Render thread time reduced by 50% and GPU time decreased by 25%!! That means 2x and 1.5x times faster!!! 
But, wait, have you noticed anything at all? I have already opened some projects since 5.4 preview, some ācommonsā and some with a lot of nanite and vsm, like the Electric Dreams. And I have noticed speed/performance improvements in approximately theā¦ 0% of my projects. In general, slightly performance drops, in fact.
Please, give me some good numbers!
I also didnāt notice any performance boost at all. Took City Sample, Valley of Ancients. Nothing. I tried to look for any plugins, even experimental stuff, but no. They said itās out-of-box improvements. So, hopefully to see any details on those updates
If I recall correctly, the City Sample is heavily CPU bottlenecked by the game thread, so I wouldnāt be surprised if render thread optimizations donāt do anything.
On my project, instead of improving performance, I got a sawtooth frame time graph
UPD. If anyone has the same problem - disable Real Time Capture in your SkyLight
Thanks for the info, I was having the same issue. Turning off the Real Time Capture did indeed stop the FPS saw tooth. Iām still testing but the issue does not seem to return if I turn the Real Time Capture back on. odd.
Regards
Last time that I checked all the features, departments and rendering aspects ending up in a tug of war regarding performance stagnating any GPU performance increase.
Anyone can view what got faster and what got slower here, this is preview but Iām working on updating to āstableā 5.4.
My guess is the GPU performance improvements are extremely conditional(abusing 5.3 for no good reason other than making abusive 5.3 scenarios faster in 5.4). What would improve performance in UE title releases would to be creating better optimization workflows for developers(better LODs algorithms, precomputed hierarchy meshlets, LOD streaming based on a global speed variable).
There is only so much performance they can give us without redoing several designs which I doubt Epic will fund any time soon. They keep iterating on fundamental broken technologies with almost no changes every time. Lumen works the same way, we work with it the same way. Nanite and TSR, again we flip them on and have no options for anything(TSR doesnāt have enough variables to fix itās fundamental issues).
The numbers given by epic games are vs 5.0 not 5.3.
some of my unoptimized test maps have seen big performance degradation and i removed them from the projects. mostly foliage spammed maps. running the highest tier setup i can do. all other test cases have seen and continue to see improvements in test visuals and performance.
Did a small test with Nanite, thought I should share here since this thread will already contain people keeping an eye on performance.
I REPORT: 1fps gain in 5.4(in the custom, optimized benchmark)
No Editor(launched game from project). Test were done with a copied project switched from 5.3 to 5.4 then each variable visible in the 5.3 where applied(black bars are scalability I decided needed to be raised or lowered, old cvers not related to the test). Each test was done after a fresh system restart for maximum performance. 1440p on a 12GB 3060.
