https://youtu.be/whPSD8sdx-0?feature=shared

<aside> 💡

!TREAT this as a history lesson, not a ground truth.

While most of the information is true and the overall fundamental ideas are in here, the white paper on (NVIDIA) GPU Architecture does not explain the exact architecture so there might be some wrong / omitted information.

</aside>

Highest layer

==

cache : type of fast extra memory, much smaller than DRAM

When accessing DRAM memory, some are saved in Cache memory (for faster access in the future)

In one GPC(Graphic Processing Unit),

• Raster Engine: generates pixel information from triangles
• 6 TPC(Texture Processing Cluster)s
• 16 ROP(render output units) divided into two partitions (8 each)

In one TPC (Texture Processing Cluster),