large-scale GPU projects, such as GPU accelerated simulations and distributed training of large deep neural networks.

Projects that can benefit from LUMI’s large scale and GPU capacity are encouraged to apply for compute time on LUMI from the Dutch share.
Interest in the specific hardware or software aspects of LUMI is also a valid motivation to apply for compute time on LUMI. LUMI offers the following hardware and software technologies that differ from our national system Snellius.


 


LUMI HardwareLUMI Software
  • AMD MI250x GPUs
  • AMD EPYC 7763 (Milan) CPUs (LUMI-C)
  • HPE Slingshot-11 interconnect in Dragonfly architecture
  • HPE Cray compilers, libraries and tools
  • AMD debugging and profiling tools
  • Central software stack provided and maintained by the LUMI User Support Team




Comparison LUMI to Snellius


For large-scale CPU-only projects and projects that strictly require the availability of NVIDIA GPUs, the national system Snellius is better suited. Snellius has a relatively large CPU partition with AMD CPUs of a later generation and contains NVIDIA GPUs.

The following table provides a (non-exhaustive) overview of reasons to apply for compute time on LUMI from the Dutch share, as well as a guide to decide whether LUMI or Snellius is the most suitable system for your project:



LUMISnellius
  • Large-scale GPU projects
    • That benefit from LUMI's scale
  • Projects targeting AMD GPUs
    • Porting code to AMD GPUs
    • Assessing performance portability between GPU architectures
  • Projects investigating Slingshot-11 interconnect / Dragonfly topology
    • Benchmarking Slingshot-11 interconnect
    • Investigating effect of Dragonfly topology on application performance
  • Projects interested in HPE or AMD software
    • Using or assessing HPE Cray compilers, libraries and tools
    • Using or assessing AMD debugging and profiling tools
  • Small-scale CPU-only projects
    • Unless they target LUMI-specific hardware or software


  • Large-scale CPU-only projects


  • Projects that strictly require NVIDIA GPUs






Application and review process

LUMI Pilot projects can be requested as part of a Small Compute Application directly via SURF. Applications can be made through our Service Desk, via the link "Small Compute applications (NWO)".
These requests will only be assessed on technical feasibility by SURF and are usually handled within 2 weeks.
An overview of the resources that can be requested can be found here.

It is important to understand that the requested resources need to be justified. You must detail how many CPU core-hours, GPU-hours and TB-hours you require and how you plan to use them.

For more information on LUMI's technical details, please see the LUMI documentation.





LUMI Regular projects can be requested as part of a Large Compute Application. Applications can be made through NWO's ISAAC portal. These requests will undergo a scientific review and be assessed on technical feasibility.

These proposals will be peer-reviewed by and discussed within the NWO Committee for the Scientific Use of Supercomputers (WGS), which has been mandated by the NWO board to grant HPC resources. The number of referees depends on the actual amount of computing time you apply for. NWO may ask you for clarification of your request, based on comments of the referee(s). During the next scheduled WGS meeting, a granting decision will be taken. This procedure may take 4-8 weeks, since referees and a meeting of the WGS are involved. The NWO office (to be confirmed by the WGS) is in general able to provide you with a preliminary amount of computing time, which is around 10-20% of what you applied for.

If your proposal is accepted, you will receive a granting letter. SURF will receive a copy of the proposal and granting decision and subsequently will contact you with directions on how to obtain your account(s) on LUMI.

LUMI pilot before LUMI regular project

In order apply for a LUMI regular project, it is required that the applicant has had prior access to LUMI. The applicant is required to demonstrate that their application(s) can make effective use of LUMI's hardware and scales well to the target scale.

The required benchmarking and scalability results can be obtained as part of a LUMI Pilot or one of the EuroHPC JU access calls (see below).




Account creation

Once you have applied for LUMI resources and your application has been granted, a project will be created in the Puhuri Portal.

The Principal Investigator (PI) of your project will receive an email with an invitation to create an account and accept the project invitation. The PI can then invite other users to the project.

Not all members are permitted

Note that a project PI/co-PI must not invite members to a project from an organization located in countries which are under US Export Control or EU sanctions.

To log in to the Puhuri Portal you have to:

  • Register to MyAccessID.
  •  Search and select your institute’s name in MyAccessID and log in using your institute's credentials.
  • You will be logged in through SURFconext.
  • After accepting the LUMI General Terms of Use and the MyAccessID Acceptable Use Policy, you can register your public key in MyAccessID.
  • After your key has been synchronized to LUMI, you can log in to the system.


For detailed instructions on setting up your account and keypairs, see the first steps page in the LUMI documentation.





Project extensions

LUMI regular and LUMI pilot projects have a maximum duration of one year. In well-motivated cases, projects can be extended up to 3 months. To request an extension, do the following:





Access through EuroHPC JU

Researchers and industrial users can also submit an application for compute resources from EuroHPC's share of LUMI resources through EuroHPC JU. These calls can be interesting for researchers that request:

  • Extreme amounts of compute time that is not available from the Dutch share.
  • A short project (max 3 months) with few resources to demonstrate the scalability of their code on LUMI. Such scalability experiments can be used to support a request for a LUMI regular project from the Dutch share.

Note that projects that apply for access through the EuroHPC calls:

  • Are reviewed by the EuroHPC JU and not by SURF.
  • Will only be reviewed after pre-defined cut-off dates.
  • Are not eligible for specialized support by SURF (see Support and training).

More information on the EuroHPC JU calls can be found here.



  • No labels