Intel AIGC experience: Core thin and light notebooks can also play large language models! Arc graphics card is extremely efficient
1. Foreword: The AI field should not forget the existence of Intel
The most popular industry this year is undoubtedly AIGC, which is AI-led generative content, including ChatGPT, AI drawing, AI poetry and other functions. It can be said that graphics cards that can only play games but do not support AIGC no longer meet the needs of the times.
Now that NVIDIA is soaring in the AI era, many students may mistakenly believe that only NVIDIA graphics cards support AI applications such as Stable Diffusion.
But actually,There is another manufacturer whose investment in AI is not inferior to NVIDIA, and has begun to deploy AI many years ago. It is the chip giant Intel.
In the upcoming 14th generation Core Meteor Lake processor, the VPU (Visual Processing Unit) AI unit is designed, which will be the first time that Intel has put an independent AI hardware unit in the CPU. The advantage of being integrated inside the CPU is that it can have extremely low latency, which is better than GPU when processing some low-load AI.
What is even more unimaginable is that the VPU can also provide hardware acceleration for Stable Diffusion without the need for an expensive discrete graphics card. This also means that the upcoming 14th-generation Core thin and light notebook will embrace AI in a more complete manner.
In order to allow more Intel users to experience AI, Intel also launchedAI Application Demo “Intel Large Language Model”it can help with some daily life as well as some tasks at work.
This demo is specially optimized for the 12th/13th generation Core processors with mixed architecture, that is, the mainstream Intel processors on the market can run directly with thin and light ones, and will have a very good experience.
Not only the CPU, Intel has also made a lot of efforts in the GPU field. at the beginning of the year,Intel launched the OpenVINO-based, AI drawing open source model Stable? Diffusion, which can use the open source image editing software GIMP, so that Intel Arc GPU can realize creative drawing by inputting keywords.
Soon after, Intel introducedSolution based on OpenVINO PyTorch backendso that Arc A770/A750 GPU can directly support Stable Diffusion WebUI, and its acceleration capability is much stronger than that of NVIDIA GPU at the same price.
In the following, we will conduct a comparative test, so that everyone can understand the AI performance of Intel Arc GPU more intuitively.
2. Intel’s large language model experience: 13th generation Core Duo thin and light notebooks are easy to handle
Here we chose a thin and light notebook certified by Intel Evo – ThinkPad X1 Yoga for testing, which is equipped with an Intel Core i5-1340P processor.
Intel’s large language model application has eight modules including “chat assistant”, “sentiment analysis”, “Chinese translation”, “story creation”, “outline generation”, “information extraction”, “food guide”, and “tourism planning”, and And the ability to write code.
In “Chat Assistant”, we tried to let AI write a heptadic poem about the Anshi Rebellion. Although the result was not good, there are still some verses that are eye-catching and can be used as a reference for creation.
“Generate outline” function, you only need to say what kind of work you want to write, and AI will give you an outline of the story.
“Chinese translation” can translate a paragraph of Chinese, and your question, AI is also answered in English.
For other functional modules, interested students can explore and experience by themselves.
3. Stable Diffusion drawing test: Arc A770 16GB is significantly ahead of RTX 4060
At present, most of the Stable Diffusion WebUI on the Windows side use the “Huishi” launcher, but it is a pity that this launcher currently only supports NVIDIA GPUs, AMD GPUs cannot be recognized at all, and Intel Arc GPUs can be recognized, but hardware acceleration cannot be enabled.
Therefore, if you want to use an Intel graphics card to accelerate Stable Diffusion, you need to run it in a Linux environment.
Of course, it does not mean that a Linux system must be installed.An easier path is to use Docker (containers).
Our test this time is based on the Docker version of Stable Diffusion WebUI. The graphics cards involved in the test include 4 dessert graphics cards with a price of 2,000 yuan from A/I/N.
It can be seen that the Arc A770 16GB can indeed support hardware acceleration, the GPU utilization rate is 99%, the memory usage exceeds 8GB, and the power consumption is also full of 190W TGP.
In order to control the video memory usage below 8GB so that other graphics cards can also be tested smoothly, we set the output resolution to 640*480.
The test results are somewhat surprising.
From the perspective of game performance, RTX 4060 is much stronger than the two Arc GPUs, but Stable Diffusion depends more on video memory. In this regard, RTX 4060 is quite powerful due to being castrated, especially since the video memory interface is only 128Bit, and the capacity is also low. Only 8GB, because it is not particularly suitable for accelerating Stable Diffusion.
On AMD’s side, although RX 6700 XT has 12GB large-capacity video memory and 192Bit bit width, due to AMD’s current lack of AI support, the performance of its GPUs in this area is not satisfactory.
The performance of the two Intel GPUs exceeded expectations, even the Arc A750 with a lower positioning is close to the RTX 4060.The performance of Arc A770 has greatly surpassed its opponents, leading by as much as 20%.
In addition, we limit the output resolution. If you use a higher resolution, or install some other plug-ins for AI image restoration, 8GB of video memory is far from enough, and the advantage of Arc A770 16GB will be even greater.
4. Summary: AI drawing may wish to consider 2000 yuan Intel Arc A770 16GB
Before, when a classmate asked me to recommend a graphics card that can run AI, I would recommend the corresponding N card according to the budget, but now, I will also consider the I card.
As we all know, Stable Diffusion has very strict requirements on memory capacity and speed, and there are many restrictions on memory capacity of 8GB and below, such as using high-resolution graphics.
Therefore, if you want to play AI drawing seriously, 16GB or 24GB video memory will bring more gameplay and better experience, which has also led to the hot sale of second-hand RTX 3090 24GB graphics cards.
Of course, even for second-hand RTX 3090, the price is not affordable for ordinary consumers.
Looking at the current GPU market, Intel Arc A770 16GB can be said to be a graphics card specially designed for AI. At a price lower than RTX 4060, this graphics card provides twice the memory capacity and memory interface width of the latter, which can treat the stable diffusion memory anxiety of AI players to a certain extent.
Although the Arc A770 16GB game performance is at the same level as the RTX 3060, the acceleration performance of its AI drawing is more than 20% stronger than that of the RTX 4060.
In contrast, NVIDIA has relatively few graphics card models with 16GB of video memory and the price is high. For example, the price of RTX 4080 is as high as 9,000 yuan, which makes many AI players sigh. Even the RTX 4060 16GB, which was launched not long ago, is priced at around 4,000 yuan, almost twice that of the Arc A770 16GB.
Therefore, for AI players with insufficient budget, Intel Arc A770 16GB is indeed better than most graphics cards in terms of cost performance.
As for AMD, its AI road is still very long!

On the other hand, AI applications often require powerful computing power, and performance is not the advantage of thin and light notebooks, so few people will associate thin and light notebooks with AI.
Intel let us see a different side,On the Evo-certified thin and light notebooks equipped with the 13th generation Core P series, Intel’s large language model can run very smoothly. Increase work efficiency.
Whether it is daily office work or business trips, taking out the portable Evo thin and light notebook can complete some light-load AI reasoning work, which was unimaginable in the past.
[End of this article]If you need to reprint, please be sure to indicate the source: Kuai Technology
Responsible Editor: Liu Yun






