Digital Foundry Article Technical Discussion [2024]

UE5.4 is tested by Digital Foundry in Fortnite and City Sample demo and compared against UE5.0

-CPU performance is a lot higher (50% to 80% higher depending on the scene/system).
-Shader stutters and traversal stutters are still a major problem that affects frame pacing.
-Hardware Ray Tracing now allows for high resolution reflections and more bounces for emissivie lights.

A very good and much needed increase in performance. It makes UE5 marginally more palatable, however the stutter issues remain and that's disappointing. Epic needs to clean this up and automatically fetch all the shaders for a pre-compilation step. Devs should not have to even worry about this and it should be handled automatically by the engine. Honestly, this should be one of their highest priorities because it's almost a decade now this has been an unaddressed issue. UE4 struggled with this from the get go and now we see it again on UE5. If they can do these things, the engine would probably get less backlash. Performance still needs to improve and hopefully they keep at it.
 
I'd be curious to know how fortnite compares to other games in terms of the number and total size of shaders. There were 1726 skins as of February. Then there's pick axes, back bling, dive trails, weapon wraps that all might have unique shaders. That's on top of whatever shaders are needed for the BR map and any other objects that are available in Creative.

This is my install:

1715359120770.png
 
UE5.4 is tested by Digital Foundry in Fortnite and City Sample demo and compared against UE5.0

-CPU performance is a lot higher (50% to 80% higher depending on the scene/system).
-Shader stutters and traversal stutters are still a major problem that affects frame pacing.
-Hardware Ray Tracing now allows for high resolution reflections and more bounces for emissivie lights.


Nice improvements...except for you know, everything else. Looks like #stutterstruggle will be a popular PC gaming hashtag for the next few years at least, jfc.

I'd be curious to know how fortnite compares to other games in terms of the number and total size of shaders.

I'm sure it's up there with some of the most heavy shader games on the market, if not at the top. Which is why it's the perfect testbed for how the engine deals with stutter.

If your premier game with access to the best technical resources available, the one that every other publisher has been chasing with dreams of similar endless cash flow, can't run smoothly on your own engine on the most expensive gaming rigs on the market, then either it speaks to a glaring deficiency with the PC as a hardware platform (or its API's) with its very need to compile shaders when combined with a large amount of varying assets, or your engine is still ill equipped to deal with it, years on. Or a combination thereof.

Or hell, decades on if we count since Unreal 3, which also had these issues.

As @Remij has also said, this just further reiterates the need for a kitchen-sink approach to this, and for me that also includes some form of distribution of at least the PSO's to generate the binary caches for games offline, something like Steam's fossilize (albeit hopefully working more consistently). Yeah I know the hurdles in getting something like that, and it's probably a discussion best continued in the shader thread regardless, but it's clear that we're likely going to go through another era where we can't rely on the engine maker/game developer/publisher to address this sufficiently within the game, so we'll need an army of shader butlers to at least suffer for the rest of us - ideally. :(
 
Last edited:
UE5.4 is tested by Digital Foundry in Fortnite and City Sample demo and compared against UE5.0

-CPU performance is a lot higher (50% to 80% higher depending on the scene/system).
-Shader stutters and traversal stutters are still a major problem that affects frame pacing.
-Hardware Ray Tracing now allows for high resolution reflections and more bounces for emissivie lights.

One thing I did notice immediately is that the visuals were definitely not a 1:1 match, with the new UE5.4 generally looking worse.

Particularly around the 4 minute mark, with the performance comparisons under the viaduct - the Lumen light boiling was crazy, even at a distance, versus in the UE5.0 version you had to look pretty closely to see it happening at all.

 
One thing I did notice immediately is that the visuals were definitely not a 1:1 match, with the new UE5.4 generally looking worse.

Particularly around the 4 minute mark, with the performance comparisons under the viaduct - the Lumen light boiling was crazy, even at a distance, versus in the UE5.0 version you had to look pretty closely to see it happening at all.


Alex talks about that near the end of the video.
 
I'd be curious to know how fortnite compares to other games in terms of the number and total size of shaders.
Me too, if that 7.12gb is typical then someone like myself with many hundreds of games installed may have to sacrifice terabytes of storage and I'm guessing it prefers nvme storage.
 
There's a big difference between just compiling a project between different revisions of the engine and seeing how they perform... and building a game.

I feel like we've already known for a while that there are certain PSOs which can't currently be auto-gathered and pre-compiled. RT PSOs can't atm.. and I don't know if/when that will change anytime soon.. but judging from what I've seen, this improvement is pretty massive already. I feel like in a typical game, it would probably cover most cases of comp stutter and result in an extremely good experience. I may be wrong, but we'll see.

Regardless, this test/video is one I've been waiting for and I'm glad to see it. It just goes to show that for as much as they're doing on their end (Epic), we're still going to need publishers/studios to take the issue seriously on their end as well.

PC gamers have shown very recently how well they can come together and achieve a goal.. if it can be done there, it can be done here as well.
 
Particularly around the 4 minute mark, with the performance comparisons under the viaduct - the Lumen light boiling was crazy, even at a distance, versus in the UE5.0 version you had to look pretty closely to see it happening at all.


Even though Alex did acknowledge it, its worth pointinf it out again. The constan boiling jumped at me often throughout the video. Its not a small quality regression. I consider it pretty major, to be honest.
 
Wow. The noise on those lights is really rough. Like 1/32th res particles/DOF blur filter. Significant change in behaviour.
 
Yeah, I thought the light boiling under the underpass was much more agregious than the reflections Alex was drawn to. I suppose what more can you expect from a DF staffer with such a massive ray tracing bias. Shameful. 😁
 
I'm wondering if this is a case where the engine upgrade to an existing project without modified assets/settings is significantly amplifying the boiling errors here. I just can't imagine that's going to be representative of new UE5 games going forward, it's so egregiously obvious and completely destroys the visuals.
 
I'm wondering if this is a case where the engine upgrade to an existing project without modified assets/settings is significantly amplifying the boiling errors here. I just can't imagine that's going to be representative of new UE5 games going forward, it's so egregiously obvious and completely destroys the visuals.
Unfortunately, it's already a problem for UE5 titles in scenes with indirect lighting or scenes lit directly by emissives. It's one of the draw backs of the current Lumen implementation. Night time in City sample is entirely lit by emissives, which is why it suffers the most. In UE5.4 emissives and small light got expanded coverage to contribute to lighting the scene, which exposed them to this Lumen flaw.
 
Last edited:
I'm wondering if this is a case where the engine upgrade to an existing project without modified assets/settings is significantly amplifying the boiling errors here. I just can't imagine that's going to be representative of new UE5 games going forward, it's so egregiously obvious and completely destroys the visuals.

I think its improbable that assets have much to do with it. At this point in time, game tech is at a relative PBR standardization paradigme regarding that stuff. I think some of the Lumen optimizationz were simply not "non-destructive" towards final output. If their final gather aproach is still the same as the inital technique, its due to an ovehaul. Modern stochastic sampling+denoising GI aproaches are working miracles considering the low sampling rates they have to work with. I think UE5 might be behind on that area.

Their choice of generating low-res screenspace-bound probes always felt kind of hacky from the start. There are more robust ways to go about it.
 
Last edited:
I think its improbable that assets have much to do with it. At this point in time, game tech is at a relative PBR standardization paradigme regarding that stuff.
There's almost certainly content things going on here too. I haven't looked at the specific cases but it's very common for people to fudge things even in a PBR context to make it look good. It would not surprise me in the slightest if - for instance - those lights have some unrealistically-ridiculous radiance value that wasn't an issue at the time based on how the balance of Lumen's screen traces and other things were integrated together.

Of course it's definitely possible that while things advance in one area they can regress in others (ex. I think it's fair to note in some of those examples that TSR maybe looks a bit worse in 5.4 CitySample than the original release as well - but some of that is a balance with how much better it looks in other cases like transparency now). It's probably a combination of both.

But certainly the fact that we have effectively demo content that gets crunched to look as good as possible with a snapshot in time of the engine (which may or may not even correspond to a public release) and then rarely gets updated later is a problem that does come up a bunch. Of course this is all on Epic in this case as it owns both the engine and the content; I think it would be great if we could find some cycles to go back and apply the latest best practices to stuff like CitySample, but finding time from art teams to do that is even more difficult than engineering as they tend to move from project to project in high demand.
 
There's anger, bafflement and frustration this week as John, Rich and Alex discuss Microsoft's decision to close one studio that created one of the greatest immersive games ever made and another studio that's just delivered a BAFTA-winning, critically acclaimed rhythm action game... so what is going on at Xbox right now? Beyond that, the team discuss the latest Switch 2 revelations, the new Asus ROG - or R.O.G. - Ally, Hellblade 2 PC specs and Sony's massive Helldivers 2 own-goal.

0:00:00 Introduction
0:01:38 News 01: Microsoft shutters Tango Gameworks, Arkane Austin
0:44:37 News 02: Nintendo confirms Switch sequel, potential components uncovered
1:03:42 News 03: Asus announces new ROG Ally
1:16:13 News 04: Sony backtracks on Helldivers 2 PC PSN requirements
1:25:25 News 05: Ninja Theory reveal Hellblade 2 PC specs
1:36:15 News 06: Sand Land released!
1:44:17 Supporter Q1: Could the Switch 2 connect to a TV wirelessly?
1:46:52 Supporter Q2: How much longevity will video cards with 8-10 GB of VRAM have?
1:53:41 Supporter Q3: Should next gen consoles have much larger RAM allotments?
1:57:40 Supporter Q4: Do you think mouse and keyboard support will become more common on consoles?
2:00:49 Supporter Q5: How feasible is it to update an older UE5 game to UE 5.4?
2:03:50 Supporter Q6: Could Intel beat AMD in discrete GPU sales by 2029?
2:06:40 Supporter Q7: What’s your favourite immersive sim, and why is it Prey (2017)?
 


0:00:00 Introduction
0:01:38 News 01: Microsoft shutters Tango Gameworks, Arkane Austin
0:44:37 News 02: Nintendo confirms Switch sequel, potential components uncovered
1:03:42 News 03: Asus announces new ROG Ally
1:16:13 News 04: Sony backtracks on Helldivers 2 PC PSN requirements
1:25:25 News 05: Ninja Theory reveal Hellblade 2 PC specs
1:36:15 News 06: Sand Land released!
1:44:17 Supporter Q1: Could the Switch 2 connect to a TV wirelessly?
1:46:52 Supporter Q2: How much longevity will video cards with 8-10 GB of VRAM have?
1:53:41 Supporter Q3: Should next gen consoles have much larger RAM allotments?
1:57:40 Supporter Q4: Do you think mouse and keyboard support will become more common on consoles?
2:00:49 Supporter Q5: How feasible is it to update an older UE5 game to UE 5.4?
2:03:50 Supporter Q6: Could Intel beat AMD in discrete GPU sales by 2029?
2:06:40 Supporter Q7: What’s your favourite immersive sim, and why is it Prey (2017)?

@Dictator
"Does the game default to a 21:9 aspect ratio like it does for console" "I can't even think of a game, off the top of my head, that would do that"

The PC version of Armed and Dangerous (2003) does that. It displays a letterboxed image at all aspect ratios.
Even in Eyefinity mode:

ingame_3x1_0.jpg
 
1:25:25 News 05: Ninja Theory reveal Hellblade 2 PC specs
Regarding this point, DF crew were curious why would the developer increase CPU requirements when increasing the resolution of the game? the problem is this new trend we are witnessing where certain games will increase LOD and shadow draw distance upon increasing the resolution, thus putting more burden on the CPU.
 
Last edited:
Back
Top