Remember that Nanite helps cull Nanite, so even though your landscape is/might-be Nanite and more expensive up-front, it can help claw back through culling (and hopefully preventing overdraw),
2 Likes
Yes and the weird thing about it is, that the higher the poly count, the more efficient nanite can cluster cull and prevent overdraw. So more details = higher performance.
This is a mistake that was continously done in this thread, where people just enabled nanite on traditional optimized meshes with large triangles and then complained about overdraw.
Yes, Nanite can not efficiently cull a mesh when the mesh has only 50 triangles 
2 Likes
A new update just dropped with unreal engine 5.5. @JustinDarlington
They introduced a new feature called âMegalightsâ, which basically removes the shadow cost of lights if you have a Raytracing GPU.
This disables Virtual Shadow Maps and completely replaces the system.
Also they did some big improvements to the nanite render pipeline.
I would definitely recommend upgrading your project!
As a comparison here are the old benchmarks from 5.4
5.8ms per Frame, Nanite VisBuffer at around 1.8ms and Light cost at around 1.6-2.2ms, depending on where you are at the scene.
Now here are the new profiler stats. Just updated the Project and enabled Megalights, no other modifications done.
Nanite is now around 1.2ms and the lights are calculated as one Pass in the RaytracingScene for a cost of 0.3ms.
We can now generate the whole image at 4.4ms, which gives a framerate of ~230FPS on 2k resolution. All while having incredible mesh detail, ultra dense foliage and over 40 real time lights in the scene.
That is just incredible IMO.
Sadly i can not test it rn on my VR headset as it is currently not available.
But definitely looking forward to my next game projects 
EDIT: Btw, just like nanite it seems that Megalight scales with screen resolution and not with amount of lights. I did some tests with 4k Rendering and a few hundred lights.
Only when increasing the resolution, the light cost went up. On 4k its around 2.1ms, which is still much faster than VSM.
Using Upscalers 75% it is 1.5ms . So again, upscalers exponentially improve performance here
8 Likes
I guess they listed to the OP! Good thing we have this thread to kick 'em in the pants.
Thanks for the updated results, the video looks amazing.
Thatâs awesome! I definitely want to utilize MegaLights but am afraid of shifting to an RT only system. Those performance numbers are amazing! Iâm planning on trying it out once I complete my game in 5.4.
Whatâs your system specs?
1 Like
I may be late to the conversation, but my dude, I am no developer, I am a gamer with IT background in collage
Epic suggestions is just that, SUGGESTIONS, It doesnât stop from studios from shipping games with it ANYWAY, So it being âBetaâ doesnât do anything but just an excuse of it being not finished. It doesnât help that EPIC doesnât put any BETA label on it.
3 Likes
âIf itâs stupid but works, it ainât stupid.â
guys im starting to think op doesnât have 5 years of experience nor a dev team 
.
i thought his vids were funny but this threat is a whole another level, so please keep quoting his post since he cant delete those.
2 Likes
Not enough people make collages nowdays
1 Like
@TheKJ after reading this thread, my question is what do you propose indie game devs should do about this problem? Switch to Unity?
1 Like
With a raytracing GPU is there a benefit with a single light source ? Like a dense forest and the sun ?
Which I think as a directional light I seem to remember the Megalights doesnât incorporate this type of light.
There are many scenarios which would have these type of lighting, sunny city streets - it could have very dense geometry, and in need of performance, but yet a single directional light source.
But you get increased blur.
I admit itâs interesting but will it require Nanite ?
I still donât Nanite as an indie (not dense enough scenes, I tend to lose performances). If it kills virtual shadowmaps, AWESOME !
5.5 test.
After working on some test, I stumbled upon another false claim presented in Epic performance presentations and other commenters from the UE community.
âHigh poly makes nanite faster, because of cluster culling, which will make overdraw betterâ
That is not true, at all.
While Iâm not going to post the hours worth of work and view modes and standalone test, it was always the same each time. Higher poly (which canât just be subdivided, no each triangle has to have significant difference or Nanite would crunch down to a single cluster in seconds) are heavier on Nanite.
The test just consisted of replacing the low poly cubes with a displaced version with just enough poly density to keep small clusters for âbetterâ culling. No, what happened is every nanite related measurement(thereâs like three) each increased by 45% in ms(.50ms-.76ms). Donât give me lies talking about how Nanite has a fixed cost, it scales with complexity like everything else.
I even took a severally displaced version in case anyone wanted to keep claiming âmore tris is betterâ, cluster culling was spot on to the pixel. Awful perf.
Also, remember that I tested these in several modes outside editor, and Iâm just showing overdraw to save myself the time and give enough context about the layers and how insignificant the view mode is. And these are small scale test that will be even more impactful on real scenes
Your welcome.
Again.
Why are you doing timing tests on a pre-release?
2 Likes
Why shouldnât you?
Because it in no way defines how the final release will perform, misleading if anyone sees this after the release as it doesnât even mention its the pre-release.
2 Likes
The entire engine is a work in progress. If you donât know that then thereâs a bigger problem. It wonât be finished until they announce it and move on to UE6. Until then, expect a lot of performance issues.
Overdraw is only an issue for Translucent/Masked objects. Any mesh using a standard material isnât going to cause overdraw. Not to mention that the engine culls blocked meshes and Nanite also does. All of your claims are completely wrong.
1 Like
What PC system are you using? Iâm surprised to see how much of an impact the performance has on your system. Youâre avoiding testing in real level scenarios, but I conducted some similar basic tests in 5.4 version. Despite an 80x increase in mesh triangles, the performance remains nearly the same. All spheres are static meshes with nanite enabled.
Sphere with first layer of displacement: ~19k Triangles.
Sphere with second layer of displacement: ~1,500,000 Triangles
4 Likes