Tech Focus: Cyberpunk 2077 RT Overdrive - How Is Path Tracing Possible on a Triple-A Game?

preview_player
Показать описание
How far we've come! Path tracing is the ultimate form of ray tracing and somehow, in just a few years, we've gone from a path-traced Quake 2 to a path-traced Cyberpunk 2077! How is this possible? In this Tech Focus video, Alex explains the combination of hardware improvements since the launch of RTX 20 Series, working in combination with radically more advanced software and the rise of AI-generated pixels - everything coming together to push in-game visuals further than they ever have before.

Рекомендации по теме
Комментарии
Автор

I studied Computer Science 25 years ago and back then lightmaps were the big thing. Raytracing was just a dream. Now a game dev i feel this graphics improvement has been amazing. We've been waiting a long time for this quality to finally reach realtime.

aloluk
Автор

Thanks for covering this. Seeing this paradigm shift in real-time rendering is nothing short of amazing.

dominicdibagio
Автор

The fact that we're getting a Triple-A game to implement Path-tracing and be remotely playable (let alone 60 FPS+ abeit with upscaling) is just nothing short of a miracle.

Nekrosmas
Автор

Greatly appreciated to see videos like these! There are so many technologies at work to produce the results you're talking about. I thought I understood what Cyberpunk 2077 was doing with its overdrive mode, but I had not heard about Restir whatsoever until now. I love learning what's going on under the hood!

Doodleschmit
Автор

For someone who doesnt know much about the deeper things of how this technology works, the analysis is just as impressive as the actual showcase! Great job presenting all this, I cant imagine it's easy to condense and organize all this info

Tyler-hseu
Автор

Going from Quake 2 RTX to Cyberpunk 2077 Overdrive in only 4 short years tells me in 10 years path tracing will be close to the norm. That really is pretty fascinating knowing that milestone is achievable in 4-5 more GPU generations.

CaptToilet
Автор

DF: You're probably the ONLY channel on the Tube that does things like this. Others mention and regurgitate what's already been said or known. YOU BRING FRESH Fruits to the table and that's why I subscribe and thumb up you - EVERYTIME - when I watch your video

lil----lil
Автор

Alex, you can turn SER and OMM on and off in Portal RTX from the developer menu. It's possible that it can be switched off in Cyberpunk via a mod. There are multiple mods that deal with ray counts as well.

cpt.tombstone
Автор

Another important factor is how BVH travessing inherently works. It has a logarithmic time complexity, which means that if the number of triangles increases by 1000x, the BVH traversal won't take 1000x as long, it might only increase by 2-3x.

TheGoncas
Автор

at 5:29, SIMD is listed as "Simultaneous Instruction / Multiple Data" but that is incorrect, SIMD stands for single instruction, multiple data, as one instruction is carried out on for example 32 registers in modern GPU architectures.

cpt.tombstone
Автор

Alex's tech focus videos are absolutely underrated.

PlaylistWatching
Автор

I loooove the in depth technical details in this video! Thx Alex and DF for making me and fellow nerds happy! 😊

_PsychoFish_
Автор

I'm glad DF makes high quality videos like these, talking about the many technologies that are used to make path tracing work in a triple A game like Cyberpunk 2077 without cutting much of the technical parts. I might not understand 100% of what was said, but I'm glad I know at least names and overview of techs used

Ferdam
Автор

Great video Alex, I did not expect you to give an introduction to restir and NRC. That was a pleasant surprise. I also did not know the game doesn't use omm. Very exciting times we're in I must say.

bananaboy
Автор

Love these insightful dives into tech. Thank you Alex!

TwinOpinion
Автор

Alex, you're the only one who pronounces the name of CD PROJEKT RED correctly! Respect and greetings from Poland!
Great educational video. Thank you.

wojstube
Автор

This is *fascinating.* Thanks for the lesson, Alex!

joeperez
Автор

Just a small nitpick: at 8:29 ReSTIR stands for Reservoir (based) SpatioTemporal Importance Resampling, hence why it's "Re"servoir based "S"patio"T"emporal "I"mportance "R"esampling, i.e. ReSTIR. Maybe y'all left that out because it's a little more complicated but it makes the acronym make sense.

AndrewPetersonGameDev
Автор

Compared to the views other channels get, this channel is quite possibly very much underrated. I didn't know about and I absolutely regret it. It's the best among the lot of channels which only provide surface facts without providing a proper feel. For eg. I was actually able to visualize how reSTIR works or how increased L2 cache would affect frame latency and fps in case of ray traced games because of how random rays are traced and textures are interpreted per frame. I can't possibly subscribe you more. Would have loved to even pay to support your channel had I not been a student and had a working job.

nikhilchouhan
Автор

Large caches are useful for ray-tracing because of the huge acceleration structures (and geometry) required to be consulted during ray traversal, by the aforementioned incoherent rays, and it is expensive to wait on fetches from VRAM. While multi-core GPUs do a great job at hiding these types of latency, they have their limits.

Caustic (later purchased by Imagination) had a solution for this, which could be implemented in software to improve cache efficiency: batch rays that are travelling in roughly the same direction, and compute all rays in these batches in a go. This works because there is a high likelihood of traversing the same structural nodes and intersecting with the same set of geometry, improving cache hits dramatically.

Another solution is to utilize imposters, reducing the size of both structures and geometry while providing quality high quality results. I believe UE's Lumen does this (as some SDF amalgam) at least in its software mode.

SeanLumly