NVIDIA Jetson AGX Thor: High-performance AI developer platform for robotics.
2070 FP4 TFLOPS: Delivers powerful computational performance for complex AI tasks.
Jetson AGX Thor Developer Kit: Ideal for building physical AI and humanoid robot systems.
128 GB LPDDR5X Memory: Ensures fast and efficient data processing for advanced robotics applications.
Jetson Blackwell: Integrates the latest Blackwell GPU for next-gen AI and machine learning.
Tutorials:https://www.yahboom.net/study/AGX-Thor
NVIDIA Jetson AGX Thor developer kit is NVIDIA's latest generation robot computing platform. It is based on the Blackwell GPU architecture and provides up to 2070 FP4 TFLOPS of AI computing power at 130W power consumption. Compared to the previous generation Jetson AGX Orin, Thor has improved performance by 7.5 times and energy efficiency by 3.5 times. Equipped with 128GB LPDDR5x RAM and 14 core Arm Neoverse V3AE CPU, Thor can easily run various generative AI models -- from visual language action models to mainstream large language models and visual language models.Thor enables robots to interact with humans and the physical world in real-time and intelligently. Through multi instance GPU (MIG) technology, a single GPU can be divided into multiple isolated instances to simultaneously perform tasks such as environment awareness, language interaction, and action planning.In terms of software ecology, Thor is fully compatible with NVIDIA's cloud to edge software stack, including the Isaac robot development platform, Metropolis visual AI framework, and Holoscan real-time sensor processing framework.
1.Equipped with NVIDIA Blackwell GPU architecture, Jetson AGX Thor provides up to 2070 FP4 TFLOPS of AI computing power, which is 7.5 times higher than the previous generation Jetson AGX Orin.
2.Equipped with 128GB 256 bit LPDDR5x RAM and a bandwidth of 273GB/s. Provides sufficient memory space for large-scale Transformer model inference, high concurrency video encoding and decoding, and multi-sensor data fusion, enabling robots to simultaneously handle multiple tasks such as vision, speech, and motion planning without lagging.
3.Support Blackwell architecture's multi instance GPU technology, which can divide a single GPU into multiple isolated instances and run tasks such as environment awareness, language interaction, and action planning simultaneously.
4.Thor is not only a hardware development board, but also a complete development ecosystem. It supports NVIDIA Isaac robot development platform, Isaac GR00T humanoid robot basic model, Metropolis visual AI framework, and Holoscan real-time sensor processing framework, seamlessly integrating with the CUDA ecosystem.
