Please explain clearly the structure of the GPU

I want to know exactly the GPU structure. On the datasheet, there are 4 TPCs, 512 Cuda cores, and 64 Tensor cores. I am curious about the structure that can identify the relationship between TPC, Cuda cores, and Tensor cores. Please let us know if there is any material expressed in pictures.

Hi @hosung0608.kim, you may be interested in the Volta Architecture Whitepaper for more information. That whitepaper is about V100, but the GPU in Jetson AGX Xavier is essentially a scaled-down version with otherwise similar architecture.

There is also this Xavier blog with more details: https://developer.nvidia.com/blog/nvidia-jetson-agx-xavier-32-teraops-ai-robotics/

Thank you for your reply. Is the GPU structure structured as shown in the attached picture?

It is similar, just with fewer TPCs. There is this diagram of Jetson AGX Xavier GPU from the blog:

How are the 4 TPCs configured? Isn’t the group of SMs becoming TPC?

My understanding is that the SMs are grouped the same on Xavier as other Volta architectures - 2 SM’s per TPC.