Current Generation Games Analysis Technical Discussion [2022] [XBSX|S, PS5, PC]

No, its not faster. You're mistaken it. Their software mode is not using BVHs, they are using signed distance fields. In some instances, SW-Lumen can be even faster than HW-Lumen. These are different methodes, it's not like running a DXR game like Control on Pascal!
It is faster... quality for quality the software version will never be as good as the HW version.

I'm at ~60fps in UE5.1 now with hardware Lumen (so better quality than SW Lumen) on a 3060ti at 1440p, I have zero reasons to opt for the SW route on my set-up.

And the software side of it also seems a little pointless to me, what's it going to be used for?
 
No, its not faster. You're mistaken it. Their software mode is not using BVHs, they are using signed distance fields. In some instances, SW-Lumen can be even faster than HW-Lumen. These are different methodes, it's not like running a DXR game like Control on Pascal!
So what you are saying is software lumen is more lightweight in exchange for being less accurate right? And hw lumen is accelerated by GPU RT hardware and more accurate but costs more?

It sounds almost contradictory because one would generally assume the RT HW inside the GPU would make hw lumen faster than a software solution due to making it easier to do. but I guess their software solution is just that refined
 
Is that how it works? Lumen performance can be accelerated with the on GPU RT HW? Seems like something the consoles could take advantage of to make lumen more performant in games no?

HW RT improves Lumen quality not performance. The quality improvement is mostly for high resolution reflections like mirrors. If I’m not mistaken UE5 docs essentially imply you must use HWRT for mirror like reflections as the SDF that software Lumen uses is too coarse.
 
So what you are saying is software lumen is more lightweight in exchange for being less accurate right? And hw lumen is accelerated by GPU RT hardware and more accurate but costs more?

It sounds almost contradictory because one would generally assume the RT HW inside the GPU would make hw lumen faster than a software solution due to making it easier to do. but I guess their software solution is just that refined

Theres simply not much reason to go with a GPU that lacks hw acceleration as there are no drawbacks there. In software its going to be about as fast (when same GPU class) and when hw RT is used its going to be faster.... UE5 iterations will be using hardware ray tracing, too.
 
It is faster... quality for quality the software version will never be as good as the HW version.

I'm at ~60fps in UE5.1 now with hardware Lumen (so better quality than SW Lumen) on a 3060ti at 1440p, I have zero reasons to opt for the SW route on my set-up.

And the software side of it also seems a little pointless to me, what's it going to be used for?
That is because you're likely running it in an empty project or sample project. Then HW-Lumen is just as fast as SW-Lumen while offering better quality.

The Matrix Demo is also specifically created for the usage of HW-RT which is why it's only 7% or so slower than SW-RT. With many instances of rocks and foliage, HW-RT might be a lot slower.
So what you are saying is software lumen is more lightweight in exchange for being less accurate right? And hw lumen is accelerated by GPU RT hardware and more accurate but costs more?

It sounds almost contradictory because one would generally assume the RT HW inside the GPU would make hw lumen faster than a software solution due to making it easier to do. but I guess their software solution is just that refined
Yes indeed, that is precisely it.

You would assume that if both methodes would be using triangle based Raytracing, but that is not the case. SW-Lumen is not a triangle based solution that could be accelerated by Ray accelerators/ RT cores, it is a different methode.

Theres simply not much reason to go with a GPU that lacks hw acceleration as there are no drawbacks there. In software its going to be about as fast (when same GPU class) and when hw RT is used its going to be faster.... UE5 iterations will be using hardware ray tracing, too.
I am sorry to say it, but that's wrong. There are drawbacks with HW-Lumen compared to SW-Lumen. You can find them here: https://docs.unrealengine.com/5.0/en-US/lumen-technical-details-in-unreal-engine/

Trust me guys, I am in the same boat as you are. It hurts me to type this. I wish HW-Lumen would be the much superior solution in all cases, but it is not. UE5 will ruin the fun parade of HW-Raytracing. I've seen already many projects on Twitter using SW-RT only. And according to @chris1515 analysis, Tekken and Silent Hill might be too, if their current state is anything to go by.
 
Last edited:
That is because you're likely running it in an empty project or sample project. Then HW-Lumen is just as fast as SW-Lumen while offering better quality.

The Matrix Demo is also specifically created for the usage of HW-RT which is why it's only 7% or so slower than SW-RT. With many instances of rocks and foliage, HW-RT might be a lot slower.

Yes indeed, that is precisely it.

You would assume that if both methodes would be using triangle based Raytracing, but that is not the case. SW-Lumen is not a triangle based solution that could be accelerated by Ray accelerators/ RT cores, it is a different methode.

Tracing SDF can be accelerated by HW RT. This is for path tracer but it can work for Realtime too and a different representation of a scene. It can work for console and HW-intersection tracing bounding volume is four time faster than HW-intersection tracing triangle on consoles and it can work with a BVH for LOD(fused BVH), this is not possible on DXR but possible on consoles. The final intersection step use a custom compute shader because current intersection HW-RT can't accelerate intersection to SDF.;) The paper is very recent and very interesting. This is a research paper of Nvidia now Alex Evans ex Media Molecule is working there.


On an NVIDIA RTX 3090, the fastest method uses the GPU's ray tracing hardware to trace against a bounding volume hierarchy, built around all non-empty voxels, and then applies either an analytic cubic solver or a repeated linear interpolation for voxel intersection.
 
Last edited:
HW RT improves Lumen quality not performance. The quality improvement is mostly for high resolution reflections like mirrors. If I’m not mistaken UE5 docs essentially imply you must use HWRT for mirror like reflections as the SDF that software Lumen uses is too coarse.
Now I think I get it...so software lumen is largely for performance and hw lumen is for accuracy. But even with software lumen devs will be compelled to use hw RT for reflections because software lumen cannot properly replicate reflections on that level?

So even using software lumen it's likely to be a hybrid approach in the game?
 
Now I think I get it...so software lumen is largely for performance and hw lumen is for accuracy. But even with software lumen devs will be compelled to use hw RT for reflections because software lumen cannot properly replicate reflections on that level?

So even using software lumen it's likely to be a hybrid approach in the game?

I think so. Theoretically you could try using software Lumen with a high resolution SDF for reflections but that could end up being slower than HWRT.
 
Now I think I get it...so software lumen is largely for performance and hw lumen is for accuracy. But even with software lumen devs will be compelled to use hw RT for reflections because software lumen cannot properly replicate reflections on that level?

So even using software lumen it's likely to be a hybrid approach in the game?
I don't know why Epic stated this in their documentary. Mirror like reflections are totally possible with SW-Lumen. There is a project that made heavy use of this:


Sure, the geometry that is being reflected is not as highly detailed as with HW-RT, but that is not something the average gamer is going to notice. I've made a post where I compared mirror like reflections of both solutions somewhere in this forum.
 
Sure, the geometry that is being reflected is not as highly detailed as with HW-RT, but that is not something the average gamer is going to notice.

You can say the same thing about screen space reflections. The standards of an “average gamer” isn’t a very high bar for this stuff.
 
The quality improvement is mostly for high resolution reflections like mirrors.
No. It's for all.

GI: gets to apply on dynamic objects and skinned meshes, it gets extended range covering (up to 1000m), and is of higher resolution as it supports Ray Hit, instead of the less accurate Surface Cache. Indoors scenes get a boost in quality as a result of this. Software GI can't do any of this.
Reflections: they also get far more higher resolution (support Hit Ray), they work on dynamic and skinned objects, Software reflections don't support any of this.
Shadows: RT shadows are better than Shadow Maps when not using Nanite

Mirror like reflections are totally possible with SW-Lumen
No. See above.

But even with software lumen devs will be compelled to use hw RT for reflections because software lumen cannot properly replicate reflections on that level?
Yes. See here:

-Better GI quality and precision in open world games, as hardware Lumen can cover a 1000m distance, and has better resolution and less light leaks.
-Better dynamism, as hardware Lumen (global illumination and reflections) support skinned and dynamic objects (people, vehicles, animals, .. etc), software Lumen can't.
-Better interior scenes, as hardware Lumen supports much higher resolution (rays per pixels), which vastly improves light bounce and reflections in interiors. And renders vastly more details.
-Better shadows for horror games that don't rely on Nanite (that use interior scenes), as hardware ray traced shadows allows for a vast number of shadow casting lights, much higher number of shadows, and a vastly better contact hardening, Virtual Shadow Maps can't offer any of that.
-Exotic use cases, translucency, translucent shadows, caustics, .... etc.
 
No. It's for all.

GI: gets to apply on dynamic objects and skinned meshes, it gets extended range covering (up to 1000m), and is of higher resolution as it supports Ray Hit, instead of the less accurate Surface Cache. Indoors scenes get a boost in quality as a result of this. Software GI can't do any of this.
Reflections: they also get far more higher resolution (support Hit Ray), they work on dynamic and skinned objects, Software reflections don't support any of this.
Shadows: RT shadows are better than Shadow Maps when not using Nanite

Good point on skinned meshes. GI and shadows are better too but much harder to notice. You don’t need crazy high resolution for decent multi bounce diffuse GI.
 
HW Lumen quality is so far a head of SW Lumen in UE5 it's not even funny.


By the time we start getting true next generation UE5 games the GPU's that would benefit from SW Lumen will be nothing but eBay fodder.
UE 5.1 made some sigificiant improvements to SW-Lumen.

And even then, you're not going to look at a ball of steel while you are playing a game all the time. Saying the quality is so far ahead is an exaggeration when there will only be a handful of scenes in a real game where you will be able to spot differences, especially in an open world. This is not a Metro Exodus -> Metro Exodus Enhanced Edition situation at all. SW-Lumen is so competent, that it absolutely obliterates SSR and SSGI.
 
Saying the quality is so far ahead is an exaggeration when there will only be a handful of scenes in a real game where you will be able to spot differences, especially in an open world.

They're massively noticeable in The Matrix demo.

And it's not an exaggeration, there's literally a video I posted above that shows how far a head HW solution is.
 
For usage on PC the answer is fairly straightforward. Devs will give players options for the highest end no compromise stuff to push their hardware up into the high end market.

But I am curious as to what the most likely scenario for console is to reach 30 or especially 60 in ue5. I feel like devs will have to make careful choices and tradeoffs in order to make games work which I find very interesting.

Ue5 seems to have trivialized the polygon budget issue of previous generations even for the consoles(considering how many can be on screen at a time now without performance loss) but replaced it with raytracing vs software lighting/reflections/shadows considerations that have to be fully balanced.

Seems like even in software though it's a lot better than traditional solutions that are faked and maybe more labor intensive to artists as well(I can only think of how tlou 1 and 2 baked their light to fake natural light seeping into certain areas)
 

Attachments

  • 91009cba16e4863219dcfacfc2bfa63c--year--uni.jpg
    91009cba16e4863219dcfacfc2bfa63c--year--uni.jpg
    29.4 KB · Views: 15
For usage on PC the answer is fairly straightforward. Devs will give players options for the highest end no compromise stuff to push their hardware up into the high end market.
Yes indeed, on PC, UE5 titles will offer Ultra settings that activate better global illumination and reflections through hardware ray tracing automatically, as it's just a flip of a switch, with literally no extra work on part of the developers. They might even scale those up to higher degrees. UE5 supports Epic and Cinematic settings, those ramp up the rays per pixel count, and offer more bounces for global illumination and reflections as well as more resolution.
 
Of course but we don’t actually have an objective measure of value that we can use. Rasterized FPS/$ certainly doesn’t capture all of the value to the consumer.
Indeed, but that just adds into the 'mystique' of marketing, as no-one really knows what they really want from a GPU. As such it can't be a bad thing someone is presenting an argument in favour of AMD against the entrenched market leader, even if that argument doesn't apply to a lot of PC gamers. In particular, nVidia is commanding ever increasing price premiums that people are buying into like they're designer sunglasses. The perception of "highest price == best product" needs to be broken for balance to be restored to the GPU sector.
 
Indeed, but that just adds into the 'mystique' of marketing, as no-one really knows what they really want from a GPU. As such it can't be a bad thing someone is presenting an argument in favour of AMD against the entrenched market leader, even if that argument doesn't apply to a lot of PC gamers. In particular, nVidia is commanding ever increasing price premiums that people are buying into like they're designer sunglasses. The perception of "highest price == best product" needs to be broken for balance to be restored to the GPU sector.

It’s not a bad argument but given the market has rejected it over and over maybe it’s time to try something new.
 
Back
Top