GPUNet: Optimized for Qualcomm Devices

GPUNet is a machine learning model that can classify images from the Imagenet dataset. It can also be used as a backbone in building more complex models for specific use cases.

This is based on the implementation of GPUNet found here. This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the Qualcomm® AI Hub Models library to export with custom configurations. More details on model performance across various devices, can be found here.

Qualcomm AI Hub Models uses Qualcomm AI Hub Workbench to compile, profile, and evaluate this model. Sign up to run these models on a hosted Qualcomm® device.

Getting Started

There are two ways to deploy this model on your device:

Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

Runtime	Precision	Chipset	SDK Versions	Download
ONNX	float	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a16	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
ONNX	w8a8	Universal	QAIRT 2.42, ONNX Runtime 1.24.3	Download
QNN_DLC	float	Universal	QAIRT 2.45	Download
QNN_DLC	w8a16	Universal	QAIRT 2.45	Download
QNN_DLC	w8a8	Universal	QAIRT 2.45	Download
TFLITE	float	Universal	QAIRT 2.45	Download
TFLITE	w8a8	Universal	QAIRT 2.45	Download

For more device-specific assets and performance metrics, visit GPUNet on Qualcomm® AI Hub.

Option 2: Export with Custom Configurations

Use the Qualcomm® AI Hub Models Python library to compile and export the model with your own:

• Custom weights (e.g., fine-tuned checkpoints)
• Custom input shapes
• Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for GPUNet on GitHub for usage instructions.

Model Details

Model Type: Model_use_case.image_classification

Model Stats:

• Model checkpoint: Imagenet
• Input resolution: 224x224
• Number of parameters: 10.49M
• Model size (float): 45.28MB
• Model size (w8a8): 21.3MB

Performance Summary

Model	Runtime	Precision	Chipset	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit
GPUNet	ONNX	float	Snapdragon® 8 Elite Gen 5 Mobile	0.507 ms	0 - 36 MB	NPU
GPUNet	ONNX	float	Snapdragon® X2 Elite	0.463 ms	24 - 24 MB	NPU
GPUNet	ONNX	float	Snapdragon® X Elite	1.121 ms	24 - 24 MB	NPU
GPUNet	ONNX	float	Snapdragon® 8 Gen 3 Mobile	0.782 ms	0 - 59 MB	NPU
GPUNet	ONNX	float	Qualcomm® QCS8550 (Proxy)	1.043 ms	0 - 2 MB	NPU
GPUNet	ONNX	float	Qualcomm® QCS9075	1.357 ms	1 - 3 MB	NPU
GPUNet	ONNX	float	Snapdragon® 8 Elite For Galaxy Mobile	0.6 ms	0 - 30 MB	NPU
GPUNet	ONNX	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	0.359 ms	0 - 49 MB	NPU
GPUNet	ONNX	w8a16	Snapdragon® X2 Elite	0.385 ms	12 - 12 MB	NPU
GPUNet	ONNX	w8a16	Snapdragon® X Elite	0.988 ms	12 - 12 MB	NPU
GPUNet	ONNX	w8a16	Snapdragon® 8 Gen 3 Mobile	0.58 ms	0 - 64 MB	NPU
GPUNet	ONNX	w8a16	Qualcomm® QCS6490	98.781 ms	22 - 31 MB	CPU
GPUNet	ONNX	w8a16	Qualcomm® QCS8550 (Proxy)	0.826 ms	0 - 2 MB	NPU
GPUNet	ONNX	w8a16	Qualcomm® QCS9075	0.985 ms	0 - 3 MB	NPU
GPUNet	ONNX	w8a16	Qualcomm® QCM6690	49.699 ms	30 - 38 MB	CPU
GPUNet	ONNX	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.431 ms	0 - 44 MB	NPU
GPUNet	ONNX	w8a16	Snapdragon® 7 Gen 4 Mobile	38.163 ms	30 - 38 MB	CPU
GPUNet	ONNX	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.351 ms	0 - 48 MB	NPU
GPUNet	ONNX	w8a8	Snapdragon® X2 Elite	0.311 ms	12 - 12 MB	NPU
GPUNet	ONNX	w8a8	Snapdragon® X Elite	0.744 ms	12 - 12 MB	NPU
GPUNet	ONNX	w8a8	Snapdragon® 8 Gen 3 Mobile	0.47 ms	0 - 62 MB	NPU
GPUNet	ONNX	w8a8	Qualcomm® QCS6490	16.703 ms	1 - 13 MB	CPU
GPUNet	ONNX	w8a8	Qualcomm® QCS8550 (Proxy)	0.618 ms	0 - 14 MB	NPU
GPUNet	ONNX	w8a8	Qualcomm® QCS9075	0.726 ms	0 - 3 MB	NPU
GPUNet	ONNX	w8a8	Qualcomm® QCM6690	10.166 ms	6 - 14 MB	CPU
GPUNet	ONNX	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.394 ms	0 - 38 MB	NPU
GPUNet	ONNX	w8a8	Snapdragon® 7 Gen 4 Mobile	7.657 ms	6 - 15 MB	CPU
GPUNet	QNN_DLC	float	Snapdragon® 8 Elite Gen 5 Mobile	0.565 ms	1 - 35 MB	NPU
GPUNet	QNN_DLC	float	Snapdragon® X2 Elite	0.652 ms	1 - 1 MB	NPU
GPUNet	QNN_DLC	float	Snapdragon® X Elite	1.39 ms	1 - 1 MB	NPU
GPUNet	QNN_DLC	float	Snapdragon® 8 Gen 3 Mobile	0.906 ms	0 - 56 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® QCS8275 (Proxy)	4.686 ms	1 - 30 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® QCS8550 (Proxy)	1.274 ms	0 - 14 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® SA8775P	1.7 ms	1 - 33 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® QCS9075	1.571 ms	1 - 3 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® QCS8450 (Proxy)	2.44 ms	0 - 61 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® SA7255P	4.686 ms	1 - 30 MB	NPU
GPUNet	QNN_DLC	float	Qualcomm® SA8295P	2.265 ms	0 - 32 MB	NPU
GPUNet	QNN_DLC	float	Snapdragon® 8 Elite For Galaxy Mobile	0.702 ms	1 - 35 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® 8 Elite Gen 5 Mobile	0.429 ms	0 - 45 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® X2 Elite	0.566 ms	0 - 0 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® X Elite	1.225 ms	0 - 0 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® 8 Gen 3 Mobile	0.763 ms	0 - 58 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCS6490	3.226 ms	0 - 2 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCS8275 (Proxy)	2.482 ms	0 - 41 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCS8550 (Proxy)	1.072 ms	0 - 2 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® SA8775P	1.286 ms	0 - 43 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCS9075	1.233 ms	2 - 4 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCM6690	6.51 ms	0 - 163 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® QCS8450 (Proxy)	1.459 ms	0 - 61 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® SA7255P	2.482 ms	0 - 41 MB	NPU
GPUNet	QNN_DLC	w8a16	Qualcomm® SA8295P	1.666 ms	0 - 39 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® 8 Elite For Galaxy Mobile	0.522 ms	0 - 41 MB	NPU
GPUNet	QNN_DLC	w8a16	Snapdragon® 7 Gen 4 Mobile	1.288 ms	0 - 42 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.287 ms	0 - 44 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® X2 Elite	0.36 ms	0 - 0 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® X Elite	0.738 ms	0 - 0 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® 8 Gen 3 Mobile	0.452 ms	0 - 55 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCS6490	1.928 ms	2 - 4 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCS8275 (Proxy)	1.409 ms	0 - 39 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCS8550 (Proxy)	0.602 ms	0 - 2 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® SA8775P	0.797 ms	0 - 41 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCS9075	0.691 ms	0 - 2 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCM6690	3.502 ms	0 - 43 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® QCS8450 (Proxy)	0.856 ms	0 - 58 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® SA7255P	1.409 ms	0 - 39 MB	NPU
GPUNet	QNN_DLC	w8a8	Qualcomm® SA8295P	1.112 ms	0 - 37 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.344 ms	0 - 42 MB	NPU
GPUNet	QNN_DLC	w8a8	Snapdragon® 7 Gen 4 Mobile	0.789 ms	0 - 39 MB	NPU
GPUNet	TFLITE	float	Snapdragon® 8 Elite Gen 5 Mobile	0.558 ms	0 - 47 MB	NPU
GPUNet	TFLITE	float	Snapdragon® 8 Gen 3 Mobile	0.904 ms	0 - 74 MB	NPU
GPUNet	TFLITE	float	Qualcomm® QCS8275 (Proxy)	4.708 ms	0 - 44 MB	NPU
GPUNet	TFLITE	float	Qualcomm® QCS8550 (Proxy)	1.266 ms	0 - 2 MB	NPU
GPUNet	TFLITE	float	Qualcomm® SA8775P	1.757 ms	0 - 45 MB	NPU
GPUNet	TFLITE	float	Qualcomm® QCS9075	1.579 ms	0 - 27 MB	NPU
GPUNet	TFLITE	float	Qualcomm® QCS8450 (Proxy)	2.402 ms	0 - 79 MB	NPU
GPUNet	TFLITE	float	Qualcomm® SA7255P	4.708 ms	0 - 44 MB	NPU
GPUNet	TFLITE	float	Qualcomm® SA8295P	2.278 ms	0 - 44 MB	NPU
GPUNet	TFLITE	float	Snapdragon® 8 Elite For Galaxy Mobile	0.71 ms	0 - 47 MB	NPU
GPUNet	TFLITE	w8a8	Snapdragon® 8 Elite Gen 5 Mobile	0.239 ms	0 - 45 MB	NPU
GPUNet	TFLITE	w8a8	Snapdragon® 8 Gen 3 Mobile	0.33 ms	0 - 56 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCS6490	1.599 ms	0 - 15 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCS8275 (Proxy)	1.121 ms	0 - 39 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCS8550 (Proxy)	0.429 ms	0 - 10 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® SA8775P	0.637 ms	0 - 42 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCS9075	0.525 ms	0 - 14 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCM6690	3.033 ms	0 - 40 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® QCS8450 (Proxy)	0.684 ms	0 - 58 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® SA7255P	1.121 ms	0 - 39 MB	NPU
GPUNet	TFLITE	w8a8	Qualcomm® SA8295P	0.903 ms	0 - 37 MB	NPU
GPUNet	TFLITE	w8a8	Snapdragon® 8 Elite For Galaxy Mobile	0.274 ms	0 - 38 MB	NPU
GPUNet	TFLITE	w8a8	Snapdragon® 7 Gen 4 Mobile	0.617 ms	0 - 38 MB	NPU

License

• The license for the original implementation of GPUNet can be found here.

References

• GPUNet: Searching the Deployable Convolution Neural Networks for GPUs
• Source Model Implementation

Community

• Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
• For questions or feedback please reach out to us.