Zen, CUDA, and Tensor Cores, Part I: The Silicon

This article is the first part of a miniseries exploring the differences between Zen cores, CUDA cores, and Tensor cores in modern computing. It aims to provide a detailed understanding of these core types by examining their physical structures on silicon chips, specifically comparing an AMD Ryzen 5 7600 CPU and an NVIDIA RTX 4090 GPU. The author highlights that while the Zen 4 core and CUDA cores serve different functions, they share similarities in layout and design. The article discusses the complexity of identifying specific cores within the chips, noting that the Zen 4 core is significantly larger than the CUDA cores. The analysis reveals that the GPU's architecture is more intricate, with various components such as Graphics Processing Clusters (GPCs) and Texture Processing Clusters (TPCs) that house the CUDA and Tensor cores. The author emphasizes the challenges in accurately measuring core sizes due to limited resolution in die shots and the lack of detailed information from manufacturers. Ultimately, the article concludes that while Zen 4 cores are more complex, CUDA and Tensor cores are smaller and less intricate, setting the stage for further exploration in subsequent parts of the series.

- The article compares Zen, CUDA, and Tensor cores in modern CPUs and GPUs.

- It highlights the physical structure and layout differences between these core types.

- Zen 4 cores are larger and more complex than CUDA and Tensor cores.

- The analysis faces challenges due to limited resolution in die shots.

- The series aims to deepen understanding of core designs and their functions.