Another of the most important novelties that NVIDIA presented during GTC 2021, along with NVIDIA A2, was Jetson AGX Orin, a compact supercomputer that, as we see in the cover image, has a very small size, but is nevertheless the most powerful in its category, thanks to the integration of a new GPU based on the Ampere architecture.
As many of our readers will know, Ampere architecture marked a major breakthrough in the graphics industry thanks to the shader duplication by SM (from 64 to 128), to the introduction of the third generation tensor cores and second-generation RT cores, and also the jump to Samsung’s 8nm node, which improved both thermal and energy efficiency.
The NVIDIA Jetson AGX Orin is aimed at industries such as artificial intelligence, robotics, autonomous systems, medical devices and many other solutions aimed at what we know as “edge computing.” At the hardware level, this small supercomputer is equipped with a CPU with twelve Cortex-A78 cores 64-bit with 3MB L2 cache and 6MB L3 cache running at 2 GHz, it has a total of 32 GB of LPDDR5 memory with a bandwidth of 204.8 GB / s and it comes with an Ampere graphics solution that has from 2,048 CUDA cores (shaders). It also has 64 tensor cores to accelerate workloads based on artificial intelligence and inference.
According to the data provided by NVIDIA, this equipment also has 64 GB of eMMC 5.1 storage capacity, can work with up to 6 cameras, supports 8K resolution at 60 FPS, comes with 1 GbE connector, four 10 GbE connectors and can be configured in 15 watt, 30 watt and 50 watt versions. Obviously, configurations with a higher consumption allow to offer a higher level of performance. The GPU reaches a peak of 1 GHz of work frequency, and in total the Jetson AGX Orin can offer a power in INT8 of up 200 TOPs (200 billion operations per second, or trillion Anglo-Saxon).
As we can see, the improvement that NVIDIA has achieved at the hardware level compared to previous generations is very large, and the setup that the Jetson AGX Orin mounts is simply impressive for its small size. If you want to delve into its technical level, I recommend you take a look at this “white paper” that NVIDIA has recently published. In it you will find all the information.
