NVIDIA Jetson AGX Thor Developer Kit 128GB LPDDR5x, 2070 FP4 TFLOPS, 1TB NVMe SSD, Wi‑Fi 6E
NVIDIA Jetson AGX Thor Developer Kit 128GB LPDDR5x, 2070 FP4 TFLOPS, 1TB NVMe SSD, Wi‑Fi 6E
Yahboom
픽업 사용 가능 여부를 로드할 수 없습니다.
Overview
The NVIDIA Jetson AGX Thor developer kit is a Blackwell-architecture computing platform for robotics and edge AI development. It provides up to 2070 FP4 TFLOPS of AI compute and is specified as 7.5x higher performance than the previous-generation Jetson AGX Orin, with 3.5x higher energy efficiency. The kit is equipped with 128GB LPDDR5x memory (256-bit) with 273GB/s bandwidth to support large Transformer inference, high-concurrency video encode/decode, and multi-sensor data fusion.
Blackwell multi-instance GPU (MIG) is supported, allowing a single GPU to be partitioned into multiple isolated instances for concurrent workloads such as environment awareness, language interaction, and action planning. The platform integrates with the CUDA ecosystem and supports NVIDIA software stacks including Isaac, Isaac GR00T, Metropolis, and Holoscan.
Key Features
- AI performance: up to 2070 FP4 TFLOPS
- GPU architecture: NVIDIA Blackwell
- CPU: 14-core Arm Neoverse V3AE
- Memory: 128GB LPDDR5x, 256-bit; memory bandwidth 273GB/s
- Power consumption: 40–130W (130W also stated for 2070 FP4 TFLOPS operation)
- Storage (noted): 1TB NVMe SSD; SSD enclosure is provided for writing a system image
- Wireless (on-board): Wi‑Fi (2.4GHz/5GHz/6GHz) up to 2402Mbps; Bluetooth 5.3; installed inside the device before shipment
- System image (noted): SSD is supplied with a pre-installed system image; the system image mentions AI visual functions, ROS cases, offline large model cases, and OpenClaw course content
Specifications
| Product | Jetson AGX Thor Developer Kit |
| GPU | NVIDIA Jetson T5000 with Blackwell GPU |
| AI compute | Up to 2070 FP4 TFLOPS |
| CPU | 14 cores (Arm Neoverse V3AE) |
| Memory | 128GB LPDDR5x, 256-bit |
| Memory bandwidth | 273GB/s |
| Power consumption | 40–130W |
| Cooling | Heatsink and fan (fan and radiator) |
| Storage (noted) | 1TB NVMe SSD |
| Wireless | Wi‑Fi 2.4GHz/5GHz/6GHz up to 2402Mbps; BT 5.3; Wi‑Fi 6E module noted |
| External I/O (noted) | 2x USB‑A; 2x USB‑C; RJ45 connector (5GbE); DP connector; HDMI 2.1 connector; QSFP28 connector |
| Buttons/indicators | Power button; flashing button; reset button; power indicator light |
| Internal headers / interfaces (noted) | Audio (10Pin); Automation Interface (12Pin); CAN Bus (2x13Pin); USB Type‑C Debugging Interface; JTAG (10Pin); RTC (2Pin) |
| Chassis dimensions | 243 x 112.40 x 56.88 mm (units: mm); also noted length 240.86 mm |
| 3D model | 3D model file will be provided |
Performance Data (as stated)
Generative reasoning comparison notes
- Compared to Jetson AGX Orin, Jetson AGX Thor offers up to 5x faster for generative reasoning.
- With FP4 and speculative decoding, developers are stated to achieve an additional 2x performance acceleration on Jetson AGX Thor.
- Real-time multimodal AI example notes: using Qwen2.5‑VL‑3B VLM and Llama3.2 3B LLM, initial Token (TTFT) responses are stated to be below 200 ms, and each Token (TPOT) output time is stated to be far below 50 ms.
Tokens/s table (Jetson AGX Orin vs Jetson AGX Thor)
| Family | Model | Jetson AGX Orin (Tokens/s) | Jetson AGX Thor (Tokens/s) | Acceleration Multiplier |
|---|---|---|---|---|
| Llama | Llama 3.1 8B | 112.33 | 150.8 | 1.34 |
| Llama | Llama 3.3 70B | 7.38 | 12.64 | 1.71 |
| Qwen | Qwen3-30B-A3B | 76.69 | 226.42 | 2.95 |
| Qwen | Qwen3-32B | 16.84 | 79.1 | 4.7 |
| DeepSeek | DeepSeek-R1-Distill-Qwen-7B | 180.41 | 304.76 | 1.69 |
| DeepSeek | DeepSeek-R1-Distill-Qwen-32B | 16.96 | 82.63 | 4.87 |
| Qwen (VLM) | Qwen2.5-VL-3B | 216 | 356.86 | 1.65 |
| Qwen (VLM) | Qwen2.5-VL-7B | 154.02 | 252 | 1.64 |
| Llama (VLM) | Llama 3.2 11B Vision | 44.22 | 69.63 | 1.57 |
| GR00T (VLA) | GR00T N1 | 18.5 | 46.7 | 2.52 |
| GR00T (VLA) | GR00T N1.5 | 15.2 | 41.5 | 2.74 |
Ports & Functional Distribution (as labeled)
- DC power supply
- USB Type‑C interface (can be used as a power input interface)
- HDMI
- USB Type‑A interface
- DP interface
- Ethernet port
- QSFP28 connector
- Fan and radiator
System Image & Software Environment (as stated)
- SSD has pre-installed system image.
- Jetson AGX Thor comes with a 1TB empty SSD before shipment and an SSD enclosure is provided for writing the system image file.
- A 2TB SSD with a system image file is also mentioned. If choosing the 2TB SSD version kit, the original 1TB SSD that comes with the official kit is stated to be retained; the 2TB SSD in the package must be replaced by the user, and the original packaging will not be removed by default.
- Ubuntu 24.04-based Jetson system image is stated, with support for NVIDIA CUDA 13.0, TensorRT 10.13, cuDNN 9.12, and OpenCV 4.13.
Optional / Supported Modules (as described)
AI large model voice module (support stated)
- High-performance microphone and cavity speaker
- Far-field pickup, echo cancellation, and environmental noise reduction
USB industrial camera (support stated)
- 5MP
- Supports stable output of 30fps at 2592 x 1944
- 105° diagonal field of view (horizontal 81.8°, vertical 66°C as stated)
- Distortion-free images; dust-proof shell (as stated)
Tutorial Video
Support
For technical questions or help choosing the right configuration for the Jetson AGX Thor developer kit, contact support at https://rcdrone.top/ or support@rcdrone.top.
Details
The included 1TB NVMe SSD ships with a pre-installed system image, and the onboard wireless card supports Wi‑Fi 6E plus Bluetooth 5.3.
Build voice-enabled assistants or vision pipelines using supported peripherals like an AI voice module and USB industrial cameras.
Performance charts summarize generative reasoning gains and multimodal throughput targets for edge inference workloads.
Core platform highlights include up to 2070 FP4 TFLOPS AI compute, 128GB LPDDR5x memory, and a 14-core Arm CPU.
A clear port-and-header map helps plan integration, from USB and display outputs to CAN bus, debug, and other internal interfaces.
Mechanical drawings list the chassis dimensions in millimeters, and a 3D model file is available for enclosure design.
The updated system image is based on Ubuntu 24.04 and references common accelerated libraries in the CUDA software stack.
In the box: the Jetson AGX Thor developer kit, power supply, and necessary cables for initial bring-up.
The Jetson AGX Thor 128GB developer kit ships with a power adapter and USB‑C data cable, with an optional 1TB/2TB SSD and SSD enclosure, while the Vision Advanced Kit adds items like a DP‑to‑HDMI cable, USB industrial c
Jetson AGX Thor 128GB kit options bundle the board with a power adapter, USB‑C cable, DP‑to‑HDMI lead, USB industrial camera, SSD enclosure, USB 3.0 hub, and optional NVMe SSD and touchscreen add‑ons.
