On-premise CT inspection just got significantly faster. Manufacturers running Lumafield Triton or Neptune in secure environments can now match the analytical depth and throughput of any cloud-connected system, without data ever leaving the facility. Lumafield's new edge compute cluster is powered by the NVIDIA DGX Spark, which is built around NVIDIA's GB10 Grace Blackwell Superchip. The result is scaled CT for NDI / non-destructive testing.

CT inspection has become dramatically more powerful and more pervasive. Lumafield has been at the center of that shift, bringing automated, AI-driven CT to the production floor and making 100% inspection a practical reality for manufacturers across industries. More scanning means more data, and more advanced analysis requires more compute. The cloud handles that demand well for many organizations. For users who can’t send their data offsite, the DGX Spark now provides the same computational muscle on-premise.

That's especially true for quality engineers working under ITAR requirements, FDA data governance policies, or internal data sovereignty mandates. For those teams, full analysis capability, including automated defect detection, porosity analysis, dimensional measurement, and assembly verification, is now available without any data leaving the building. Full capability. Full control.

Purpose-built compute for CT at any scale

The NVIDIA GB10 Grace Blackwell Superchip delivers one petaFLOP of AI compute alongside 128GB of unified memory, where the CPU and GPU share a single memory pool rather than operating as separate systems with a bottleneck between them. For CT reconstruction and AI inference, this means faster analysis and faster time to insight at the line.

Triton Performance can complete a full CT scan in under 10 seconds. The DGX Spark configuration is built to keep pace, processing each scan and returning a quality decision before the next part arrives. Early performance testing shows throughput of 100 UPH / units per hour under continuous production conditions.

A single DGX Spark unit handles demanding inspection workloads for most environments. For the highest-throughput lines, a Triton configuration can support up to eight independent DGX Spark units, each delivering one petaFLOP of AI compute and 128GB of memory, running parallel reconstruction processes to keep pace with your line.

For Neptune users in secure lab and R&D environments, the DGX Spark delivers the same on-premise compute advantage. That means faster reconstruction and deeper analysis without data leaving the facility.

Made for the industries that can't compromise

In aerospace and defense, the internal geometry of a part can be as sensitive as the part itself. On-premise CT inspection ensures that data never leaves infrastructure outside your direct control, a requirement for many classified and export-controlled programs.

In medical device manufacturing, sealed assemblies and implantable components require tightly controlled, traceable inspection data resident within validated systems.

In semiconductor and advanced electronics manufacturing, proprietary geometries and process parameters represent core intellectual property. A single leaked process geometry can expose your entire development roadmap to a competitor. Keeping inspection data local eliminates an entire category of supply chain exposure.

In all of these contexts, Triton and Neptune deliver high-resolution CT inspection, detecting voids, inclusions, porosity, misalignment, dimensional drift, and assembly errors, with full data residency.

More capable where it counts

Lumafield’s on-premise systems with automated part handling, Ultra-Fast CT, Voyager integration, and lights-out operation, are now supercharged to serve manufacturers whose facilities demand that intelligence live on-site.

If your compliance framework or security posture has kept high-performance automated CT inspection out of reach, we'd like to show you what is now possible. Contact us to discuss your requirements.