Explore how two revolutionary AI hardware products are reshaping the future of enterprise intelligence
In today’s rapidly evolving artificial intelligence landscape, selecting the right computing platform has become crucial for business success. As a leader in AI computing, NVIDIA has introduced two powerful products designed for different application scenarios: Jetson Thor (based on the Thor chip) and DGX Spark.
This article provides a detailed comparison of these two products’ technical features, application scenarios, and commercial value to help you make the best choice for your business.
I. Core Positioning and Design Philosophy
NVIDIA Jetson Thor is an edge computing platform specifically designed for physical AI and robotics. Its name derives from the Norse god of thunder, symbolizing immense power. Jetson Thor aims to serve as the “brain” for next-generation robots, supporting real-time operation of large language models, vision language models, and vision language action models.
NVIDIA DGX Spark, on the other hand, is a desktop-level AI supercomputer, hailed as “the world’s smallest AI supercomputer.” It compresses data-center-level computing power into a compact body measuring just 150×150×50.5mm, designed specifically for developers, research institutions, and small businesses that need to develop and deploy AI models locally.
II. In-Depth Technical Specification Comparison
To help you better understand the differences between the two products, here is a detailed technical specification comparison table:
| Feature | NVIDIA Jetson Thor | NVIDIA DGX Spark |
|---|---|---|
| Core Architecture | Blackwell architecture GPU | GB10 Grace Blackwell superchip |
| AI Performance (FP4) | 2070 TFLOPS | 1000 TFLOPS |
| AI Performance (FP8) | 1035 TFLOPS | Not explicitly specified |
| CPU Configuration | 14-core Arm® Neoverse®-V3AE 64-bit CPU | 20-core ARM (10×Cortex-X925+10×Cortex-A725) |
| Memory System | 128GB LPDDR5X (273GB/s bandwidth) | 128GB LPDDR5X (273GB/s bandwidth) |
| Storage Support | 1TB NVMe + M.2 Key M slot | 1 or 4TB NVMe M.2 storage |
| Video Processing | 6x 4Kp60 (H.265) encode/4x 8Kp30 decode | Not explicitly emphasized |
| Power Range | 40W-130W | Not specified (desktop-level device, relatively high) |
| Connectivity | Extensive I/O options including QSFP slots, multi-gigabit RJ45 | 4x USB TypeC, 10GbE Ethernet, ConnectX-7 |
| Price Range | $3499 (development kit) | Approximately $3000 (domestic price) |
From the technical specifications, it’s evident that Jetson Thor leads significantly in AI performance, especially reaching 2070 TFLOPS at FP4 precision, compared to DGX Spark’s 1000 TFLOPS.
Jetson Thor also features rich video processing and hardware accelerators, such as the third-generation programmable vision accelerator (PVA), optical flow accelerator, etc. These specialized optimizations make it particularly suitable for robotics and autonomous driving scenarios.
III. Application Scenarios and Industry Solutions
Jetson Thor Applicable Scenarios:
- Humanoid Robot Development: Thor supports real-time operation of LLM, VLM, and VLA models with over 7x performance improvement, significantly advancing general-purpose robotics.
- Industrial Automation: In industrial automation, Jetson Thor can be used to achieve autonomous navigation and precise operation of robots.
- Intelligent Autonomous Driving Systems: The DRIVE Thor version is designed for autonomous vehicles and has already been adopted by automakers like BYD and GAC Aion.
- Medical Robotics: In the medical field, it can be used for precise control of surgical assistance robots.
DGX Spark Applicable Scenarios:
- Local AI Model Development and Debugging: DGX Spark enables researchers and developers to seamlessly conduct deep learning and AI development without relying on powerful cloud computing resources.
- Sensitive Data Processing: Suitable for scenarios requiring localized processing of sensitive data, such as healthcare and finance, meeting privacy compliance requirements.
- Generative AI Applications: Real-time debugging of generative AI applications (like text and image generation), training and inference of complex AI models like physical simulations.
- Education and Research: Provides relatively low-cost, high-performance AI computing resources for universities and research institutions, accelerating AI innovation and research.
IV. Software Ecosystem and Development Support
Both platforms benefit from NVIDIA’s powerful software ecosystem:
Jetson Thor runs the NVIDIA AI software stack, including:
- NVIDIA Isaac for robotics
- NVIDIA Metropolis for visual intelligence
- NVIDIA Holoscan for sensor processing
- Support for seamless cloud-to-edge experience
DGX Spark comes pre-installed with the NVIDIA AI software stack, including:
- CUDA-X AI platform
- NeMo framework
- RAPIDS data science acceleration tools
- Support for NVIDIA NIM microservices
Both products support simulation testing via NVIDIA Omniverse, forming a “trinity architecture of DGX (training), Omniverse (simulation), and Jetson Thor (inference).”
V. Commercial Value and Return on Investment
Jetson Thor is renowned for its outstanding energy efficiency, with an overall power consumption of only 40 to 130 watts and an energy efficiency ratio 3.5 times that of the previous Orin generation. This means it provides powerful computing capabilities while significantly reducing operational costs, making it particularly suitable for robotics and autonomous driving scenarios that require large-scale deployment.
DGX Spark offers data-center-level computing power at a relatively low threshold, with the domestic 128GB+1TB version starting at around 30,000 RMB. This enables small and medium-sized enterprises and individual developers to obtain supercomputing-level power at consumer-level costs, dramatically lowering the barrier to AI innovation.
Conclusion: How to Choose the Right Platform for Your Business
The choice between Jetson Thor and DGX Spark depends on your specific application needs:
- If you focus on physical AI, robotics, or autonomous driving and require powerful edge computing capabilities with high energy efficiency, Jetson Thor is your ideal choice.
- If your main work involves AI model development, training, and debugging, requiring local high-performance computing resources to handle sensitive data or reduce cloud dependency, DGX Spark will better meet your needs.
- For large enterprises, both products can form a complete AI development and deployment pipeline: develop and train models on DGX Spark, conduct simulation testing through Omniverse, and finally deploy to the Jetson Thor platform for operation.
The NVIDIA Jetson AGX Thor developer kit is now available, priced at $3,499. The Jetson T5000 production modules and Thor developer kit can be purchased through TWOWIN Technology.Twowin is NVIDIA NPN Partner, 11 years of industry experience.
NVIDIA Thor Developer Kit has arrived, please click here for consultation
