Nvidia says its new Jetson Thor robot brain can outpace rivals, and the first reviews suggest that’s true, and it will only get better

Nvidia Jetson AGX Thor debuts with Blackwell GPU, 128GB memory, and 1TB storage
Early reviews describe a capable platform offering serious performance improvements over Jetson Orin
Reviewers agree it will appeal to developers needing powerful hardware for projects

Nvidia recently launched the Jetson AGX Thor Developer Kit, a $3,499 platform designed for robotics and edge AI development – and it’s had a warm initial reception from reviewers.

At its heart is the Jetson T5000 module built on the Blackwell architecture, which combines a GPU with 2,560 CUDA cores, 96 Tensor cores, and a 14 core Arm Neoverse CPU.

It is paired with 128GB of LPDDR5x memory, offering over 270GB per second of bandwidth, and 1TB of onboard storage. Connectivity options include USB C, USB A, HDMI 2.1, Wi Fi 6E, Bluetooth, Gigabit Ethernet, and a 100GbE port.

“Gobs of horsepower”

The first reviews of the kit are now in, and they suggest Nvidia has built an impressive option for developers despite its higher price compared to the Jetson Orin.

HotHardware’s testing showed the Jetson AGX Thor to be a strong performer, even with limited comparisons. Nvidia’s Arm64 containers ran smoothly, but testing against other Blackwell hardware wasn’t possible, and the older Orin kit failed to complete workloads.

The gap in capability was clear, though, with Orin closer to an RTX 3050 and Thor approaching RTX 5070 levels.

Large language models performed well in testing. As HotHardware points out, “LLMs are one area where the Jetson excels, and it needs to since humanoid robots are expected to mix language with visual inputs.”

The review concluded that the kit has “gobs of horsepower” for robotics and AI projects, noting, “If you want to run very large AI models in a friendly multi-tasking environment using Nvidia’s software stack, the Jetson AGX Thor Developer Kit is a great new tool for your toolchest. The good news is that it handles all of those tasks with style and aplomb. And the device will likely get even better over time as Nvidia continues to refine and update its software stack with additional edge AI capabilities.”

ServeTheHome’s review found performance came close to matching Nvidia’s claims, including 149.1 tokens per second on Llama 3.1 8B versus the expected 150.8.

CPU multi-threaded performance placed it near an AMD Ryzen AI 7 350 or Mac Mini M4, which was considered sufficient given its GPU focus.

In benchmark testing, as expected, Thor consistently outpaced Orin across every model. Gains on smaller workloads such as Qwen 2.5-VL 7B and Llama 3.1 8B were modest, with Thor coming in around 1.3 times faster.

Deepseek-R1 7B showed a larger improvement at about 1.5 times the speed. The most dramatic difference came with Qwen 3 32B inference, where Thor nearly reached five times the performance of Orin, highlighting its strength when running larger and more demanding models.

While power draw may challenge battery systems, ServeTheHome concluded the Thor offers the compute and memory needed for advanced robotics. It also managed to identify the included 1TB SSD as a WD/SanDisk SN5000S.

Both reviews described the Jetson AGX Thor as a capable step forward for edge AI and robotics projects and praised its mix of compute power, memory capacity, and developer tools, while noting that software updates will be needed to unlock all of its poential.

As ServeTheHome put it, the new kit 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. ”