AMD launches Peano, an open-source LLVM compiler for Ryzen AI XDNA and XDNA2 NPUs

[harlan4096]

Peano is an open-source compiler for Ryzen AI

AMD is finally taking its NPUs seriously. 



The AMD Ryzen 7000 series, known internally as Phoenix, introduced the Xilinx AI accelerator unit (marketed as XDNA) to consumer series processors. Over a year have passed since its debut, and upgrades to the silicon with the Hawk Point version boasting an even faster NPU, it wasn’t until the recent announcement of the Ryzen AI 300 series that AMD revealed an open-source LLVM compiler specifically designed for this unit.

AMD Ryzen AI processors:

• Ryzen 7000 “Phoenix”: XDNA up to 10 TOPS
  
• Ryzen 8000 “Hawk Point: XDNA up to 16 TOPS
  
• Ryzen AI 300 “Strix Point”: XDNA2 up to 50 TOPS ?
  

The importance of built-in NPU units will continue to grow, with AMD, Intel, and soon Qualcomm all participating in a new race for the fastest AI acceleration on their devices natively. These new dedicated units for machine learning will be utilized by upcoming Windows technologies and serve as accelerators for third-party tools, such as locally run large language models (LLMs).
 

“On behalf of AMD, I’m pleased to announce the open sourcing of an LLVM backend for AMD/Xilinx AI Engine processors. These processors exist in a number of devices including RyzenAI SoCs. The repository currently focuses on supporting the AIE2 architecture implemented by the XDNA accelerators in “Phoenix” and “Hawk Point” devices.
…
Note that these accelerators include an array of processors, while the LLVM backend only supports a single processor. Support for devices as a whole is available in open source tools based on MLIR.
— Stephen Neuendorffer, AMD/Xilinx

The Peano project may kickstart the popularity of writing AMD NPU-native applications that can leverage these new processing units for large language models. AMD is actively showcasing how to use Ryzen and Radeon-powered systems with tools like LM Studio, a simple app with an intuitive GUI that can currently only utilize integrated and discrete GPUs, not NPUs. AMD provides several examples of how to use Ryzen AI NPUs for software development here, though this is not intended for end users.
...

Continue Reading