ZED Cameras and NVIDIA Jetson Thor: High-Performance Perception at the Edge

Get ready! Jetson Thor support for ZED cameras is arriving in ZED SDK 5.1, launching Q4 2025!

From AGX Orin to Thor: performance that unlocks new workloads

Jetson AGX Thor features a Blackwell-generation GPU delivering up to 2,070 FP4 TFLOPs (1,035 FP8 TFLOPs), unlocking significantly higher performance compared with AGX Orin:

• 7.5× more AI compute, enabling larger models or multiple perception workloads to run simultaneously.
• ~4× better AI TFLOPs per watt, allowing more efficient multi-sensor processing without increasing power consumption.
• 2.6× faster Arm Neoverse CPU with 14 cores, speeding up preprocessing and overall system orchestration.
• Faster data handling with the increased 256-bit 128 GB LPDDR5X RAM with 273 GB/s bandwidth
• Four 25 GbE links, supporting high-bandwidth sensor capture and real-time data fusion.
• Dual video engines, handling up to 4× more 1080p30 H.265 streams, ideal for multi-camera setups.

The Jetson Thor module is powered by the new Jetpack SDK 7, based on Ubuntu 24.04, which introduces the new CUDA 13 and TensorRT 10.

Together, these capabilities let engineers run advanced perception pipelines on a single module, enabling robots to process more sensors, respond faster, and operate more intelligently in real environments.

How ZED cameras can leverage Thor

ZED 2i streaming live on ZED Explorer with Thor Dev Kit

Integrating ZED cameras with Jetson AGX Thor provides a complete perception architecture for robotics. The system can process real-time depth, motion tracking, and object recognition, running multiple pipelines concurrently while leaving compute headroom for planning and control tasks.

A key enabler is NVIDIA’s Multi-Instance GPU (MIG) technology, which partitions a single GPU into isolated instances so that perception, mapping, and planning workloads can run independently in parallel. By providing each task with dedicated compute resources and memory, MIG prevents workloads from interfering with one another, helping maintain consistent performance even under demanding conditions.

In addition, the platform’s high-bandwidth memory and 25 GbE interfaces simplify multi-camera capture and sensor fusion, ensuring that large amounts of sensor data can be processed efficiently. Complementing this, Holoscan sensor-to-GPU pipelines reduce latency by keeping data on the GPU instead of transferring it through the CPU.

The Jetson AGX Thor development kit natively supports Stereolabs USB cameras, including ZED 2i and ZED Mini. Full GMSL2 connectivity for the ZED X family will be enabled through our partners’ solutions.

Where this matters in industry

The combination of ZED cameras and Jetson AGX Thor supports a wide range of robotics applications:

• Manufacturing: robots can handle inspection, bin picking, and worker safety on the same module. With higher compute and video throughput, they stay accurate and responsive, boosting productivity without compromise.
• Humanoids: projects in this field demand vision for locomotion, recognition, and manipulation all at once. ZED depth and Thor’s MIG partitioning keep these pipelines running in parallel, enabling safe, natural collaboration.
• Logistics and warehousing: AMRs and forklifts gain wide‑field 3D vision for aisle navigation, pallet handling, and dock operations. Thor’s efficiency lets fleets scale camera counts while keeping mapping and obstacle avoidance responsive.
• Smart facilities and infrastructure: safety zones, flow analytics, and asset tracking can be consolidated onto a single system, reducing hardware sprawl across large sites.

Let’s build

If you’re planning a Thor‑based platform in manufacturing, logistics, or large facilities, Stereolabs cameras provide the perception backbone to deploy robots that do real work.

Let’s talk about your pipeline, evaluation setup, and multi‑camera design with ZED cameras on NVIDIA Jetson AGX Thor.