Let's talk about Specs

[DonFerrari]

SvennoJ said:

Hynad said:

Reeks of bottleneck? The speed at which data transfers from read to memory is incredibly fast. RAM can be filled in two seconds. TWO SECONDS!

It gets unloaded, then loaded again in two seconds. Wrap your head around that! O_o

What good does that do when frames render in 16 ms. Wrap your head around that, it takes 120 frames to fill the RAM. One hundred and twenty!!!

Compromises will have to be made. 1080p SDR to 4K HDR is 8x increase in memory. Ray tracing is a memory hungry process. Procedural generation is a memory hungry process. Living/changing environments, very memory hungry. Persistent worlds (all changes stay) very memory hungry. More variation in assets and textures, need more memory. Megatexures, need a lot of memory. Longer draw distance, bigger fov, need more memory. Memory is one thing you can never have too much off and the one thing you can't upgrade during the gen. (Well you can, but it's pointless cause parity needs to be maintained)

The SSD can help, but it's still at least 50 times slower than memory, with compression.

It's enough to get what we're used to in 4K and HDR. But I wouldn't expect any revolutionary new dynamic worlds etc.

The image don't completely change every frame, and the RAM increased. We haven't seem yet how this balancing will work and how much constrain we will have at the end of the gen.

They are taking every step to minimize the RAM need for texture and others (like taking in consideration the distance draw for the quality of texture, only showing the visible part, not rendering what is behind you).

And on RTX Cerny mentioned that they have achieved some good results with little load, but of course it won't be full RTX for most games, it will vary depending on the other loads and intentions of the dev.