Intel Ultra Notebook NPU Capability Experience: The AI ​​Trend Is Unstoppable

The future work goal may be to deploy an AI assistant that can not only dub, chat, and help you draw. However, whenever we use AI applications (such as AI cutout, AI background replacement), we seem to easily encounter the "bottleneck" problem of laptops: not only will it start to freeze, but it will also make noises, and even the battery life will decrease significantly. Shorten

Therefore, Intel announced its AI PC plan in the middle of this year and quickly launched the Meteor Lake architecture Core Ultra processor at the end of the year, which is regarded as the largest generation processor upgrade in history. The specific features are as follows:

1. Intel uses a 4-process process to improve performance and energy efficiency by more than 20%

New three-level core structure, equipped with AI computing unit NPU

Integrated graphics performance doubled

Among them, the most eye-catching one is the newly added "NPU" (Neural Network Processor)

We can compare it to the AI ​​assistant in the processor, which exists to improve the user’s AI application experience

At the press conference, Intel introduced the NPU of the Ultra processor, which is suitable for continuous artificial intelligence and sharing artificial intelligence load

It is conceivable that artificial intelligence computers with NPU will improve user experience in the following aspects:

1. Faster speech recognition and voice assistant response speed

2. Smoother face recognition and unlocking

3. Improve image recognition and processing efficiency

4. Provides more intelligent background beautification and video effects

Even allows large AI models to run smoothly

Recently, we received an ASUS Lingyao 142024 thin and light notebook, which is equipped with a Core Ultra 155H processor with Meteor Lake architecture. We immediately conducted AI performance and application testing

When using UL Procyon AI Inference Benchmark for inference performance benchmark testing, it was found that the best performance of Meteor Lake’s AI inference capabilities was Arc iGPU (core graphics card), followed by NPU, and the worst performance was CPU. This benchmark covers a variety of neural network models such as MobileNetV3, InceptionV4, YOLOV3, DeepLabV3, Real-ESRGAN and ResNet50

By testing the power consumption of the entire machine, it was found that the AI ​​computing performance of the NPU is much higher than that of the CPU and higher than that of the GPU (Intel has stated that the overall energy consumption ratio of the NPU can be increased by up to 8 times)

This shows that NPU is very suitable for mobile devices with extremely high energy efficiency requirements such as notebook computers. It can be used to process continuous low-computation artificial intelligence applications to reduce the power consumption of the entire machine and improve battery life

When our laptop is equipped with a neural processing unit (NPU), there may be two obvious feelings:

1. When using a program that can perform AI calculations, the battery life is not easy to "collapse" - the NPU's AI calculation energy efficiency is higher.

2. The smoothness of AI applications supporting AI operations has been improved - NPU can reduce the computing pressure on the CPU and GPU and give full play to their performance

However, due to confidentiality requirements during the product testing phase, I have not been able to successfully test the product in practical applications. Most of the applications currently on the market (including Team, Cutting, SD, LLM, etc.) do not yet support NPU hardware acceleration and still mainly rely on CPU and GPU computing

When using a virtual background in a Teams video conference, it still relies on CPU and GPU acceleration without enabling NPU

With the popularity of NPU processors, we can foresee that most AI applications will be quickly adapted, thereby further improving the overall energy efficiency of the machine. However, we can only say that "the future can be expected" at present, and we are still unable to determine the actual adaptation results

In summary:

NPU is in principle more suitable for completing low-precision calculations in parallel than general-purpose CPUs or GPUs. In addition, NPU has the characteristics of high bandwidth and high capacity in terms of storage, large data throughput and low response latency. Therefore, NPU plays a positive role in improving the efficiency of AI applications

NPU is like an ultra-efficient AI computing expert. It can not only make your laptop last longer, but also smoothly run a variety of cool AI applications, such as virtual backgrounds and real-time beauty. This is the most significant role that NPU plays in personal computer equipment

PC devices with NPU computing units will theoretically have more powerful artificial intelligence computing capabilities, be able to efficiently process complex artificial intelligence models, and achieve rapid calculation and feedback

At the Ultra conference, Intel stated that Ultra processors already have the ability to run large AI models with 20 billion parameters, and personally demonstrated AI applications such as local large language models and Vincentian graphs

Currently, the accuracy and complexity of results generated by local AI are not comparable to the larger number of parameters of cloud-based generative AI applications

Looking to the future, I believe that as more and more AI applications support NPU, the battery life and performance of notebooks will be significantly improved (provided that you often use such AI computing applications)

With the improvement of NPU unit computing power and the enhancement of PC device storage capacity, it will be possible to implement a powerful local AI large model on a personal computer in the future, with hundreds of billions of parameters, which can be comparable to cloud AI. Smart writing or image generation capabilities. This is not an unattainable goal

Ultimately, we will continue to pay attention to the implementation of NPU acceleration applications on the official website, and will further evaluate the detailed evaluation of Core Ultra processor products in terms of AI performance and power consumption performance. Please continue to pay attention to the latest content of PConline

The above is the detailed content of Intel Ultra Notebook NPU Capability Experience: The AI ​​Trend Is Unstoppable. For more information, please follow other related articles on the PHP Chinese website!