# AI software overview and architecture
Qualcomm^®^ Linux AI software allows you to optimally deploy pre-trained, deep learning models on Qualcomm hardware accelerators, such as Neural Processing Unit (NPU), Graphic Processing Unit (GPU), and
Central Processing Unit (CPU). Qualcomm Linux contains software development kits (SDKs), APIs, sample applications, development tools, and third-party frameworks support, such as GStreamer and TFLite, for application development.
Multimedia AI applications use the Qualcomm IM SDK GStreamer Plugin APIs to take advantage of Qualcomm AI Engine Direct
and Qualcomm Neural Processing Engine APIs that provide lower-level hardware acceleration for AI use cases, as shown in the following figure:
**Top-level AI hardware and software blocks of the Qualcomm Linux AI stack**
The key components of the Qualcomm Linux AI software are as follows:
- [AI applications:](https://docs.qualcomm.com/doc/80-70022-15/topic/sample-apps.html) Sample applications based on Gstreamer that you can [customize](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-15B/customize-reference-app.html).
- [GStreamer plugins](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-50/gst-plugin-architecture.html):
Qualcomm Linux software offers GStreamer-based, machine learning plugins for accelerating AI inference using tools such as, TFLite
and Qualcomm^®^ Neural Processing Engine SDK, along with GStreamer plugins for pre- and postprocessing.
- Qualcomm AI software consists of two SDKs to accelerate AI workloads.
The [Qualcomm Neural Processing Engine SDK](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-2/)
and [Qualcomm AI Engine Direct](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/) provide tools and
libraries to optimally accelerate AI models on multiple hardware accelerators.
- Qualcomm system-on-Chip (SoCs) offer three hardware cores for AI loads.
- [Neural Processing Unit (NPU):](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-15B/run-models.html)
The NPU, also referred to as Qualcomm® Hexagon™ Tensor Processor (HTP) or DSP/HMX, is suitable for running AI workloads
with low-power and high-performance.
To get optimized performance, quantize pre-trained models to one of the supported precisions.
- [Graphics Processing Unit (GPU):](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-15B/run-models.html)
Qualcomm® Adreno™ GPU is suitable for running AI workloads with medium-power and medium-performance.
AI workloads are accelerated with OpenCL kernels.
You can also use the GPU to accelerate model pre- and postprocessing.
- [Central Processing Unit (CPU):](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-15B/run-models.html)
Use AI inferencing on the CPU to benchmark model accuracy or performance against other hardware accelerators.
You can also use the CPU to run model pre- and postprocessing.
## Architecture
The following image shows available AI runtimes, such as ONNX runtime and LiteRT, the available hardware
accelerators, such as the neural Processing Unit (NPU), graphics Processing Unit (GPU), and central Processing Unit (CPU), and the backend libraries for each accelerator.
### AI hardware accelerators
You can accelerate AI workflows using the following hardware cores:
- Qualcomm^®^ Hexagon™ Tensor Processor (HTP): The HTP, also known as NPU, DSP, or HMX, is suitable to run AI workloads with low-power and high-performance.
To get optimized performance, quantize pre-trained models to one of the supported precisions.
- Qualcomm^®^ Adreno™ GPU: Suitable to run AI workloads with medium-power and medium-performance.
AI workloads are accelerated with OpenCL kernels.
You can also use the GPU to accelerate model pre- and postprocessing.
- Qualcomm^®^ Kryo™ CPU: Use AI inferencing on the CPU to benchmark model accuracy and performance against other hardware accelerators.
You can also use the CPU to run model pre- and postprocessing.
### AI software
Choose from the following software to deploy AI workloads.
You must convert pre-trained models into an executable format with the SDK before running them, except for TFLite models.
Pre-trained models (with the exception of LiteRT models) need to be converted to an executable format with the selected AI Stack SDK before running them.
The TFLite Delegate allows you to directly run LiteRT models.
- [Qualcomm Neural Processing Engine (SNPE)](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-2):
An all-in-one SDK that provides C, C++, and Java APIs to support heterogenous computing, system-level configurations, and direct AI workloads to all accelerator cores.
Provides you with flexibility, such as inter-core collaboration support.
- [Qualcomm AI Engine Direct (QNN)](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50):
Lower-level, highly customizable unified APIs that speed up AI models on all AI accelerator cores with individual libraries.
Use to directly target a specific accelerator core or delegate workloads from popular runtimes, including Qualcomm Neural Processing Engine SDK, LiteRT (formerly TensorFlow Lite), and ONNX runtime.
Low-level SDK provides more functionality and debugging abilities.
- [AI Model Efficiency Toolkit (AIMET)](https://quic.github.io/aimet-pages/releases/latest/index.html):
Open-source library to optimize (compressing and quantizing) trained neural network models.
This is a complex SDK designed to generate optimized quantized models.
## Next steps
- [AI Hub](https://aihub.qualcomm.com/get-started)
- [Qualcomm AI Model Efficiency Toolkit](https://quic.github.io/aimet-pages/releases/latest/index.html)
- [Qualcomm Neural Processing Engine](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-2/)
- [Qualcomm AI Engine Direct](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/)
- [AI Engine Direct: TFLite Delegate](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/tflite_delegate.html)
- [Qualcomm Intelligent Multimedia SDK](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-50/example-applications.html)
Last Published: Sep 15, 2025
[Previous Topic
AI/ML documentation](https://docs.qualcomm.com/bundle/publicresource/80-70022-15/topics/home.md) [Next Topic
GStreamer ML plugin and AI SDK APIs](https://docs.qualcomm.com/bundle/publicresource/80-70022-15/topics/interfaces.md)