AMD has been on a software / container / documentation spree recently around ROCm and today are announcing another software addition to the ROCm portfolio: AITER.
AITER is the “AI Tensor Engine for ROCm” as a collection of high performance AI operators to accelerate different AI workloads. AMD describes AITER as:
“AITER serves as a unified platform where customers can easily find and integrate optimized operators into their existing frameworks—be it private, public, or custom-built…With AITER, AMD simplifies the complexity of optimization, enabling users to maximize performance while providing flexibility to meet diverse AI requirements.”
AITER offers both C++ and Python interfaces, is designed to integrate into various workloads and systems, and is all nicely integrated with the AMD ROCm compute stack.
AMD in announcing AITER is showing off some rather significant gains from DeepSeek-V3 leveraging AITER:
Those wanting to learn more about AITER can do so via the AMD ROCm blog. AITER is available under an MIT license on GitHub.
