| guide | Generative AI (AIGC) is undoubtedly the hottest topic and application at the moment, and various software and hardware manufacturers are fully committed to it. |
As an industry leader with powerful hardware and extensive ecology, Intel is also vigorously promoting the implementation of PC AIGC applications, providing comprehensive support from hardware to software.
Intel runs into the era of AI PC! The 13th generation Core supports large models with 16 billion parameters, and the Arc graphics card speeds up by 54%
Currently on the PC platform, Intel's 12th/13th generation Core, Arc A series graphics cards, and the upcoming next-generation Core can meet AIGC's demand for high computing power.
Coupled with the construction of software ecology and the optimization of models, AIGC scenarios can be applied on desktops, thin and light notebooks, all-around notebooks, game notebooks and other devices, thus greatly improving the efficiency of popular daily life and work.
Among them, based on the OpenVINO PyTorch backend solution, through the Pytorch API, the community open source model can run well on Intel client processors, integrated graphics cards, discrete graphics cards, and dedicated AI engines.
For example, the Automatic1111 WebUI implemented by Stable Diffusion has been able to run FP16 floating-point precision models on Intel processors and graphics cards through the above-mentioned methods, and supports functions such as Vincent graph, graph-generated graph, and partial repair.
Intel runs into the era of AI PC! The 13th generation Core supports large models with 16 billion parameters, and the Arc graphics card speeds up by 54%
Furthermore, through model optimization, Intel reduces the model's demand for hardware resources, thereby increasing the model's inference speed, allowing the community's open source model to run well on PCs.
Right now, Intel’s 13th-generation Core, through XPU acceleration, low-bit quantization, and other software-level optimizations, can allow a large language model with up to 16 billion parameters to run on a personal computer with a memory capacity of 16GB or above through the BigDL-LLM framework .
Models that have been verified include but are not limited to: LLAMA/LLAMA2, ChatGLM/ChatGLM2, MPT, Falcon, MOSS, Baichuan, QWen, Dolly, RedPajama, StarCoder, Whisper, etc., and support Windows and Linux operations through Transformers and LangChain API interfaces system.
Core Ultra will integrate an independent AI acceleration unit called "VPU" through a separate module architecture to provide PC users with AI-driven new functions and new applications, such as automatic reframing in Adobe Premiere Pro, scene editing detection, etc. , and more efficient machine learning acceleration.
Currently, Intel is cooperating with PC industry partners to accelerate the implementation and popularization of AI acceleration applications, and future follow-up baseband products will further expand AI capabilities.
Intel runs into the era of AI PC! The 13th generation Core supports large models with 16 billion parameters, and the Arc graphics card speeds up by 54%
On the other hand, Intel Arc graphics card can already support Stable Diffusion better through driver and patch updates, and the performance improvement effect is quite significant.
Tom's Hardware tested Arc A770 16GB and Arc A750 respectively, using Automatic1111 WebUI OpenVINO.
Before that, the performance of Arc A770 16GB and Arc A450 lagged behind RTX 4060 by about 20%, but now they have increased by 54% and 40% respectively, and are 25% and 6% ahead of RTX 4060 respectively.
You know, the game performance of the two is actually far inferior to the RTX 4060, but now the AI efficiency is higher, which shows that Intel's hardware performance and software optimization are quite impressive.
Looking at the performance of AMD graphics cards, it highlights Intel's progress even more. The efficiency of the RX 6000 series is still ridiculously low, and the RX 6800 is only a little more than half of the Arc A750.