Last January, AMD launched its Ryzen 7040 chips, the first x86 processors with an integrated neural processing unit (NPU) for AI workloads, beating out Intel. Intel’s first chips to include an NPU, the Core Ultra “Meteor Lake” chips, are set to arrive on December 14th after significant delays. Today, AMD is announcing the Ryzen 8040 series chips, its next batch of AI-equipped laptop hardware, and sharing its future AI roadmap.
The Ryzen 8040 chips, led by the 8-core Ryzen 9 8945HS, are reportedly up to 1.4 times faster than their predecessors for Llama 2 and AI vision model performance. They are also up to 1.8 times faster than Intel’s high-end 13900H chip for gaming and 1.4 times faster for content creation. The real test will be comparing them to Intel’s new Core Ultra chips, which AMD wasn’t able to benchmark.
AMD’s NPU will be available on all of the Ryzen 8040 chips except for two low-end models, the six-core Ryzen 5 8540U and quad-core Ryzen 3 8440U. Their Ryzen 7040’s NPU, AMD XDNA, can reach 10 TOPS, while the 8040’s NPU can hit 16 TOPS. Looking into 2024, AMD teased its next NPU architecture, codenamed “Strix Point,” which will offer “more than 3x generative AI NPU performance.”
It’s important to note that both AMD and Intel are behind Qualcomm in bringing NPUs to Windows PCs, as its SQ3 powered the ill-fated Surface Pro 9 5G. Ultimately, the far more compelling competitor for Intel and AMD is Apple, which has integrated Neural Engines in its hardware since 2017.
Like Intel, AMD is also urging developers to incorporate NPU features into their apps. Today, it’s unveiling the Ryzen AI Software platform, which will allow developers to optimize pre-trained AI models to run on Ryzen AI hardware. AMD’s platform will also assist those models in running on Intel’s NPUs, similar to how Intel’s AI development tools will ultimately help Ryzen systems.
