MPG Cloud

In the context of MPG Cloud various cloud, storage and HPC resources are provisioned in addition to the respective core offers for Max Planck Society by GWDG.

CPU-/GPU-Knoten

For hosting cloud-VMs in the OpenStack environment of GWDG the following nodes can be used:

• 24 CPU nodes (64 AMD EPYC Rome/Milan cores, 2048 GB RAM)
• 8 CPU nodes (192 AMD EPYC Genoa cores, 1536 GB RAM)
• 8 GPU nodes (120 Intel Emerald Rapids cores, 1024 GB RAM, 8 TB SSD, 4x NVIDIA H100 GPU)

Storage

For MPG Cloud the Ceph infrastructure of GWDG with 9.1 PB HDD and 1.4 PB NVMe SSD capacity can be used.

GPU inference nodes

The new GPU inference nodes support large AI models like DeepSeek-R1-0528 and are run with the SAIA platform of GWDG, which is also used to host the Chat AI service:

• 2 GPU nodes (104 Intel SapphireRapids cores, 2048 GB RAM, 2x 3,84 TB NVMe-SSD, 8x NVIDIA H200 GPU (141 GB VRAM))

More information

The planned strategy for MPG cloud includes dedicated resources which can be booked for a fixed timespan. 25% of the operational cost is billed. Further information for this option will be available at this location.

Additional resources, HPC Cloud and storage, are located at MPCDF.