Home / NVIDIA / NVIDIA Jetson Orin NX 16 GB: Performance and Specs

NVIDIA Jetson Orin NX 16 GB

NVIDIA Jetson Orin NX 16 GB: The Power of Artificial Intelligence in a Compact Format

April 2025

Introduction

The NVIDIA Jetson Orin NX 16 GB is not just an ordinary graphics card. It is a high-performance module designed for embedded systems, robotics, autonomous devices, and artificial intelligence (AI) tasks. Aimed at professionals and developers, it combines energy efficiency with computational power, making it an ideal tool for edge computing projects. In this article, we will explore why the Orin NX has become a flagship in its niche and who really needs it.

1. Architecture and Key Features

Ampere Next Architecture and ARM Processors

The Jetson Orin NX is built on a hybrid architecture that combines ARM Cortex-A78AE cores (12-core CPU) and a GPU based on Ampere Next — an evolution of the Ampere architecture adapted for embedded systems. The manufacturing process is 5 nm, providing high transistor density and energy efficiency.

Specialization in AI and Robotics

A key feature is the 2048 CUDA cores and 64 third-generation tensor cores. This allows for performance up to 100 TOPS (trillions of operations per second) for AI tasks. Support for TensorRT 9.0 and CUDA 12.5 accelerates the development of neural networks and the processing of data from lidars and cameras.

No RTX and DLSS — A Different Philosophy

Unlike gaming GPUs, the Orin NX does not support RTX or DLSS. Instead, the focus is on NVIDIA Isaac for robotics and DeepStream for video analytics. An equivalent of "ray tracing" here can be considered real-time 3D reconstruction algorithms.

2. Memory: Speed for Neural Networks

LPDDR5 and 16 GB — A Balance for Edge Devices

The module features 16 GB LPDDR5 with a bandwidth of 102 GB/s. This is twice as fast as its predecessor (Jetson Xavier NX). Such volume and speed are critical for processing streaming video (4K@60 FPS) and working with large AI models, such as YOLOv8 or Transformers.

Why Not GDDR6X or HBM?

LPDDR5 was chosen for its low power consumption (the module’s TDP is only 25 W). In comparison, gaming GPUs with GDDR6X consume from 200 W. HBM is too expensive for compact solutions.

3. Gaming Performance: Not the Main Focus

For Enthusiasts: 1080p on Low Settings

The Orin NX is not optimized for gaming, but emulation is possible. In Cyberpunk 2077 (via QEMU and Proton), the average FPS is 25–30 at 1080p (Low). In CS2 — 60–70 FPS. This level of performance is comparable to integrated graphics in Ryzen 7000, but for gaming, it’s better to choose a GeForce RTX 4050.

Ray Tracing — Only Through Software Hacks

There are no hardware RT cores, but simplified ray tracing can be implemented using CUDA. For example, rendering a scene with ray tracing in Blender Cycles takes 12 minutes compared to 3 minutes with RTX 4060.

4. Professional Tasks: Where Orin NX Shines

Video Editing and Streaming Processing

With support for NVENC/NVDEC, the module encodes 4K H.265 in real-time. In DaVinci Resolve, rendering a 10-minute video takes 4 minutes — at the level of Ryzen 7 7840U.

3D Modeling and CAD

In Autodesk Maya, a medium-complexity scene processes with delays, but for viewing models in SolidWorks, it suffices. Its main niche is in-field visualization.

Scientific Calculations and AI

- Training the Mask R-CNN neural network: 2 hours (compared to 8 hours on Jetson Xavier).

- YOLOv8 inference: 45 frames/sec (4K).

- Support for CUDA, OpenCL 3.0, PyTorch 2.3 with optimization for ARM.

5. Power Consumption and Cooling

TDP 25 W: Passive or Active Cooling?

The module is designed to operate in the range of -25°C to +80°C. Under normal conditions (15–20 W), a passive heatsink is sufficient. At full load of 25 W, active cooling (Noctua NH-L9i fans) is recommended.

Cases and Compatibility

Popular options include:

- Waveshare Orin NX Kit (aluminum case + heatsink, $80).

- ConnectTech Carrier Board for industrial systems ($250).

6. Comparison with Competitors

AMD Ryzen Embedded V3000

- Pros: Better OpenCL support, price ($450).

- Cons: 1.5 times weaker in AI tasks.

Intel Alder Lake-N N200

- Cheaper ($300), but lacks CUDA and Tensor Cores.

Within the Brand: Jetson AGX Orin

- AGX Orin is more powerful (275 TOPS), but costs more ($1999) and is larger.

7. Practical Tips

Power Supply and Peripherals

- Minimum 65 W (with overhead for peripherals).

- Use NVMe SSDs through M.2 adapters.

Software Compatibility

- OS: Linux Ubuntu 24.04 LTS with JetPack 6.0.

- Drivers: Regularly update via SDK Manager.

Caution with Adapters

HDMI 2.1 is supported only through DisplayPort adapters.

8. Pros and Cons

Pros:

- Best-in-class AI/TOPS performance per watt.

- Compact size (70×45 mm).

- Support for ROS 2 and Isaac Sim.

Cons:

- Price $699 (as of April 2025).

- Difficulties running x86 applications.

9. Final Conclusion: Who Is Orin NX For?

This module is designed for:

- AI Engineers developing autonomous robots or drones.

- Industrial Designers in need of a mobile workstation.

- Startups in computer vision (such as smart cameras).

If you're looking for a GPU for gaming or studio-level 3D rendering — this is not your choice. But for projects where compactness, energy efficiency, and AI acceleration are crucial, the Jetson Orin NX 16 GB is unrivaled.

Basic

Label Name

NVIDIA

Platform

Professional

Launch Date

February 2023

Model Name

Jetson Orin NX 16 GB

Generation

Tegra

Bus Interface

PCIe 4.0 x4

Transistors

Unknown

Tensor Cores

Tensor Cores are specialized processing units designed specifically for deep learning, providing higher training and inference performance compared to FP32 training. They enable rapid computations in areas such as computer vision, natural language processing, speech recognition, text-to-speech conversion, and personalized recommendations. The two most notable applications of Tensor Cores are DLSS (Deep Learning Super Sampling) and AI Denoiser for noise reduction.

TMUs

Texture Mapping Units (TMUs) serve as components of the GPU, which are capable of rotating, scaling, and distorting binary images, and then placing them as textures onto any plane of a given 3D model. This process is called texture mapping.

Foundry

Samsung

Process Size

8 nm

Architecture

Ampere

Memory Specifications

Memory Size

16GB

Memory Type

LPDDR5

Memory Bus

The memory bus width refers to the number of bits of data that the video memory can transfer within a single clock cycle. The larger the bus width, the greater the amount of data that can be transmitted instantaneously, making it one of the crucial parameters of video memory. The memory bandwidth is calculated as: Memory Bandwidth = Memory Frequency x Memory Bus Width / 8. Therefore, when the memory frequencies are similar, the memory bus width will determine the size of the memory bandwidth.

128bit

Memory Clock

1600MHz

Bandwidth

Memory bandwidth refers to the data transfer rate between the graphics chip and the video memory. It is measured in bytes per second, and the formula to calculate it is: memory bandwidth = working frequency × memory bus width / 8 bits.

102.4 GB/s

Theoretical Performance

Pixel Rate

Pixel fill rate refers to the number of pixels a graphics processing unit (GPU) can render per second, measured in MPixels/s (million pixels per second) or GPixels/s (billion pixels per second). It is the most commonly used metric to evaluate the pixel processing performance of a graphics card.

14.69 GPixel/s

Texture Rate

Texture fill rate refers to the number of texture map elements (texels) that a GPU can map to pixels in a single second.

29.38 GTexel/s

FP16 (half)

An important metric for measuring GPU performance is floating-point computing capability. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy.

3.760 TFLOPS

FP64 (double)

An important metric for measuring GPU performance is floating-point computing capability. Double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy, while single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.

940.0 GFLOPS

FP32 (float)

An important metric for measuring GPU performance is floating-point computing capability. Single-precision floating-point numbers (32-bit) are used for common multimedia and graphics processing tasks, while double-precision floating-point numbers (64-bit) are required for scientific computing that demands a wide numeric range and high accuracy. Half-precision floating-point numbers (16-bit) are used for applications like machine learning, where lower precision is acceptable.

1.918 TFLOPS

Miscellaneous

SM Count

Multiple Streaming Processors (SPs), along with other resources, form a Streaming Multiprocessor (SM), which is also referred to as a GPU's major core. These additional resources include components such as warp schedulers, registers, and shared memory. The SM can be considered the heart of the GPU, similar to a CPU core, with registers and shared memory being scarce resources within the SM.

Shading Units

The most fundamental processing unit is the Streaming Processor (SP), where specific instructions and tasks are executed. GPUs perform parallel computing, which means multiple SPs work simultaneously to process tasks.

1024

L1 Cache

128 KB (per SM)

L2 Cache

256KB

TDP

25W

Vulkan Version

Vulkan is a cross-platform graphics and compute API by Khronos Group, offering high performance and low CPU overhead. It lets developers control the GPU directly, reduces rendering overhead, and supports multi-threading and multi-core processors.

1.3

OpenCL Version

3.0

OpenGL

4.6

DirectX

12 Ultimate (12_2)

CUDA

8.6

Shader Model

6.7

ROPs

The Raster Operations Pipeline (ROPs) is primarily responsible for handling lighting and reflection calculations in games, as well as managing effects like anti-aliasing (AA), high resolution, smoke, and fire. The more demanding the anti-aliasing and lighting effects in a game, the higher the performance requirements for the ROPs; otherwise, it may result in a sharp drop in frame rate.

Benchmarks

FP32 (float)

Score

1.918 TFLOPS

Compared to Other GPU

FP32 (float) / TFLOPS

GeForce GTX 760 OEM Rebrand

2.005 +4.5%

GeForce GTX 1050 Ti Max Q

1.943 +1.3%

Jetson Orin NX 16 GB

1.918

Radeon R9 270 1024SP

1.856 -3.2%

Playstation 4 GPU

1.806 -5.8%