Computex 2024: AMD’s next generation of processors for AI laptops

Starting with the Ryzen AI 9 HX 370 and Ryzen AI 9 365.

At Computex 2024, AMD announced a new generation of Ryzen notebook processors for generative AI workloads: the Ryzen AI 300 Series. This is a rebranding of the top-of-the-line Ryzen 9 chips. The new name still includes the HX suffix that AMD introduced in 2022, but it doesn’t indicate how many watts the chip consumes. Instead, HX will stand for «top of the stack» or the best and fastest Ryzen AI 300 chip. The new Ryzen AI chips are built on AMD’s latest architectures for neurons, integrated graphics and general processing: XDNA2 for the NPU, RDNA 3.5 for the iGPU, which now has up to 16 compute units, and Zen 5 for the CPU. The first two processors in the series are — Ryzen AI 9 HX 370 and Ryzen AI 9 365. Both have 50 TOPS NPUs, but the HX variant is the more expensive of the two.

The HX variant is the more expensive of the two.

The Ryzen AI 9 HX 370 — is a 12-core/24-thread chip with a maximum clock speed of 5.1GHz, 36MB of cache and Radeon 890M graphics. The Ryzen AI 9 365 — is a 10-core/12-threaded chip with a maximum clock speed of 5.0GHz, 34MB of cache and Radeon 880M graphics. The Ryzen AI 9 300 series also has the highest TOPS compared to other NPU chips already on the market or coming soon. Qualcomm’s Snapdragon X series has 45 TOPS, Apple’s M4 — 38, and AMD’s latest-generation Ryzen 8040 chips — just 16. Intel’s Meteor Lake Ultra 7 165H has about 10 TOPS (but the upcoming Lunar Lake will supposedly level the playing field).

An Intel Meteor Lake Ultra 7 165H has about 10 TOPS.

AMD claims its XDNA2 NPU has five times the processing power and twice the energy efficiency of last-generation chips due to the unique architecture of the «FP16 block» or NPU, which can handle both 8-bit (INT8) and 16-bit (FP16) generative AI workloads without quantization. Quantization — a common method of making AI models more energy efficient (something Qualcomm uses in its smartphone chips) or converting input values from a large data set to output values in a smaller data set. Unfortunately, this can also reduce the accuracy of AI models. But AMD’s new chips will presumably be able to perform both types of math equations without conversion, so AI workloads will be processed both quickly and accurately.

Starting in July 2024, the Ryzen AI 300 will appear in some of the Copilot Plus PCs (laptops) announced at Microsoft’s recent Surface event: the Asus Vivobook S 15 and HP OmniBook. Other laptops with artificial intelligence include — MSI’s Stealth A16, Summit A16, Prestige A16 and Creator A16; Asus’ Zenbook S 16, Vivobook S 14 and 16 and ProArt P16/X13; and Asus’ Rog Zephyrus G16 and Tuf A14/A16 gaming laptops; and Lenovo’s other ThinkBook, ThinkPad and Yoga laptops.

An AI 300 will also be available in some PCs (laptops) with artificial intelligence.