At Computex 2024, AMD introduced its latest generation of Ryzen laptop processors designed for generative AI workloads: the Ryzen AI 300 series. This series represents a rebranding of AMD’s top-tier Ryzen 9 chips, retaining the HX suffix to denote “top of stack” performance rather than indicating power consumption. The Ryzen AI 300 series is AMD’s answer to Qualcomm’s ARM-based Snapdragon X Elite and Intel’s x86 Lunar Lake chips.
AMD Ryzen AI 300 series — The technical lowdown
The new Ryzen AI chips leverage AMD’s cutting-edge architectures for neural, integrated graphics, and general processing: XDNA2 for the neural processing unit (NPU), RDNA 3.5 for the integrated GPU (iGPU) with up to 16 compute units, and Zen 5 for the CPU. AMD is kicking things off with the Ryzen AI 9 HX 370 and Ryzen AI 9 365. Both feature a 50 tera operations per second (TOPS) NPU, with the HX variant positioned as the higher-end model.
The Ryzen AI 9 HX 370 boasts a 12-core/24-thread configuration, a maximum clock speed of 5.1GHz, 36MB cache, and Radeon 890M graphics. In comparison, the Ryzen AI 9 365 features a 10-core/20-thread setup, a 5.0GHz max clock speed, 34MB cache, and Radeon 880M graphics.
Notably, the Ryzen AI 300-series stands out in the market with the highest TOPS among NPU-configured chips. Qualcomm’s Snapdragon X series NPU can pull 45 TOPS, Apple’s M4 38 TOPS, while the upcoming Lunar Lake series’ NPU can clock over 40 TOPS, Intel claims. AMD’s previous Ryzen 8040 series was only 16 TOPs so that gives a sense of how big of an update the Ryzen AI 300-series is, at least on paper. Having said that, although TOPS is a notable performance metric, it is not the sole indicator of a system’s ability to handle AI tasks. Various factors influence AI performance, including architecture efficiency, software optimisation, and workload characteristics.
But moving on, AMD claims that the XDNA2 NPU architecture delivers five times more compute capacity and twice the power efficiency compared to the previous generation. This improvement is attributed to its unique FP16 “block” architecture, capable of handling both 8-bit performance (INT8) and 16-bit accuracy (FP16) generative AI workloads without requiring quantisation. Quantisation, used by competitors like Qualcomm, can increase power efficiency but often reduces AI model accuracy. AMD’s new architecture aims to process AI workloads quickly and accurately without this trade-off.
AMD Ryzen AI 300 series — Market availability
The Ryzen AI 300 series is ready for prime time and will start shipping from July 2024. It will power some of the same Copilot Plus PCs that were announced at Microsoft’s recent Surface event including the Asus Vivobook S15 and HP OmniBook.
Follow FE Tech Bytes on Twitter, Instagram, LinkedIn, Facebook.
