Digital Foundry Article Technical Discussion [2021]

Status
Not open for further replies.
And this is why 30fps isn't going anywhere. People are loving their 60fps or bust mantras nowadays, but as soon as a developer shows up and blows us away with what the hardware can do with a 30fps target, people will pick up their jaws and proceed to buy the game and enjoy it perfectly fine, just as they did last generation.
 
And this is why 30fps isn't going anywhere. People are loving their 60fps or bust mantras nowadays, but as soon as a developer shows up and blows us away with what the hardware can do with a 30fps target, people will pick up their jaws and proceed to buy the game and enjoy it perfectly fine, just as they did last generation.
But this time, like many are already doing, they should be able to do performance (up to 120hz) and quality modes thanks to the much better CPUs. Apparently they said that demo is not hammering that much the CPU so I think a performance mode in those games should be possible as long as they are not afraid of making resolution sacrifices (and also as we have seen no performance modes, or no RT, in more and more XSS games because resolution is already low enough).
 
But this time, like many are already doing, they should be able to do performance (up to 120hz) and quality modes thanks to the much better CPUs. Apparently they said that demo is not hammering that much the CPU so I think a performance mode in those games should be possible as long as they are not afraid of making resolution sacrifices (and also as we have seen no performance modes, or no RT, in more and more XSS games because resolution is already low enough).

Yup, I'll be extremely surprised if most developers don't continue to offer at the very minimum a 60 FPS mode even if their "quality" mode is 30 FPS.

One mode for people that like gameplay and one mode for advertising and people that like to zoom in on the image. :)

Regards,
SB
 
But this time, like many are already doing, they should be able to do performance (up to 120hz) and quality modes thanks to the much better CPUs. Apparently they said that demo is not hammering that much the CPU so I think a performance mode in those games should be possible as long as they are not afraid of making resolution sacrifices (and also as we have seen no performance modes, or no RT, in more and more XSS games because resolution is already low enough).

Yep. I'd be curious to see what kind of framerate PS5 and Series X could do in this Matrix demo if they were running at Series S resolution. They said they have headroom on the cpu for more expansive gameplay, but do they have headroom on the cpu for gameplay plus 60fps.
 
But this time, like many are already doing, they should be able to do performance (up to 120hz) and quality modes thanks to the much better CPUs. Apparently they said that demo is not hammering that much the CPU so I think a performance mode in those games should be possible as long as they are not afraid of making resolution sacrifices (and also as we have seen no performance modes, or no RT, in more and more XSS games because resolution is already low enough).
Many are already doing this because hardly any games are pushing the systems yet. Of course the overhead to hit 60fps is easily there in basically any cross-gen game out there. This is not necessarily gonna be the case for any developer who really wants to push what these machines can do, as of course plenty will. There's nothing special about this hardware that ensures that 60fps is always gonna be an option unless devs all choose to ensure it is. CPU's are good, but CPU's can be pushed, too. Devs here might say it's not hitting the CPU super hard, but that's still different than saying they can cut the frametime target to below 50% of what it is now(as it's not even close to running at a solid 30fps here) without bottlenecking the CPU.

I think right now, 60fps modes seem 'the way to play' in most games cuz the compromises for choosing it are so small in most cases. That is going to change. And I dont think a developer will be punished for making a 30fps-only game when they can demonstrate that the end result is worth it. I'd also be disappointed if every game this generation is playable at 60fps on console. Would just tell me not a single developer has really pushed the hardware.
 
And I dont think a developer will be punished for making a 30fps-only game when they can demonstrate that the end result is worth it.

If it happens, they may not get punished for it but just like with the PS4 generation, I expect to see quite a few people complaining that the game was only 30 FPS. I heard a lot of that with games like Bloodborne, HZD and other games that were locked to 30 FPS.

Heck, I'd expect even more complaining with the PS5 generation as now more people have gotten to experience what 60 FPS brings to a game WRT gameplay. It's now not just PC gamers who played at 60+ FPS on PC complaining that they are forced to play at 30 FPS. It'll also be console gamers who are getting used to 60 FPS gaming who will be complaining about being "forced" to play at 30 FPS.

Regards,
SB
 
some clarifications from an epic dev

vDOYzgd.jpg

pNfCre9.jpg
 
So UE5 transitioned from tracing against SDF in software for Lumen to tracing against triangle approximations (bounding boxes) in hardware, right?

I like the fact that Digital Foundry addressed the skeptics who doubted the need for hardware RT to even exist, I guess the skeptics are proven wrong now by the performance and fidelity Lumen receives from hardware RT, Lumen now runs much faster, doesn't suffer from screen space artifacts, and can do more precise diffuse and indirect lighting. Hardware RT reflections also comes into play to flesh things out in a performant way.


We are sure it's the Devs saying that it's faster and not their speculation right?


Here he says that Hardware raytracing overall is about 50% slower than software RT while it's more accurate. I assume he didn't mean that (it's actually faster since the work is offloaded to the raytracing hardware) but that the total cost is more.

If so then that's a pretty impressive boost in performance from some months ago and I'd love to hear more about how they achieved it.

Also curious if it's combined with their software solution in some way.
 
We are sure it's the Devs saying that it's faster and not their speculation right?
Yeah, Digital Foundry interviewed the devs and go these info directly from them, also Xbox and PlayStation posted blogs about using ray tracing acceleration, also the devs specifically mentioned that the Coalition team helped shave 0.5ms from the hardware ray tracing cost on Xbox.

Here he says that Hardware raytracing overall is about 50% slower than software RT while it's more accurate
That was an older build, things changed now.
 
Article @ https://www.eurogamer.net/articles/...kens-previewing-the-future-of-gaming-graphics

Inside The Matrix Awakens: a vision for the future of real-time graphics
Digital Foundry talks to Epic about the demo and the evolution of Unreal Engine 5.

Epic: Just like a movie's Director of Photography would place a light card next to an actor for improved lighting in a scene, Epic does this with its characters in The Matrix Awakens. "For example, when Trinity is in the car, and when you know the light doesn't bounce because the seats are a bit too dark, we just add a white card [off-screen]. Lumen analyses the white card and you see that light bouncing back," explains Jerome Platteux.​

I called this back in April and some people were arguing. If you're going to use employ properly modelled light, you're obviously going to have employ the same movie trickery to solve your lighting problems that RT creates.
 
I called this back in April and some people were arguing. If you're going to use employ properly modelled light, you're obviously going to have employ the same movie trickery to solve your lighting problems that RT creates.
Excepting now the tricks are virtual, so can be applied magically to only affect some objects and you don't have to worry about the camera catching a glimpse of the 'card' in a reflection.

But then if you're using such tricks, how is that better than adding a little light-source as existing games do? Is it just a matter of bringing game techniques in line with movies to unify the techniques and design processes?
 
Excepting now the tricks are virtual, so can be applied magically to only affect some objects and you don't have to worry about the camera catching a glimpse of the 'card' in a reflection. But then if you're using such tricks, how is that better than adding a little light-source as existing games do? Is it just a matter of bringing game techniques in line with movies to unify the techniques and design processes?

The tricks not being visible work when the game controls the camera but falls apart when the player has full agency. As for adding light sources, I assume the it's the same reason they always use diffused reflected light in movies. Epic could have added an invisible low light-source here but didn't. ¯\_(ツ)_/¯
 
That was an older build, things changed now.

There was a bit of overconfidence in the early UE5 tech videos that SDF traces would be good enough and hardware RT would just be icing on the cake. It seems that didn’t pan out and triangle RT is needed to get a reasonable result.

Excepting now the tricks are virtual, so can be applied magically to only affect some objects and you don't have to worry about the camera catching a glimpse of the 'card' in a reflection.

But then if you're using such tricks, how is that better than adding a little light-source as existing games do? Is it just a matter of bringing game techniques in line with movies to unify the techniques and design processes?

Those lighting tricks are mostly relevant for cut scenes where the devs are in control of the camera and want to touch up the scene just like in a movie. For regular gameplay the fake lights are placed all over the place for a different reason - to fake GI. That’s no longer necessary. There will still be cases where lights are placed for dramatic effect but just like in real life it should require far less of them.
 
Yeah, Digital Foundry interviewed the devs and go these info directly from them, also Xbox and PlayStation posted blogs about using ray tracing acceleration, also the devs specifically mentioned that the Coalition team helped shave 0.5ms from the hardware ray tracing cost on Xbox.
That was an older build, things changed now.

It seems things have changed, and I'm very impressed by the strides they have made from the EA build, which is why I wanted to be 100% wasn't them extrapolating from what they heard from the devs and was indeed straight from the horse's mouth.

Like I said though, it makes me very curious about the changes that have been made since then and how software tracing now fits into the equation

There was a bit of overconfidence in the early UE5 tech videos that SDF traces would be good enough and hardware RT would just be icing on the cake. It seems that didn’t pan out and triangle RT is needed to get a reasonable result.

For me, it's less about Hardware RT now being fast enough and necessary and more about the fact that it's faster than the software while also more accurate. Seems like you could just throw away the SWRT now. I wonder if this is due to a huge focus on HWRT and SWRT having more room to be optimized. No assumptions, just very curious. But I'll take pretty games over answers.

The tricks not being visible work when the game controls the camera but falls apart when the player has full agency.

We've had fake light sources for ages I'm sure you can sneak in some invisible light bouncers even when the player has control
 
Last edited:
We've had fake light sources for ages I'm sure you can sneak in some invisible light bouncers even when the player has control
I'm sure it's possible, but at what effort? The original discussion was predicated on devs being able to save time once RT could just be dropped in. But in actual fact, they will still be effort will be in faking what is not actually realistic - but without it looking weird.

Does this save dev time? ¯\_(ツ)_/¯. I guess we'll have to wait for the first rounds of GDC talks to hear from them. :yes:
 
what is not actually realistic - but without it looking weird.
MAybe we should start embracing natural light, as in some cinematic approaches (dogma, etc.). I understand sometimes a certain visual (or even sound) vision is what the director wants to achieve, but I'm a bit tired of this dissonance between real life and how we think real life MUST look and be heard in movies, series or even static pictures.

Most realistic fiction to us is based on unnatural lighting, unnatural acting and unnatural voices: people and stuff are not highlighted by white cards in real life, what we understand as professional/good acting many times has nothing to do with how most people act in real life and what we understand as professional sounding voices have nothing to do with how most people talk in real life.
 
I'm sure it's possible, but at what effort? The original discussion was predicated on devs being able to save time once RT could just be dropped in. But in actual fact, they will still be effort will be in faking what is not actually realistic - but without it looking weird.

Does this save dev time? ¯\_(ツ)_/¯. I guess we'll have to wait for the first rounds of GDC talks to hear from them. :yes:
Typically cinematic teams are separate from level/world builders. And often, places where cinematics occur are not often where the real world environment is. Camera angles, animations, lighting, environment are all controlled etc.

I guess this is up to the team to decide how best to do it; but if the goal is reduce workloads based on lighting, world/game lighting is likely where the majority of savings will be as typically it’s about unified lighting and grounding. Making the player believe they are immersed in the world.
 
I wonder if adding an invisible light source is more expensive than using a surface to bounce light in the case where you’re already doing global dynamic gi.
 
I wonder if adding an invisible light source is more expensive than using a surface to bounce light in the case where you’re already doing global dynamic gi.
this probably depends on what you can afford for the situation, I suspect a new light source is much more costly in terms of pure calculation.
 
Status
Not open for further replies.
Back
Top