Zen, CUDA, and Tensor Cores - Part 1
Summary
TLDRThis video script delves into the physical differences between Zen, CUDA, and Tensor cores in CPUs and GPUs. Using AMD's Ryzen 57600 and Nvidia's RTX 4090 as examples, it explores the silicon layout to locate these cores. The script discusses the challenges in identifying CUDA and Tensor cores due to their small size and lack of detailed information from Nvidia. It also touches on the practice of binning and chiplet strategies used by manufacturers to optimize yield and performance, providing insights into the physical design and manufacturing processes behind these computing components.
Takeaways
- 😲 The video aims to provide a detailed understanding of the physical differences between Zen, CUDA, and Tensor cores in CPUs and GPUs.
- 🔍 The presenter uses high-quality die shots of an AMD Ryzen 57600 CPU and an Nvidia RTX 4090 GPU to visually compare the cores.
- 🏗 The Ryzen 57600 has six Zen4 cores, while the RTX 4090 has 16,384 CUDA cores and 512 Tensor cores, although the physical layout of these cores varies significantly.
- 🔎 The video discusses the layout and size differences of Zen, CUDA, and Tensor cores, with Zen4 cores being much larger and more complex than CUDA or Tensor cores.
- 🧩 The concept of 'binning' in chip manufacturing is explained, where chips with more defects have fewer enabled cores, affecting performance and pricing.
- 🤔 The video speculates on why there might be more physical cores than enabled ones on a chip, suggesting it's a strategy to manage manufacturing defects.
- 📊 The presenter estimates the size of CUDA and Tensor cores based on the available data, acknowledging the limitations due to the resolution of die shots.
- 🔗 The video acknowledges the contribution of Fritz Chien, who provides the high-quality die shots used in the analysis.
- 🔍 The video concludes that despite the physical differences, the functionality and performance of these cores are what truly matter for end-users.
- 🔄 The video series will continue to explore what these cores do and why their sizes and structures are optimized the way they are.
Please replace the link and try again.
Outlines

This section is available to paid users only. Please upgrade to access this part.
Upgrade NowMindmap

This section is available to paid users only. Please upgrade to access this part.
Upgrade NowKeywords

This section is available to paid users only. Please upgrade to access this part.
Upgrade NowHighlights

This section is available to paid users only. Please upgrade to access this part.
Upgrade NowTranscripts

This section is available to paid users only. Please upgrade to access this part.
Upgrade Now5.0 / 5 (0 votes)