By using this site, you agree to our Privacy Policy and our Terms of Use. Close
Eddie_Raja said:


ESRAM will always be a major bottleneck no matter what.  It is only 32MB, and a 1080p buffer takes double that space to have any amount of detail on screen.  

But again:  8.000 GB vs 0.032 GB.  Yeah I know which one matters more.


You don't fit an entire frame into the ESRAM. You go with a tiled approach, that's the entire point of it, you still get the speed benefit, whilst using a smaller memory footprint.
It will only be a bottleneck IF a develeper decides to use it... And uses it incorrectly.

You are looking at it from a perspective that it somehow holds the machine back, it doesn't, it's a feature developers can use to extract more performance, not one that hinders performance. There is a difference, learn it please.

Basically... With Tiled based rendering you do this:


The frame is divided into much smaller ones, allowing you to use a fraction of the memory in a single pass, this is why super fast but tiny caches of a few megabytes such as eDRAM and ESRAM have shown to still be extremely beneficial.

The other benefit to this is that culling can become more efficient and more aggressive which farther boosts bandwidth and processing efficiency, there is a reason why the mobile world uses it to great success, because they are bandwidth constrained.

The downside however is that ESRAM/eDRAM is costly, it consumes transisters and electricity which could be better spent on more CPU Cores or more Pixel Shaders.




www.youtube.com/@Pemalite