In reality, the PS5 Pro achieves 16.7 TFLOPs compared to the PS5's 10 TFLOPs. Meanwhile, one of the accurate pre-release leaks pointed toward the PS5 Pro achieving 300 TOPS when performing 8-bit calculations. 16-bit calculations, meanwhile, can achieve 67 TFLOPS. As Cerny clarifies, RDNA 2.X, Sony's customized AMD RDNA 2 architecture, uses many RDNA 3 features but maintains enough of the original architecture to not force code rewrites on the new hardware.

According to Mark Cerny, the biggest improvements of PS5 Pro are considered to be its new Ray Acceleration structure using BVH8 (Bounding Volume Hierarchy) and leveraging improved "Stack management in hardware," which means that graphics shader code is now better-managed, simpler, and more performant on the new hardware.

BVH refers to how bounding boxes, a common feature of 3D rendering, are used to make graphics calculations like reflections. BVH4, with bounding boxes in groups of 4, was used on PS5 for RT calculations, while PS5 Pro can now leverage BVH8 (8 bounding boxes) for its RT calculations. Similarly, the Ray Intersection Engine has doubled from checking rays against 4 boxes and 1 triangle (PS5) to 8 boxes and 2 triangles (PS5 Pro).

These improvements to ray tracing hardware in the PlayStation 5 Pro, made possible through an incredibly customized version of the RDNA 2 GPU architecture used within the PS5, give great performance gains with curved and bumpy light reflections but only moderate gains with shadows and flat reflections.

The full 37-minute video on the PlayStation 5 Pro is recommended for more technical information. It includes lots of interesting hindsight into the console market and the technologies required to compete in it.