General information

• Investigation area

  3D Materials Analytics
• Techniques

  GPU Computing / 3D-4D Tomography Visualization & Analysis
  
• Manufacturer

  Supermicro / NVIDIA
• Fabrication year

  2025
• Measured quantity

  3D volumetric reconstructions, segmented phase maps, porosity and pore connectivity distributions
• Main application

  Remotely accessible, centralized GPU server for 3D and 4D (time-resolved) X-ray tomography data visualization and analysis. Supports cross-faculty research in materials science, geoscience, engineering, and life sciences. Applications include AI/ML-driven image segmentation, deep learning model training, GPU-accelerated CT reconstruction, quantitative microstructure analysis, and large-scale volumetric data processing using both commercial (Dragonfly 3D World) and open-source tools (TomoPy, ParaView, Tomviz, ImageJ/Fiji, Python).
• In-situ, real-time compatible

  Yes
• Correlated workflow available

  Yes

Instrument specification

• Technical aspects

  Server: Supermicro A+ Server 4125GS-TNRT
  CPU: 2x AMD EPYC 9754, 2.25 GHz, 128 cores / 256 threads each (256C/512T total), Socket SP5
  Memory: 1024 GB DDR5-6400 (16x Samsung 64 GB DDR5-6400 CL52)
  GPU: 4x NVIDIA Ada Lovelace L40S, PCIe 4.0 x16
   - 48 GB GDDR6 ECC VRAM each (192 GB total)
   - 18,176 CUDA Cores, 568 Tensor Cores (Gen4), 142 RT Cores (Gen3) per GPU
   - 864 GB/s memory bandwidth per GPU
   - Max power: 350 W per GPU
  OS Disk: 1x 960 GB Samsung SSD PM9A3 U.2 NVMe
  Data Disk: 2x 7.68 TB Samsung SSD PM9A3 U.2 NVMe (15.36 TB total)
  Network: 2x RJ45 10GbE LAN + dedicated IPMI port
  Power: 2000W (2+2) redundant, Titanium-level efficiency
  Software: Dragonfly 3D World (Comet/ORS), TomoPy, ParaView, Tomviz, ImageJ/Fiji, Python/Anaconda (NumPy, scikit-image, PyTorch)
  Remote access via web-based remote desktop with GPU-accelerated 3D rendering

• In-situ capabilities

  Not applicable (compute infrastructure). However, the server processes in-situ and operando 4D tomography datasets acquired at partner instruments.
• Correlated workflow

  TOMCAT processes and analyzes 3D/4D imaging data acquired from the MAPEX X-ray microscope (XRM), synchrotron facilities (e.g. ESRF, SOLEIL), FIB-SEM, and micro-CT systems. Dragonfly 3D World supports data import from all major imaging modalities. The server enables correlated analysis pipelines combining tomography data with electron microscopy, Raman spectroscopy, and other MAPEX-CF techniques. Remote access allows collaboration between MAPEX groups (materials science, geoscience, engineering) and external partners (e.g. Hochschule Bremen, FIBRE, EXCITE network).
• Additional measurement possibilities

  Deep learning model training and inference for automated image segmentation and classification. GPU-accelerated CT reconstruction (TomoPy, ASTRA). Python/NumPy/SciPy/PyTorch scripting for custom analysis workflows. Batch processing and high-throughput analysis of large image dataset collections. Quantitative 3D analysis: porosity, grain/fiber orientation, phase fractions, wall thickness, pore connectivity. 4D (time-resolved) analysis of dynamic processes (dissolution, fracturing, fluid displacement). FAIR-compliant data storage using open formats (OME-TIFF, Zarr).

Contact

• Application scientist

  Oliver Plümper
  Fachbereich 5 , Uni Bremen
  GEO
  Phone number 0421 218 65220
  pluemperprotect me ?!uni-bremenprotect me ?!.de
• Wolf-Achim Kahl
  MAPEX Center for Materials and Processes
  GEO-Building / Room 5070
  Phone number +49 421 218 65408
  wakahlprotect me ?!uni-bremenprotect me ?!.de
• Principal investigator

  Oliver Plümper
  

Instrument location

• Building

  University server room (rack-mounted)
• Faculty

  Fachbereich 5
• Institute

  MARUM
• Institute University

  Mineralogy and Nano-Geosciences