lumi-supercomputer.eu

What It Is

LUMI (Large Unified Modern Infrastructure) is a European pre-exascale supercomputer located at the CSC data center in Kajaani, Finland, jointly supported by EuroHPC JU and the countries in the LUMI consortium. It is designed for large-scale computing needs in research and industry. The main text states that its sustained HPL performance is 380 petaflops, with resources allocated in the form of GPU-hours, CPU-hours, and storage hours.

Core Capabilities and Developer Perspective

In terms of use cases, LUMI primarily serves climate modeling, genomics and disease research, AI/deep learning, autonomous driving, materials science, big data in the social sciences, quantum algorithms, emergency computing, and similar workloads. It is not an IDE, CI system, or cloud API tool in the traditional sense, but rather a high-performance computing platform.

For languages and frameworks, the main text provides considerable detail on migration to AMD GPUs: NVIDIA CUDA applications do not run out of the box on LUMI-G and typically need to be ported to HIP. It also supports OpenMP GPU offloading (C/C++, Fortran), OpenACC (Fortran only), and C++ ecosystem tools such as Kokkos, RAJA, SYCL, and ALPAKA. CUDA Fortran can be adapted via the hipfort path. Clear API/SDK details are not disclosed; usage appears to be primarily through login nodes, a Web interface, and batch job policies.

Pricing and Resource Access

The website does not provide commercial pricing or payment methods. Resources are obtained through EuroHPC JU peer review, local allocations by consortium countries, and channels that make up to 20% of EuroHPC resources available to industry and SMEs. Non-European projects may also be able to apply through collaboration with European institutions, but the project PI must be based in the EU or an associated country.

Pros and Cons

The advantages are its powerful computing capacity, extensive research and enterprise use cases, and the availability of FAQs, service status, SLA information, training entry points, user support, and GPU porting support. It is also connected with the VTT Helmi and VTT Q50 quantum computers, making the ecosystem relatively complete. The drawbacks are the high eligibility and review thresholds, making it unsuitable for ordinary developers who want to buy GPU capacity on demand with a credit card; migrating CUDA workloads to AMD GPUs involves engineering effort; and sensitive data processing capabilities are still being prepared.

Who It’s For and Access from China

LUMI is suitable for European research teams, enterprises in consortium countries, institutions needing ultra-large-scale AI training or scientific simulation, and developers with HPC experience. Access from China is not discussed in the main text, so it is considered unknown. For Chinese teams, the practical barriers are more likely to involve eligibility, collaborative applications, cross-border compliance, and network stability rather than simple website access. Alternatives include other EuroHPC systems, national HPC centers, AWS/GCP/Azure HPC, or China’s national supercomputing centers and local cloud GPU/HPC services.