I love this stuff! You guys should do more series documenting the evolution of specific rendering techniques/attempts at certain kinds of effects, sorta like the H2O video John did a while back.
I used to be a student of the professor that invented SVOGI. Wild to think that it'd be implemented in video games through ray tracing just a few years later.
It’s still wild to me that real-time ray tracing exists in my (younger) life time. I thought I would be ranting to my grandkids “back in my day we had rasterized lighting and we liked it! None of this ray-traced 5000 fps retina projected images y’all have! Kinds these days are soft.” “Mom! Grandpas ranting about having to walk up hill both ways in the snow to pay 2 times MSRP for a used 1080 Ti again.” “Calm down dad or you’ll spike your blood pressure again.”
@@TheCrewExpendableThis is due to the use of complex algorithms such as ReSTIR GI. Roughly speaking, light sources are assigned a weight and the few rays are concentrated mainly on the areas that have the greatest weight, i.e. are the brightest and most promising. In addition, spatial and temporal information is collected and reused. This is the reason for the noise and ghosting, but is several hundered times faster. In the traditional Monte Carlo method used in online rendering, the rays are distributed randomly and evenly across the scene. This is an unbiased method and is exorbitantly more expensive, but of course produces increasingly better results the more samples and bounces are used with diminishing returns of course.
Incredible work as always Alex! Even as an industry tech guy, I still learn so much from your videos. Keep being awesome! (Now to look into the viability of converting Turok PC over to a Path Tracer...)
Sonic Unleashed is one of my favorite examples of games that used baked GI but incorporated one of the first examples of light probes to generate a GI look in dynamic objects
Global illumination is the most important part of modern game graphics in my opinion. Such a drastic change for people that can recognize fake/real lighting
The problem is that we've gotten so good at avoiding the problematic situations for traditional techniques that you really need to know the tech to understand when something's impressive.
@@SimonBuchanNz Yeah devs have gotten so good at 'faking it' that normies often don't notice. Even for a nerd like me, sometimes the difference is so slight I'd choose rasterised over RT just to get the extra performance. Depends on the game though, of course. I guess it will save loads of dev time though, and like all new tech the devs need to learn how to use it to get it looking its best. Plus, we're in a transitionary stage right now... One day, real-time ray-traced GI and reflections will become standard, like pixel shaders and normal maps are today. Will need hefty GPUs though!
We're only good at faking it in a very specific subset of games, typically games that can get away with baking most if not all of their lighting. In more dynamic games, many of the techniques that get close to full raytracing just aren't feasible due to the dynamic nature of the games, and so you're left with alternative techniques that just don't cut it, for one reason or another. GI especially is a hard one, because the only alternatives are generally slow (slower than even full raytracing, when you try to fit more geometry detail into them), and support a limited amount of bounces (1, 2 if you're lucky).
Lighting is the crucial baseline for human vision to perceive color, dimension, and depth. One of the reasons we're easily impressed by viewing a game scene with plenty of realistically arranged well-baked lightmaps, despite being static diffused lighting. Or why more stylized art direction like R&C Rift Apart can still feel incredibly sophisticated and immersive - our minds can suspend disbelief of the stylized presentation, yet still be dazzled by impressive lighting and ray-traced surfaces, as it communicates significant information and 'sense of place and time' to human perception of our surroundings.
Coming from an environment artist working in dem vidjagaems, I just want to let you know how much I appreciate the research and attention to detail that you clearly put into the video. Great job!
We (I know it's a small %) certainly respect & appreciate all the work & attention to details game devs put into games. And I obviously like watching videos like these as well which point out such details
I'm surprised there was no mention of SSGI, either the UE4 implementation found in Gears 5 and Days Gone or Pascal Gilcher's Reshade implementation. It's that missing link between ray tracing and rasterisation that perfectly demonstrates the limitations of the latter.
How does it perfectly demonstrate the limitations of rasterization? RTGI itself is already a perfect example of that and SSGI is just a downgraded version of that, and thus doesn't show the difference as much. It's still a pretty nice thing to have and is a decent solution to add on top of existing lighting if full RTGI is too demanding.
@@okaybutwhythough7456 SSGI is a post process and can only work on the limited information available in the framebuffer. It exhibits artefacts due to missing information from surfaces being off-screen or occluded causing the indirect lighting it produces to change radically as different pixels become visible, even from relatively small camera movements. It's really a very poor and unstable approximation of GI compared to any other method.
@@okaybutwhythough7456 rasterisation works by sampling the scene at each visible pixel on each visible triangle. At the time of shading it doesn't know anything about any other points on any other surfaces, let alone how much light they might be reflecting toward it. To approximate GI you therefore have to sample back a pre-lit representation of the scene passed to the shader. SSGI makes a good example of this limitation because it traces/marches rays through a representation of the scene, similar to RTGI, but is limited by that representation having been rendered into a 2D image (with depth information). It's that lost information about the scene that is the key difference between rasterisation and ray tracing.
I can't wait until the day when RTX will have so many rays possible that we'll be able to do the same kind of full scale path tracing that is even rare in CGI movies today due to how time consuming it is.
I am not sure it'll be an optimal use of resources, because a lot of the lighting can be "faked" pretty well with rasterization-based techniques. Having *fully* realtime global illumination (not just indirect lighting) is very exciting, but it could also be distracting in games and make it harder to play. Quake 2 RTX has some of that, where the lighting becomes too dynamic and the game world can go from pitch black to blindingly bright very quickly during frantic fights in tight spaces and it can have adverse effects on the gameplay IMO. Modern games use raytracing mostly to augment the rasterized image, not to replace it completely as that'd be unfeasible but also needlessly expansive even if possible - as Quake 2 RTX demonstrates IMO. I guess it is possible that the raytracing techniques and hardware will improve with time bringing both manufacturing and running cost down to the point where it becomes feasible to make raytracing-only games, and dropping a traditional rasterized lighting as a fallback. It'd be interesting to see what game designs could emerge from that, especially when the indie scene gets their hands on this technology.
@@unfa00 my point was that eventually, there will be no reason for those shortcuts, the power will be so good that it would be a waste not to go full multi bounce GI pathtracing all the time.
Man... imagine still being optimistic for a "better future" 😂 it's a race to the bottom dude IDK if you've been paying attention but success isn't achieved by those who work the hardest or push their medium to its limits. Success in 2023 comes from exploiting existing intellectual property for the lowest possible cost & selling that hackjob product to the massive audience of frachise fans (who have no standards or reservations about buy & playing a badly made game that's based on their favorite IP) LOL, wait... did you actually think anyone involved at the leadership level actually cares about making good games? Dude, almost all of those people retired/left the industry years ago. It's all about licensing existing assets & crunching disposable workers to make those licensed assets into a facsimile of a "game" with a "story" 🤣 "Imagine how good future games will look..." No. I don't think I will engage in this pointless thought experiment.
@@unfa00 Light baking is very resource and developement time intensive and will go away and be replaced by ray tracing as soon as the average gamer has hardware powerful enough to run that at normal framerate.
One small correction, many games with only direct lighting don’t have pitch black shadows (unless we’re counting ambient light color). Shadow stencils in Doom 3 have that effect but shadow mapping is a second pass and the shadows generated there are overlaid on top of a color pass that’s already been made, and the devs get to decide how transparent those shadows are. Shadow mapping still uses only direct lighting though.
I absolutely love this. Would love even more detailed breakdowns of complex topics in lighting, materials (PBR in particular), evolutions in modeling, etc. Great work!
This is incredible work, Alex. I need to watch it like three more times to retain it all, but if I was a teacher I would proudly show this to my class. Well done!
I really enjoyed this. I’d highly recommend using Cycles and Eevee in Blender to showcase these effects, would really allow precise demonstration that some of the diagrams can’t easily convey. Especially things like screen space effects which were glossed over a bit.
isn't screen space basically cube mapping ? Edit, oh, no I was thinking about something else. Yes, screen space reflections are an entire different thing using a specific screen buffer to provide the colour data to other objects in the scene but mostly for reflections.
@@TheSliderW no, it’s basically in a very basic sense copy pasting what’s on the screen, that’s why the trees disappear when you look down at the water in a game like far cry. A cube map isn’t limited by screen space.
Studying for college was never made this easy. Thank you for helping me on my exam. I never thought this channel, that I watch ever since I was a kid back in 2014 would help me in college. Thank you so so much from Portugal, DF.
I think these are very important videos in the context of this channel. I think you guys overestimate the amount of people that actually understand what you talk about in your videos! I genuinely believe a large majority of your viewers are very casual with not much deep technical knowledge, who may not even understand any of the terms or techniques you discuss... so these videos are important ways to bring those viewers in even more.
This was fantastic! I'd love to see a similar Tech focus episode looking at the evolution of shaders and the hardware they run on, from pre DX9c fixed function HW with non-programable shaders, through the DX10 era with the G80/HD series cards up to modern DX12 stuff :)
It was a bit ambiguous to explain fill color after light maps even though fill color is the oldest technique in the list (predating even the direct illumination) but very well made video! Fill color is basically act drawing things as they are without any lightning calculations.
This channel is very unique. It's hard to find a channel that gives you all the informations to understand complex topics, in a understandable way. Your work helped me to find the inspiration for my master thesis in computer graphics. Thank you (sorry for my English)
Holy shit, this is one of the best videos you guys have done, also special thanks to Alex for his in depth presentation it really seemed like a single take :o Auf Wiedersehn!
Love these videos not just for the content itself but the comment section is usually more chill with the more tech nerd types coming out and the gaming fanboys mostly gone.
Lumen is not as precise but runs a lot better and doesn't require hardware acceleration. Pretty much sums it up. It just uses and combines several different forms of tracing instead of triangle tracing like RTGI.
I may be a minority here but I generally only really understand about 30% of the topics covered by you guys at DF. However, the way you all present the information and content is done is such a great way that, regardless of how much I understand, i find it utterly fascinating and genuinely look forward to every video you guys make. Thanks
Wow. I didn't knew that indirect lighting in last of us 2 is so complex. Also I didn't knew that these techniques were discovered/invented years, no decades ago! Kudos to genius developers! Fantastic video! Thank You!
amazing video alex, been watching DF for a long time, and this demystified a ton of visual effects my eyes could see but my brain couldn't comprehend. great, great work.
Excellent video. It is much needed that some techinical stuff is explained in detail like this, so people can understand to appreciate or criticize the implementation used in their favorite games.
That was really informative and explained a ton of things I noticed from years of playing games but never bothered to research or try to understand. Fantastic video
I can watch these videos all day. So good and educational. Btw, wtf, the top 2 comments are from 1 month ago while the video was just released 1 day ago.
Know what'd be really fun to see? A ray-traced Doom 3. Substantially better quality models than Quake 2 with actual bumpmaps and such, but also it's a game with the harshest direct lighting in existence and global illumination would fix one of two primary complaints with the game (how goddamn black it is).
It'd easily be the most impressive way to show off what global illumination can do, especially as it was designed to have many moving light sources. Most other games since the late 90s at least attempted to approximate GI in some way, while Doom 3 just went "lolno, pitch-black ambient is all you get"
I understand the appeal but as TonkarzOfSolSystem said, the game would completely break as it'd lose its main feature: being afraid of what can come out of...pitch darkness. What could be interesting tho is raytraced shadows to get rid of those horrible shadow edges !
We'll eventually see all games migrate to it as sufficient raytracing performance is available on lower and lower end cards and as the market share of non-raytracing cards shrinks. Sort of like how Doom 3 more or less sounded the death knell of DX7-class cards (though the popularity of the GF2-in-disguise GF4 MX required an exception). Metro Exodus might be the first game to require it (though that's actually debatable since the "non-enhanced" edition exists as the fallback), but there will be more, at an increasing rate, until all games do.
@@sushimshah2896 Technically yes but there are games with 60fps ray traced modes on console you just gotta make some smart optimizations to make it possible.
@@theninjamaster67 consoles most likely will see RTGI along with reflections and shadows in at least 3-4 years when RT techniques evolve enough to run on the weaker RT hardware. Granted the Xbox has the potential to do it earlier if DirectML delivers something close to DLSS.
Pretty cool! Next can we have a latency explanation when online gaming? Such as: net code, ping, server tick rate, and refresh rate vs frame timing. I think we'll all eat that up😏
Lovely video! Those diagrams perfectly illustrate what each technique does. I remember learning a lot about them when mapping for both the Source 1 and Source 2 engines...
Would love to see more videos like this from the channel - very informative and approachable. Love the examples given, really showcased the points. Thanks for all the hard work!
They've covered SSGI in several videos. VXGI is Voxel Cone Tracing and as far as I know they haven't covered it, but they have covered similar solutions like CryEngine's SVOGI and Lumen, which partially uses voxels. It wasn't a full on in-depth coverage but it was still pretty decent. In short SSGI is pretty limited while VXGI/SVOGI are not as precise as full RTGI.
we had fluid simulation for years, they just dont implement on games because it is too taxative on consoles cpu, now smoke simulation is something im more interested.
