XDNA is the name for AMD's neural processing unit microarchitecture. It is based on IP blocks from Xilinx, a company which was acquired by AMD in 2023.[1]
| Design firm | AMD |
|---|---|
| Introduced | April 2023 |
| Type | Neural processing unit microarchitecture |
As of 2024, XDNA is implemented in AMD's consumer PC processors (branded as Ryzen AI), as well as the AMD Alveo V70 AI accelerator.
Generations
editThis section needs expansion with: Architectural details, from the source(s). You can help by adding to it. (August 2024) |
First
editFirst-generation XDNA, initially implemented in the Ryzen 7040 series mobile processors, provides up to 10 TOPS of processing performance. The Ryzen 8040 series (codenamed "Hawk Point"), a refresh of the Ryzen 7040 series, features a higher-clocked XDNA NPU providing 16 TOPS of performance.[2]
XDNA is also used in AMD's Alveo V70 datacenter AI inference processing card.[3]
Second
editXDNA 2 was introduced in the Strix Point Ryzen AI 300 series processors. The NPU provides up to 50 TOPS of processing power,[4][5] and the implementation is labelled by AMD as the third generation of Ryzen AI.
See also
edit- Neural Engine, Apple's neural processing unit microarchitecture
- Tensor Processing Unit, Google's AI acceleration microarchitecture
References
edit- ^ Hachman, Mark (May 15, 2023). "Why AMD thinks Ryzen AI will be just as vital as CPUs and GPUs". PCWorld. Retrieved August 24, 2024.
- ^ Svitlyk, Yuri (March 7, 2024). "What is AMD XDNA? Architecture that launches AI on Ryzen processors". Root Nation. Retrieved August 24, 2024.
- ^ "Alveo V70 AI Accelerator". AMD. Retrieved August 24, 2024.
- ^ Alcorn, Paul (July 15, 2024). "AMD deep-dives Zen 5 architecture — Ryzen 9000 and AI 300 benchmarks, RDNA 3.5 GPU, XDNA 2, and more". Tom's Hardware. Retrieved August 24, 2024.
- ^ Bonshor, Gavin (July 15, 2024). "The AMD Zen 5 Microarchitecture: Powering Ryzen AI 300 Series For Mobile and Ryzen 9000 for Desktop". www.anandtech.com. Retrieved August 24, 2024.