- Nvidia Jetson Agx Thor debuts with Blackwell GPU, 128 GB memory and 1 TB storage
- The first reviews describe a capable platform that offers serious performance improvements on Jetson Orin
- Reviewers agree that it will attract developers who need powerful hardware for projects
Nvidia recently launched the Jetson Agx Thor developer kit, a $ 3,499 platform designed for Robotics and Edge development, and has had a warm initial reception of the reviewers.
In its heart is the Jetson T5000 module built on the Blackwell architecture, which combines a GPU with 2,560 CUDA nuclei, 96 tensioner cores and a CPU of 14 CPU nucleus of ARM.
It is combined with 128 GB of LPDDR5x memory, which offers more than 270 GB per second bandwidth and 1 TB of storage on board. Connectivity options include USB C, USB A, HDMI 2.1, Wi Fi 6E, Bluetooth, Gigabit Ethernet and a 100gbe port.
“Power Gobs”
The first kit reviews are now, and suggest that Nvidia has created an impressive option for developers despite their highest price compared to Jetson Orin.
Hothardware The tests showed that the Jetson Agx Thor was a strong performance, even with limited comparisons. NVIDIA ARM64 containers worked without problems, but tests against another Blackwell hardware were not possible, and the oldest Orin kit failed to complete workloads.
However, the capacity gap was clear, with Orin closer to an RTX 3050 and Thor approaching RTX 5070 levels.
Large language models worked well in the tests. As Hothardware He points out: “LLMs are an area where Jetson stands out, and needs to do so since humanoid robots are expected to mix the language with visual entries.”
The revision concluded that the kit has “drops of power” for robotics and artificial intelligence projects, pointing out, “if you want to execute very large models in an environment of multiple friendly tasks using the NVIDIA software stack, the Jetson Agx Thor developer kit is a great tool for your tool. Continue refining and updating its software battery from Ai Edge “.
Servethehome’s The review found that the performance was close to matching Nvidia’s claims, including 149.1 tokens per second in flame 3.1 8b versus the expected 150.8.
CPU multiprocess yield placed it near an AMD Ryzen AI 7 350 or Mac Mini M4, which was considered sufficient given its GPU approach.
In the reference tests, as expected, Thor constantly exceeded Orin in each model. Profit in smaller work loads such as QWEN 2.5-VL 7B and call 3.1 8b were modest, with Thor entering around 1.3 times faster.
Deepseek-R1 7B showed an improvement greater than 1.5 times the speed. The most dramatic difference occurred with the inference Qwen 3 32b, where Thor almost reached the performance of Orin five times, highlighting his strength when larger and more demanding models are executed.
While the power can challenge battery systems, Servethehome He concluded that Thor offers the necessary calculation and memory for advanced robotics. He also identified the 1TB SSD included as a SN5000 WD/Sandisk.
Both revisions described the Jetson Agx Thor as a capable step for AI and Robotics projects of Edge and praised their combination of computing power, memory capacity and developer tools, while noting that software updates will be necessary to unlock their entire poental.
As Servethehome That, the new kit “is going to sell as Hotcakes. If you are building next -range next generation robotics, this is the platform you want to do.”
