
Nvidia introduced Jetson AGX Thor: 2560 cores for robots
Nvidia company presented a development for physical AI – Jetson AGX Thor. This isn’t just a chip, this is literally a brain for future robots. Imagine — 2560 Blackwell cores and 128 GB of RAM in one compact device!
Computing density is very cool. 2070 teraflops with only 130W consumption! For comparison — RTX 4090 delivers 82-83 teraflops at 450W consumption. This is energy efficiency that makes autonomous robots reality!
The 14-core Arm Neoverse-V3 AE processor provides control logic. Built-in optical flow accelerators allow the robot to analyze surrounding reality in real-time. Interfaces for cameras, sensors, motors — all integrated into a unified platform!
The software ecosystem is also impressive. Support for all popular language models and computer vision models. NVIDIA Isaac for robotics, Metropolis for visual agentic AI, Holoscan for sensor processing. You can create AI agents directly on edge devices with video search and summarization functions!
Modularity opens boundless possibilities. Developers get not just hardware, but an entire technology stack for robot training. From vision language action models to complex solutions for industrial automation!
It seems robots are becoming autonomous intelligent systems. But the question isn’t whether we can create intelligent machines. But whether we’re ready for coexistence with this new form of intelligence.