NVIDIA Jetson AGX Thor Developer Kit

An embedded computing platform for real-time AI and sensor fusion

Key points

• NVIDIA Jetson T5000 module with heatsink and reference carrier board
• 140 W DC power supply
• 802.11ax wireless network interface controller
• 1 TB NVMe SSD preinstalled in the M.2 Key-M slot
• Quick Start Guide

Description

The kit is built around an NVIDIA Blackwell GPU paired with 128 GB of memory and up to 2070 TFLOPS in FP4 (sparse) for running AI models, including generative models. Compute is complemented by a 14-core Arm Neoverse-V3AE CPU (64-bit) and a set of accelerators dedicated to perception workloads: PVA v3, an optical flow accelerator, and hardware video encoders/decoders.

For integration, the kit targets sensor fusion and high-throughput acquisition: multi-camera support, industrial I/O (buses and serial links), and network connectivity including a 4×25 GbE link via QSFP, a RJ45 Multi-GbE connector, and a Wi-Fi 802.11ax controller. Storage is provided by a preinstalled 1 TB NVMe SSD (M.2 Key-M), with NVMe support via PCIe and SSD support via USB 3.2.

The kit includes: Jetson T5000 module with heatsink and reference carrier board, 140 W DC power supply, 802.11ax controller, 1 TB NVMe (M.2 Key-M), and a Quick Start Guide.

Lab applications and robotics projects

• Humanoid robotics: prototyping perception → decision → action pipelines, local execution of VLA (Vision-Language-Action) models such as NVIDIA Isaac GR00T N.
• Multi-camera vision: acquisition of multiple streams, synchronization, and real-time inference with hardware encode/decode (H.264/H.265) for experimental recording or streaming.
• High-throughput sensor fusion: aggregating sensor data over 25 GbE networking and system interfaces to test fusion architectures (perception, motion estimation, tracking).
• Sensor processing and real-time pipelines: building processing chains with NVIDIA Holoscan (sensor processing), especially when stable latency is required.
• Embedded video analytics: developing visual agentic AI workloads with NVIDIA Metropolis and experimenting with agent workflows, including Video Search and Summarization (VSS).
• Robot integration and field interfaces: control and instrumentation via CAN, UART, SPI, I2C, PWM signal generation, and audio (I2S/AHUB) for test benches.