Nvidia Jetson AGX Thor debuts with Blackwell GPU, 128GB reminiscence, and 1TB storage
Early evaluations describe a succesful platform providing severe efficiency enhancements over Jetson Orin
Reviewers agree it’s going to enchantment to builders needing highly effective {hardware} for tasks
Nvidia not too long ago launched the Jetson AGX Thor Developer Package, a $3,499 platform designed for robotics and edge AI improvement – and it’s had a heat preliminary reception from reviewers.
At its coronary heart is the Jetson T5000 module constructed on the Blackwell structure, which mixes a GPU with 2,560 CUDA cores, 96 Tensor cores, and a 14 core Arm Neoverse CPU.
It’s paired with 128GB of LPDDR5x reminiscence, providing over 270GB per second of bandwidth, and 1TB of onboard storage. Connectivity choices embody USB C, USB A, HDMI 2.1, Wi Fi 6E, Bluetooth, Gigabit Ethernet, and a 100GbE port.
It’s possible you’ll like
“Gobs of horsepower”
The primary evaluations of the package at the moment are in, they usually recommend Nvidia has constructed a powerful choice for builders regardless of its greater value in comparison with the Jetson Orin.
HotHardware’s testing confirmed the Jetson AGX Thor to be a robust performer, even with restricted comparisons. Nvidia’s Arm64 containers ran easily, however testing towards different Blackwell {hardware} wasn’t doable, and the older Orin package failed to finish workloads.
The hole in functionality was clear, although, with Orin nearer to an RTX 3050 and Thor approaching RTX 5070 ranges.
Massive language fashions carried out properly in testing. As HotHardware factors out, “LLMs are one area where the Jetson excels, and it needs to since humanoid robots are expected to mix language with visual inputs.”
ServeTheHome’s evaluation discovered efficiency got here near matching Nvidia’s claims, together with 149.1 tokens per second on Llama 3.1 8B versus the anticipated 150.8.
CPU multi-threaded efficiency positioned it close to an AMD Ryzen AI 7 350 or Mac Mini M4, which was thought-about adequate given its GPU focus.
In benchmark testing, as anticipated, Thor constantly outpaced Orin throughout each mannequin. Beneficial properties on smaller workloads reminiscent of Qwen 2.5-VL 7B and Llama 3.1 8B had been modest, with Thor coming in round 1.3 instances quicker.
Deepseek-R1 7B confirmed a bigger enchancment at about 1.5 instances the pace. Essentially the most dramatic distinction got here with Qwen 3 32B inference, the place Thor practically reached 5 instances the efficiency of Orin, highlighting its energy when working bigger and extra demanding fashions.
Whereas energy draw might problem battery methods, ServeTheHome concluded the Thor presents the compute and reminiscence wanted for superior robotics. It additionally managed to establish the included 1TB SSD as a WD/SanDisk SN5000S.
Each evaluations described the Jetson AGX Thor as a succesful step ahead for edge AI and robotics tasks and praised its mixture of compute energy, reminiscence capability, and developer instruments, whereas noting that software program updates will probably be wanted to unlock all of its poential.
As ServeTheHome put it, the brand new package is “going to sell like hotcakes. If you are building high-end next-generation robotics, this is the platform you want to do it on. ”
(Picture credit score: ServeTheHome)
You may also like
