AMD ZenDNN 5.0 was rolled out this morning as the newest version of this deep neural network library that is compatible with Intel's oneDNN APIs and infrastructure. ZenDNN 5.0 is now optimized for AMD Zen 5 processors such as the [1]EPYC 9005 series. ZenDNN 5.0 also ships performance enhancements for generative large language models (LLMs) with its PyTorch plug-in.

ZenDNN 5.0 is optimized for 5th Gen AMD EPYC "Turin" processors while it will also work better than prior ZenDNN releases on the new Ryzen 9000 series Zen 5 desktop processors. ZenDNN 5.0 is also compatible with the AMD BLIS 5.0 library, EPYC-specific enhancements to MATMUL operators and related fusions, aut-tuning for BF16, performance enhancements focused on LLMs, optimized Scalar Dot Product Attention (SDPA), support for BF16 precision within the Recommender System models in PyTorch, and graph optimizations and pattern matching improvements within the PyTorch plug-in.

ZenDNN 5.0 has been tested with TensorFlow 2.16+ and PyTorch 2.0+. This morning's [2]ZenDNN 5.0 release announcement notes:

"The focus of the ZenDNN 5.0 release is on delivering support for Zen5 AMD EPYC™ architectures, as well as performance enhancements for generative LLM models through the PyTorch plug-in. The list of models supported includes architectures such as Llama2 and Llama3, Phi2, Phi3, Qwen, ChatGLM, and GPT. The release also delivers performance improvements to non generative LLM models such as BERT."

AMD ZenDNN is available under an Apache 2.0 license.



[1] https://www.phoronix.com/search/EPYC+9005

[2] https://github.com/amd/ZenDNN/releases/tag/v5.0