We may have gotten something like that, I hear the render thread was increased by 50% in 5.4
They claim a 24% increase on the GPU side but that isn’t exactly true on my side, even with hardware similar to 9thgen consoles. I imagine this comment accurately reflects Epics claim on “performance improvements”.
I’m still doing regular test to make sure, just waiting.
Refresher/TLDR to Epic Games if they are keeping an eye after the recent 15+ votes that have catapulted this thread from the original 1# feedback thread:
-
Make effects as performant as other implementations. If I can get ms timing on effects in other games, you can too via api/hardware inspectors. UE’s implementations of the same effects always fallbind other versions with little improved or worse quality in UE. Make them stable indepently of Temporal AA/Upscaling(DLSS included). That includes your terrible dithering effects and making temporal variants of your effects the start up default and not hidden from developers.
-
And make better designs related to shadows, lighting and raymarched reflections that take advantage of the common game environment being: 80% static environments with dynamic lights with precomputed movement and or small dynamic lights. Rather than systems that are made for games like FN were absolutely no information can be baked,tweaked(brightness, area etc), or interpolated by developer artist becuase Lumen fits FN which is 100% Dynamic/Destructible world combined with a constant moving sun where constant iterations on the GPU waste performance. Examples of better designs (in term of caching/performance, not quality) would be The Divisions GI, Quantum Break’s GI , and I have explained some more stuff in the Lumen thread.
-
Do not tie effects to Nanite(like PGC or VSMs) or continue mislead people with it’s very situational benefits. Foliage is one of the few kinds of popular meshes where overdraw should be tackled with something like visibility buffers but WPO explodes both VSMs and Nanite.
—So for the that reason UE(the many games that will use this engine in the future) desperately needs something like deferred texturing as HFW performs way better than any UE5.3- game with nanite. Another investment related to this area is a better LOD system that optimizes topology for overdraw and creates LODs based on a higher quality algorithm.
—In this link, I state systems that would greatly benefit the optimization workflow for draw calls via precomputation meshlets that developers could create from finished scenes with static(at least) geometry.
—In the same link I state a system that would help memory as well.