Read on Twitter
People are really dumb enough out there to be saying that the Series X's 12 teraflops worth of compute is for the cloud, and not games. Nope, it's for ray tracing, but beyond that it's for a level of geometry performance/flexibility that the PS5 simply will never be able to match
If I personally had to guess what the biggest motivating factor was for Microsoft to go with a 52 Compute RDNA 2 GPU with hardware support for every RDNA 2 feature, I would say it was because of Mesh Shaders and Amplification Shaders.
Mesh Shaders and Amplification Shaders together are on another level compared to the PS5's Geometry Engine and Primitive Shaders because Mesh Shaders together with the Amplification Shader allow the GPU operate in a more parallel nature than what's possible with Primitive Shaders
The goal is to take advantage of what GPUs do best, and to no longer limit the geometry pipeline to a more limited, inferior form of doing what they do best, a method that currently doesn't come remotely close to taking advantage of all those ALUs/Compute capability on GPUs.
But why specifically is the PS5's Geometry Engine and its Primitive Shaders feature inferior? It still relies on some of the primary weaknesses of the older, more limited geometry pipeline, one is called the Input Assembler, and the other is the fixed function tesselator.
The natural weaknesses of the current setup aka fixed function blocks can end up doing wasteful work and thus end up wasting memory and bandwidth. Developers making the move to Mesh Shaders and Amplification shaders will make far better use of a GPU's memory bandwidth.
The larger and more ALUs or Compute Units a GPU has, the better and more powerful Mesh Shaders become. This is why the 52 Compute Unit GPU. People who thought it wasn't going to be of any potential benefit shows you they don't know what they're talking about.
As developers become more accustomed to using Mesh Shaders and Amplification Shaders, Amplification Shaders will end up eventually replaced hardware tesselators all together. Games that have the resources and the budget to pursue will benefit greatly.
All the more reason Xbox purchasing Bethesda made so much damn sense. Can you imagine Starfield, Elder Scrolls 6, the next Doom or Fallout taking advantage of this tech? How about Hellblade II or some of the other big releases coming and in development.
Now do you see why Xbox making sure to fully support RDNA2, pushing DX12 Ultimate, and aiming for a unified, consistent development API between PC and Xbox was so smart? Xbox games going to PC is in light of this is an enormous advantage that many still fail to fully appreciate.
Simply put, Mesh Shaders are a game changer. The raw benefits in a console meant for the next 6-8 years having all this is pretty significant. The old pipeline being the way it was has held back game engines for years. The PS5's Geometry Engine and Primitive Shaders offer..
..some of the same benefits, and will do great things for PS5, but primitive shaders doesn't go nearly far enough, and the PS5's much smaller number of ALUs will hold it back. Remember what Mark Cerny said about keeping more than 36 CUs busy? Mesh Shaders.
https://microsoft.github.io/DirectX-Specs/d3d/MeshShader.html Read about Mesh Shaders here.
Check out what Mesh Shaders can do for Storage, RAM and bandwidth. The highest 201,972 triangle model went from 2.3MB down to 0.7MB when using Meshlets. They can be compressed too, even in cases where the device may not support meshlets. You can ship them on a disc.
Imagine the potential benefits to game sizes when this starts being used more heavily.
DXR uses more memory also, so toss in Sampler Feedback for things like textures and Mesh Shaders for geometry, quite a bit can be done to help further improve ray tracing by offering it more of the memory it requires.
