AMD has announced its comprehensive AI Platform strategy, which encompasses a range of hardware products designed to cater to AI needs from cloud to edge to endpoint.

As part of the announcement, AMD introduced the Instinct MI300X accelerator for generative AI workloads. Built on the next-gen CDNA 3 accelerator architecture, the MI300X boasts up to 192 GB of HBM3 memory to deliver optimal compute and memory efficiency. The accelerator is capable of accommodating large language models, such as the Falcon-40, a 40B parameter model, on a single MI300X accelerator. Additionally, the AMD Instinct Platform integrates eight MI300X accelerators into a standard design, providing a powerful solution for AI inference and training. The MI300X accelerator is expected to be available for sampling to key customers in Q3.

AMD also highlighted its commitment to an open and collaborative AI software ecosystem. The company showcased the ROCm software ecosystem for data center accelerators, which includes collaborations with industry leaders like PyTorch and Hugging Face. The integration of AMD’s ROCm software stack with PyTorch 2.0 enables immediate support for PyTorch on all AMD Instinct accelerators, facilitating the utilization of AI models on AMD’s accelerators.

Hugging Face, known for its open platform for AI builders, announced plans to optimize numerous Hugging Face models for deployment on various AMD platforms, including  Instinct accelerators, Ryzen and EPYC processors, Radeon GPUs, and Versal and Alveo adaptive processors.

[Image courtesy: AMD]