Skip to content

Data Analytics Nodes


This page is about a partition that is not yet available

The LUMI-D partition consists of a 12 nodes with large memory capacity and Nvidia GPUs. LUMI-D is intended for interactive data analytics and visualization. It is also a good place run pre- and post-processing jobs that require a lot of memory.

Nodes CPU Memory GPUs Disk Network
4 AMD EPYC 7742
2.25 GHz base
3.4 GHz boost
128 cores (2x64)
8 TB none 25 TB SSD 2x100 Gb/s
8 AMD EPYC 7742
2.25 GHz base
3.4 GHz boost
128 cores (2x64)
2 TB 8x NVIDIA RTX8000
48 GB of memory
14 TB SSD 2x100 Gb/s

Note: The CPUs in LUMI-D are one generation older (Zen 2 / "Rome") than in LUMI-C (Zen 3 / "Milan"). There should be no big problem with software compatibility, though, as only a few new processor instructions related to encryption and virtualization was added to the Zen 3 core. We expect that almost all programs compiled for LUMI-C (e.g. with -march=znver3) will run on LUMI-D with good performance.


The interactive GPU nodes have 8 NVIDIA Quadro RTX8000 GPUs each with 4,608 GPU cores, 576 Tensor cores, 72 RT cores and 48 GiB GDDR6 Memory. The current Nvidia driver version is [FIXME]. The CUDA Toolkit for GPU development is also installed so that it is possible to run and test CUDA code, but the main purpose is visualization of these GPUs are visualizing, not GPU computing.


  • The servers in LUMI-D have local SSDs with a total capacity of 25 TB for the CPU nodes and 14 TB for the GPU nodes [FIXME]. This storage is accessible on the path /scratch/local [FIXME]. It is a good idea to use the local disks when working with many small files (e.g. compiling), as that will be faster than using the parallel file system.
  • The LUMI-D nodes are connected with 2x 100GbE Adapters to the Cray Slingshot Network, which means that they can access to all the parallel file systems (LUMI-P and LUMI-F) with good performance.