Recently, AMD unveiled its latest Strix Point APU series, emphasizing their exceptional performance in AI large language model (LLM) applications, significantly outperforming Intel's Lunar Lake series processors. With the increasing demand for AI workloads, the competition among hardware manufacturers has become more intense. To address this market, AMD has introduced AI processors designed for mobile platforms, aiming for higher performance and lower latency.

AMD states that the Ryzen AI300 processors in the ix Point series can significantly increase the number of tokens processed per second when handling AI LLM tasks, achieving a 27% performance improvement over Intel's Core Ultra 258V. Although the Core Ultra7V is not the fastest model in the Lunar Lake series, its core and thread count is close to that of the higher-end Lunar Lake processors, demonstrating AMD's competitive edge in this field.

image.png

AMD's LM Studio tool is a consumer-oriented application based on the llama.cpp framework, designed to simplify the use of large language models. This framework optimizes x86 CPU performance, and while it does not require a GPU to run LLMs, using a GPU can further accelerate processing speeds. Tests have shown that the Ryzen AI9HX375 can achieve 35 times lower latency with the Meta Llama3.21b Instruct model, processing up to 50.7 tokens per second, compared to only 39.9 tokens per second for the Core Ultra7258V.

Furthermore, the Strix Point APUs come equipped with a powerful Radeon integrated graphics based on the RDNA3.5 architecture, which offloads tasks to the iGPU via the ulkan API, thereby enhancing LLM performance. Utilizing the Variable Graphics Memory (VGM) technology, the Ryzen AI300 processors can optimize memory allocation, improve energy efficiency, and ultimately achieve a 60% performance boost.

In comparative tests on the Intel AI Playground platform with identical settings, AMD found that the Ryzen AI9HX375 was 87% faster than the Core Ultra7258V on Microsoft Phi3.1 and 13% faster on the Mistral7b Instruct0.3 model. Despite this, comparing it to the flagship Core Ultra9288V in the Lunar Lake series would yield even more intriguing results. Currently, AMD is focused on making the use of large language models more accessible through LM Studio, aiming to enable more non-technical users to easily get started.

Key Points:

🌟 AMD Strix Point APUs offer a 27% performance improvement in AI LLM applications over Intel's Lunar Lake.

⚡ The Ryzen AI9HX375 exhibits 3.5 times lower latency in the Meta Llama3.2 model.

🚀 The LM Studio tool is designed to simplify the use of large language models, making it suitable for non-technical users.