LUMI Extreme Scale Access - Services for Research
LUMI offers several computing partitions and capacities. All the capacities are accounted in three types of units, GPU (graphics processing unit) hours, CPU hours (central processing units) and storage hours. All projects on LUMI need to be applied and resourced as a combination of these three units. A potential LUMI user has two routes to apply for the resources.
LUMI resources are divided into national resources administered by the national mechanisms and to the European-wide EuroHPC quota.
One half of the LUMI resources belong to the LUMI consortium countries. This half is divided based on the member countries´ contributions to the LUMI funding. The resources in this pool are allocated through national mechanisms and according to local policies.
Researchers affiliated with an academic research organisation or companies established or located in a LUMI consortium country can apply for these national resources. For more information, select the country.
The other half of the LUMI resources are allocated by the EuroHPC Joint Undertaking. Researchers in European countries, i.e. the EU member states and the associated countries to H2020, including LUMI consortium countries, and companies established or located in a European country, can apply for the resources in this pool.
Open calls for Finnish researchers are listed on the right hand side of this page, and the LUMI Extreme Scale Access described in detail below. More information about LUMI access is also available on the LUMI website.
LUMI Extreme Scale Access
At the moment, LUMI consists of a CPU partition (LUMI-C) and GPU partition (LUMI-G) that you can apply for Extreme Scale Access project.
The call is aimed at high-impact, high-gain innovative research, open to all fields of science justifying the need for extremely large allocations in terms of computing time (especially GPU) and data storage. The call for the Finnish Extreme Scale Access projects is organized by CSC, and it is open for researchers in Finnish higher education and research institutions. The awarded resources are part of Finland’s share of LUMI. The calls will be organized once a year.
Other access modes (e.g. Regular Access) are reserved to lower resource needs. These access modes are described on the LUMI website's Users in Finland section.
Applicants for the Extreme Scale Calls need to:
- Demonstrate that the research requires the use of extremely large allocations to reach the objectives.
- Demonstrate that the methods, softwares and tools are technically adapted to LUMI GPU. This can be shown by technical data collected from GPU systems currently available (Mahti GPU, Puhti GPU, LUMI Early Access Platform or other large accelerated systems, such as Piz Daint), see below.
- Provide a project plan, with adequate time schedule of the expected resource consumption during the lifetime of the project.
- The Extreme Scale Call projects are awarded resources in GPU hours, CPU core-hours, and storage hours. These are limited between 0.5–2 million GPU hours and max. 16 million CPU core-hours. The awarded resources have to be consumed within a year.
- The proposals will go through a technical review and a scientific evaluation. The scientific evaluation will be done by an international panel.
- The link to fill in the project application will be separate for each call.
- LUMI's resources are not awarded in billing units as for other CSC services.
Adaptation to LUMI-G in Extreme Scale Access
LUMI-G will be available in September 2022 after the second phase of LUMI installation.
LUMI Early Access Platform is now available with MI100 GPUs for code development and testing. When planning to apply for LUMI-G resources please consider:
- The codes in the proposed Extreme Scale Access projects need to be compatible with the LUMI-G architecture and its next generation AMD CPUs and, especially, GPUs.
- The proposal must justify the technical feasibility of the intended applications: explain why and how the code will be able to be run AMD GPUs on LUMI.
- Codes already GPU enabled may be ported to LUMI-G. For example, codes using CUDA can typically be converted to use HIP with modest effort and good performance.
- GPU offloading with OpenMP or OpenACC is also supported to certain extent.
For more information, please visit:
We encourage the applicants to contact email@example.com if they need any further information regarding the compatibility of their codes or support in LUMI porting.