- Nvidia Jetson AGX Thor for the first time with GPU Blackwell, 128 GB memory, 1 TB store
- Early reviews describe a platform capable of making serious performance improvements on Jetson Orin
- The auditors agree that it will resume developers who need strong projects
NVIDIA recently launched the Jetson Agx Through, a $ 3,499 platform designed to develop robots and develop Edge AI – and had a warm initial reception from auditors.
In its heart there is the Jetson T5000 unit that was built on the Blackweell structure, which combines the graphics processing unit with 2,560 cuda cuis, 96 cores, and the 14 Core Arm Noverse Core Core Unit.
It is paired with a 128 GB memory of LPDDR5X memory, providing more than 270 GB per second of the frequency domain width, and 1 TB of storage on the plane. Contact options USB C, USB A, HDMI 2.1, Wi Fi 6e, Bluetooth, Gigabit Etternet and 100GBE port.
“Gobs of Harrespowe”
The group’s first reviews are now, and they suggest that NVIDIA has built an impressive option for developers despite its high price compared to Jetson Orin.
Hothardware’s The test showed that Jetson Agx Thor is a strong performance, even with limited comparisons. Nvidia’s ARM64 containers continued smoothly, but the test against other BlackWell devices was not possible, and the older Orin group failed to complete the work burdens.
The gap in the ability was clear, with Orn approached the RTX 3050 and Thor levels approaching RTX 5070 levels.
Large language models were a good performance in the test. like Hothadware He points out that “LLMS is one region in which Jetson excels, and it should be because human robots are expected to mix the language with visual inputs.”
The review concluded that the group contains a “horse force” for robots and artificial intelligence projects, “If you want to run very large Amnesty International models in a multi -task friendly environment using a NVIDIA software staple, Jetson Agx Thor Stotes will be better by getting new tools better. Refine and update its own programs with the Edge Ai additional capabilities.”
Servethehome’s The review found that the performance is close to matching NVIDIA claims, including 149.1 icons per second on Llama 3.1 8B for 150.8 expected.
Place it in the multi -threaded central processing unit near AMD Ryzen Ai 750 or Mac Mini M4, which was considered sufficient given the GPU concentration.
In the standard test, as expected, the Orin bull is constantly exceeded through each model. The gains on smaller work burden such as QWEN 2.5-VL 7B and Llama 3.1 8B were modest, with thor about 1.3 times faster.
Deepseek-R1 7B showed a greater improvement in about 1.5 times the speed. The most dramatic difference came with QWEN 3 32B conclusion, as a bull reached almost five times the performance of Orine, highlighting its strength when operating larger and more demanding models.
While pulling energy may challenge battery systems, Servethehome He concluded that Thor offers the account and memory needed for advanced robots. I also managed to determine the SSD 1 TB as WD/Sandisk Sn5000.
Both auditors described Jetson Agx Thor as a step capable of front for AI and Robotics projects, and praised its mixture of arithmetic energy, memory capacity, and developers, noting that software updates will be necessary to cancel all of the lock.
like Servethehome In the words, the new collection is “you will sell like hot cakes. If you are building robots from the next generation of the next generation, this is the platform you want to do.”