AMD Surges Ahead: Ryzen AI Outperforms Intel Core Ultra
In the realm of AI-enhanced PCs, AMD ventured into the arena earlier than Intel, launching last year the Ryzen 7040 series processors, equipped with the Ryzen AI unit, based on the Phoenix core. This year, the introduction of the Ryzen 8040 series processors, powered by the Hawk Point core, saw the AI unit officially renamed to NPU. These also utilize the XDNA architecture, with computational capabilities boosted from 10 TOPS to 16 TOPS. AMD has devoted considerable effort towards software optimization for client-side and local AI tasks. Recent official tests have demonstrated that AMD processors, featuring Ryzen AI, surpass Intel’s Core Ultra in new AI benchmarks.
AMD compared the 15W Ryzen 7 7840U with the 28W Core Ultra i7 155H, both notebooks were equipped with 16GB LPDDR5-6400 memory and updated to the latest drivers.
The tests, conducted locally with LLMs including LLama 2, Mistral AI, code llama, and RAG, highlighted the advantages of running AI models locally in terms of privacy, as well as potential savings on subscription fees. AMD is delving further into this domain, having released a guide on operating large language models locally, which details deploying one’s own AI chatbot on-premises.
Firstly, the Mistral Instruct 7B LLM saw the Ryzen 7 7840U completing AI tasks in just 61% of the time it took the Core Ultra i7 155H.
In the Llama 2 7B chat speed test, the Ryzen 7 7840U accomplished the task in merely 54 seconds.
The Ryzen 7 7840U exhibited a performance 14% superior to its counterpart in the Llama v2 Chat 7B (Q4 KM), and a 17% higher performance in the Mistral Instruct 7b (Q4 KM). It also showcased a 79% faster first token speed in LLama v2 Chat and a 41% quicker first token speed in Mistral Instruct 7B LLM.
It is noteworthy that the power consumption limit of the Ryzen 7 7840U is lower than that of the Core Ultra i7 155H, with a slightly slower NPU speed – the Ryzen 7 7840U’s NPU is at 10 TOPS compared to the Core Ultra i7 155H’s 11 TOPS. Therefore, a Ryzen 8040 processor, matching the 28W power consumption while boosting the NPU speed to 16 TOPS, would offer a more distinct advantage.
AMD’s earlier market entry with Ryzen AI allows for more extensive market feedback, facilitating optimization of AI performance in their next-generation CPUs. The upcoming generation, leveraging the Zen 5 core Strix Point series, will upgrade the NPU to the XDNA 2 architecture, enhancing computational power to better accommodate the evolving landscape of AI applications.
