A high-performance data layer for large 3D and 4D engineering datasets. Random-access compression, direct network streaming from object storage, and GPU-native decompression — so applications access only the fields, regions, timesteps, and resolutions required for the current task.
Physics AI is constrained by massive 3D datasets that are extremely expensive to store, move, visualize, and prepare for training workflows.
Zibra system includes universal representation support (volumetric grids, volume & surface meshes, point clouds, splats) and keeps the data highly compressed using optimal lossless or error-bound compression strategies per each representation.
Instead of copying entire datasets, applications utilize random-access network streaming to fetch only the required chunks (fields, regions, points, timesteps), followed by GPU-native decompression at the exact point of consumption.
Traditional compression reduces file size. ZibraXYZ is designed to reduce the amount of data that must be read, moved, and decompressed for every operation.
Data is organized into independently accessible compressed pages, allowing applications to address only the working set they need.
Compression profiles can be adapted to the representation and numerical requirements of each field, with support for lossless and controlled-error workflows.
ZibraXYZ turns massive 3D and 4D datasets into interactive visualization workflows by streaming and decompressing only the data required for the current view.
Explore volumetric fields, meshes, slices, regions of interest, timesteps, and levels of detail while the underlying dataset remains compressed in object storage. Progressive loading makes the relevant data available first, enabling responsive exploration of datasets far larger than local workstation memory.
ZibraXYZ lets AI pipelines access selected parts of large engineering datasets directly from their compressed representation, reducing storage, preprocessing, and data-loading overhead.
A key capability is representation-selective access: a full VTU dataset can contain mesh topology, geometry, and multiple fields, while a training workflow can request only the point data it needs. For point-based models, ZibraXYZ can stream selected point attributes from the original compressed dataset without fetching the mesh topology, turning rich simulation files into efficient AI training sources.
Use the SDK to integrate compressed remote data into custom tools, preprocessing pipelines, simulation software, and internal applications.
SDK documentation →
▋
All displayed trademarks are the property of their respective owners. All company, product and service names are used for identification purposes only. Use of these names and trademarks does not imply endorsement