I’ve run some benchmarks comparing 4.27 vs 5.1, to evaluate upgrading to UE5, and I’ve found that 5.1 performance is considerably lower than 4.27, especially if you are aiming for very high FPS for competitive shooters.
All tests are on a packaged, shipping build. More info in the table. Also this is on epic settings, on lower settings (where I am less GPU bound) the gap is even wider between 4.27 and 5.1.
DX12 seems to have lower performance than DX11 when GPU bound (this was already the case in 4.27 to be fair)
Is anybody else experiencing similar results?
Note: There is a slight difference in lighting in the TPP map because I rebuilt the lightmaps and 5.1 seems to have a regression with lightmass that sky light looks a bit different. But this has no FPS implications.
I noticed a pretty large ~30 fps drop migrating from 5.0 to 5.1, but it ended up being a migration bug that was fixed by manually migrating to a blank 5.1 project.
However yeah, the template I used in 4.27 seemed to run much faster than the same template in 5.0.
Thank you for sharing! I am also starting a new project and it’s some live-streaming stuff that requires every bit of performance. Though, the operating system I’m targeting is Linux using Vulkan. I’ve also came to the conclusion to stick to 4.27 for the time being and then upgrade later.
Here’s what I’ve noticed with my project. The start up time of the engine from my C++ IDE is way faster in 4.27 than 5.0. Though, I’ve found 5.1 to be as fast as 4.27 if not even faster, which proves 5.1 is way more optimized in this regard than 5.
Comparing FPS 4.27 is much faster due to baked lights, though Lumen looks a bit nicer in the same scene.
Another factor for me, are some code plugins I’ve used in this project that have not yet been updated to 5.1, yet.
Im having issues upgrading from 5.0.3 to 5.1. Several bugs and issues with blueprints losing values on several values and performance problems among other things.
Can you explain what you called “manually migrating” so I can test if that can fix my problem?
the migration method i’m pretty sure also doesn’t work, i had the issue mentioned in this thread with 5.1 and now with 5.2 i have the same, i attempted the migration with ~250gb project and still the same in 5.2, i’m getting between 10 - 30% performance degradation and generally towards the higher end, this is a bit of a joke, god knows how many hours i’ve wasted updating my project and testing just to find out it’s always much slower; im guessing 30+, this is a pretty big deal, the issue needs to be identified and fixed
got test with some version below 4.25 and you get mutch more frustrated, i think the gap from 4.23 to 4.27 is worst than from 4.27 to 5.1. so when we compare 4.23 and 5.1 it is ridicilous.
i just watched this UE4 vs UE5 Performance Comparison - The Isle Evrima - YouTube (isle of evrima) so im confused. It turns out after upgrade on ue5 they got signif. more performance. so im confused now. does some have comment about this, or even detailed info like what they use, lumen, nanite, vs … or not? of course there is dilema about visual fidelity in each case but regardless.
UE5 needs to address the serious performance issues, I’ve been rebuilding my project from 5.2 to use UE4.27, because UE4 runs twice faster than UE5 without Lumen
UE5 Chaos vs UE4 PhysX on the demo map of Drivable Cars Advanced:
Tested on RTX 3050 4GB VRAM with 4k resulution on second monitor
UE5: about 10fps vs UE4: about 30fps
Test result for maps using large amount of construction script:
UE5: less than 1fps vs UE4: about 15fps
Tested on RTX 3050 4GB VRAM with 4k resulution on second monitor, 4GB VRAM was just enough for the same map on UE4, but was more than 4GB over budget on UE5
I couldn’t rebuild 5.2 project to UE4 due to too many new nodes used, I’m stuck with UE5 and the left option is to convert to C++ project
I forget some even more important tweaks for UE5.2.1:
Anti-Aliasing Method: None / Temporal Super → Temporal AA
Generate Mesh Distance Fields: Enabled → Disabled (Must be disabled for 50% more performance if not using Lumen)
Nanite: Enabled → Disabled (Also change redering model from SM6 to SM5 for both DX12 DX11 Vucan, from “Platforms / Windows”)
After the above optimization and other options the UE5 project has 3/4 the performance of the UE4 version
i see your test, long time ago, i have simillar results like you, i have no doubt about that. i just cant understand no mather what settings used, how they managed to get better performance with so many foliage in ue5 (forest environment)???
