- NVIDIA Jetson Agx Thor makes its debut with Blackwell GPU, 128 GB of memory and a storage of 1 to
- The first opinions describe a capable platform offering serious performance improvements compared to Jetson Orin
- The examiners agree that this will call upon developers who need powerful equipment for projects
Nvidia recently launched the Jetson AGX Thor Developer kit, a platform of $ 3,499 designed for robotics and the development of EDGE – and it had a warm initial reception of criticism.
In its heart is the Jetson T5000 module built on the Blackwell architecture, which combines a GPU with 2,560 coda cores, 96 tensor nuclei and a CPU at 14 Core Neoverse.
It is associated with 128 GB of LPDDR5X memory, offering more than 270 GB per second of bandwidth and 1 TB of integrated storage. Connectivity options include USB C, USB A, HDMI 2.1, Wi Fi 6e, Bluetooth, Gigabit Ethernet and a 100gbe port.
“Power Gobs”
The first criticisms of the kit are now underway, and they suggest that Nvidia has built an impressive option for developers despite its higher price compared to the Jetson Orin.
Hothardware The tests have shown that the AGX Thor is a solid artist, even with limited comparisons. Nvidia’s Arm64 containers took place gently, but tests against another Blackwell equipment were not possible, and the old Orin kit failed to finish the workloads.
The capacity difference was clear, however, with Orin closer to a RTX 3050 and Thor approaching RTX 5070 levels.
The models of large languages worked well in the tests. As Hothard Underlines: “LLM are an area where the jetson excels, and it must be supposed because humanoid robots should mix language with visual entries.”
The journal concluded that the kit has “powerbobs” for robotics and AI projects, noting: “If you want to execute very large AI models in a multi-tamer environment using NVIDIA software battery, the Jetson AGX Thor developer kit is an excellent tool for your device. Refine and update its software stack with additional AI capabilities. “”
Server The revision noted that the performances have been able to match Nvidia’s assertions, including 149.1 chips per second on Llama 3.1 8B against 150.8 expected.
CPU multi-threads have placed it near an AMD Ryzen AI 7 350 or Mac Mini M4, which was considered sufficient given its GPU point.
In reference tests, as expected, Thor has constantly exceeded the Orin through each model. The gains on smaller workloads such as Qwen 2.5-VL 7B and Llama 3.1 8B were modest, with Thor arriving about 1.3 times faster.
Deepseek-R1 7B has shown greater improvement to approximately 1.5 times speed. The most dramatic difference came with the Qwen 3 32B inference, where Thor has almost reached Orin’s performance five times, highlighting his strength when he performs larger and more demanding models.
Although the electric print can question battery systems, Serve Concludes The Thor offers the calculation and memory necessary for advanced robotics. He also managed to identify the SSD 1 TB included as WD / Sandisk SN5000S.
The two opinions described the Jetson Agx Thor as a step forward capable for advanced AI and robotics projects and have praised its mixture of calculation power, memory capacity and developer tools, while noting that software updates will be necessary to unlock its full potential.
As Serve In other words, the new kit “will sell like hotcakes. If you build a new, high-end robotics, this is the platform you want to do. ”
