We recently tested OREC on the HP ZGX Nano AI Station, a compact desktop workstation designed for AI, confirming compatibility with OREC.
The goal was to see whether OREC could be shipped as pre-installed on the device, ready to use from day one and operate on a device of the ZGX Nano AI Station class, either at a command center or at the tactical edge.
We have confirmed that the performance is comparable to the workstations we usually run the AI models on, opening a new range of use cases that require running OREC AI models on compact hardware.
The test validates operations closer to the sensor
Defence and intelligence users often can't send imagery to the cloud. Their SAR data stays inside their own networks, frequently in air-gapped environments. AI in defense applications has special requirements for hardware, the setup, and configuring the analytical stack.
The first finding is that for the end-users requiring it, the setup of our AI analytical capabilities can be as simple as getting a device with the software pre-installed. This gives a very fast onboarding for enabling AI analysis in the pipeline.
A device this size lets OREC run close to the sensor, as well. At a forward command post, or at a ground station receiving data from a UAV-mounted radar. Imagery doesn't have to travel back to a central facility before an analyst sees a detection.
That shortens the time between collection and a result the decision-maker can act on. It also follows where the field is moving, with AI built into the sensor chain and working at the tactical edge, where connectivity is limited or contested.
Performance: comparable to our training workstations
During the test, we ran an experimental build of OREC, the engine for detecting and classifying vessels, vehicles and aircraft in SAR imagery. The ZGX Nano is built around the NVIDIA GB10 Grace Blackwell Superchip with 128 GB of unified memory. It has a 15 × 15 cm footprint, stands about 5.5 cm tall and weighs from 1.25 kg. That is a fraction of the footprint of the workstations we normally train and use our models on. Even so, its performance held up against them.
We’re satisfied with the results, as they’ve confirmed that strong performance of OREC is not limited to larger devices.
Our thanks go to HP for providing the ZGX Nano AI Station and for the support throughout the tests.
