Nvidia Cuts Thor Robotics Chip Size in Half With New T3000 and T2000 Modules
Nvidia launched the Jetson T3000 and T2000 modules for its Thor robotics platform, reducing the size and power consumption of its AI accelerator by half compared to the existing T5000 chip while delivering 865 and 400 FP4 teraflops of performance, respectively.
The new chips, which feature 32GB and 16GB of memory, aim to lower hardware costs and move humanoid robotics out of research labs for mass-market deployment. A new software suite cuts memory usage by up to 15GB, enabling developers to run larger models on the smaller modules, with the first units scheduled to ship in the first quarter of 2027.
From the sources (6 posts)
@stockmktnewzNVIDIA $NVDA JUST SHRUNK THE SIZE OF ITS ROBOTICS CHIP IN HALF WITHOUT LOSING PERFORMANCE NVIDIA introduced two new modules, the T3000 and T2000, built on its Thor architecture for humanoid robots and edge AI, aimed at moving robotics from
@wallstengine$NVDA JUST CUT ITS THOR ROBOTICS COMPUTER TO HALF THE SIZE AND POWER The new T3000 module is roughly half the size and power of NVIDIA’s T5000, while delivering similar inference performance on multimodal AI workloads. Inside: → 865 FP4
@stocksavvyshay$NVDA unveiled two new Blackwell-powered modules called T2000 and T3000 to bring robotics and edge AI to the mass market. Leading robotics companies are already adopting Jetson Thor including $AMZN Robotics.
@polymarketmoneyBREAKING: Nvidia unveiled two new Blackwell-powered modules called T2000 and T3000 to bring robotics and edge AI to the mass market.
@stocksavvyshay$NVDA also cut the size of its Thor robotics computer in half without sacrificing performance:
@nvidiaroboticsIntroducing the new NVIDIA Jetson T3000 and T2000 modules.🎉 Expanding the Jetson Thor platform, these compact supercomputers bring Blackwell-class performance to mainstream robotics and edge AI in a more power-efficient design. Paired wit