AICE NX
Compact, fanless local computing. The J4012 reference includes 16 GB LPDDR5, dual GbE and up to 100 TOPS INT8 with sparsity.
AICEAICE · EDGE INTELLIGENCE
The intelligence layer
close to reality.
AICE is the operational edge for computer vision: it runs computer-vision inference close to enabled sources on project-sized Edge configurations and sends contextual events to AOI.
Talk to RTFicial
AICE / EDGE RUNTIME NVIDIA POWERED
REAL-TIME INFERENCE
VIDEOUSE CASE IN ACTION
AICE Next Gen: edge intelligence in real time.
From AICE Edge to the operational platform: multi-camera video, accelerated inference and contextual events running close to the source.
Video · sound available · controls enabled01WHY AICE
Not only detection.
Operational intelligence.
AICE captures, infers, detects deviations, evaluates rules, creates evidence and publishes metadata. By running close to the source, it keeps the operation responsive while reducing unnecessary data movement.
04simultaneous cameras
RTSPvideo ingestion
OFFLINElocal persistence
AICE CAPABILITIES
01
Multi-camera ingestion
RTSP streams, direct camera input and a local MediaMTX bridge for compatible and non-compatible sources.
02
Accelerated inference
GStreamer and NVIDIA TensorRT run detection models on AICE NX / DGX.
03
Rules and persistent alerts
Hierarchies, regions, thresholds and person/object tracking turn detections into open, observed and closed alerts.
AICE / NVIDIAAICE × NVIDIA
Industrial intelligence
accelerated at the edge.
AICE turns NVIDIA accelerated computing into real-time perception for industrial environments. AOI then connects every event to context, workflows and operational decisions.
Explore AICE × NVIDIA Built with NVIDIA technology. Designed for industrial reality.
AICE / EDGE RUNTIME
AICE EDGE NODENVIDIA accelerated edge computing, multi-camera inference and local evidence.
01Cameras / OT→
02AICE Edge→
03AOI Context→
04Action
REFERENCE CONFIGURATIONS
AICE Edge is sized for each project’s sources and models.
AICE DGX
AICE on NVIDIA DGX Spark with GB10, 128 GB unified memory and a 4 TB reference configuration. Up to 1 PFLOP FP4 with structured sparsity.
These are reference hardware configurations. Performance, topology, models, sources and installation are defined and validated for each project.
AICE / MULTI-CAMERA INFERENCE
PPE · ZONES · TRACKING
02FROM VIDEO TO EVIDENCE
See the operation.
Keep the proof.
AICE maintains state by region and person, creates snapshots for relevant alerts and publishes the metadata required by the operational layer.
Bounding boxes and object metadataAlert lifecycle: open, observe, closeSnapshots in object storage
03AICE / SYSTEM COMPONENTS
01aice-main
Runtime and local administration.
02aice-cam
Runtime and local administration.
03MediaMTX
Local camera relay.
04Rules engine
Operational conditions and hierarchies.
05Local evidence queue
Retries during connectivity loss.
06AOI platform
Events, telemetry and integration.
04BUILT FOR THE EDGE
Keep operating
when the network changes.
Local persistence stores snapshots and pending events, retries publication and synchronizes when connectivity returns. The edge remains useful even when the cloud is temporarily unavailable.
EDGE ONLINELocal first.
Cloud connected.
DETECT→QUEUE→SYNC
05AICE · NVIDIA POWERED
VIDEO → INFERENCE → RULES → EVIDENCE → EVENT
Every event can carry location, region, time, detected objects, bounding boxes, snapshot and operational status.
Talk to RTFicial