Modern video games employ a variety of sophisticated algorithms to produce groundbreaking
3D rendering pushing the visual boundaries and interactive experience of rich
environments. This course brings state-of-the-art and production-proven
rendering techniques for fast, interactive rendering of complex and engaging
virtual worlds of video games.
This year the course includes speakers from
the makers of several innovative games and game engines, such as Activision, EA
| Frostbite, Tencent Games, Intel, and even independent engine developers, for
the first time in the Advances’ history! The course will cover a wide range of
topics, this year diving deeply into improvements for GPU-driven pipeline
rendering with visibility buffer and related techniques, and mobile-friendly
cluster rendering; delve into the design of shader language for AAA games; and cover
global illumination topics with practical innovations for runtime surfels-based
GI, as well as a neural network-based GI, and advances in hemispherical
lighting. We will also explore improvements for dense geometry rendering, and
more.
This is the course to attend if you are in
the game development industry or want to learn the latest and greatest
techniques in the real-time rendering domain!
Welcome and Introduction to Part I
Natalya Tatarchuk (Activision)
Neural Light Grid: Modernizing Irradiance
Volumes with Machine Learning
Michał Iwanicki (Activision)
Seamless Rendering on Mobile: The Magic of Adaptive
LOD Pipeline
Shun Cao (Tencent Games)
Flexible and Extensible Shader Authoring in
Frostbite with Serac
Simon Taylor (EA | Frostbite)
Closing Notes for Part I
Natalya Tatarchuk (Activision)
Welcome and Introduction to Part II
Natalya Tatarchuk (Activision)
Announcing The Call
of Duty Open-Source USD Caldera Data Set
Michael Vance (Activision)
Variable Rate Shading with Visibility Buffer Rendering
John Hable (Visible Threshold)
Shipping Dynamic Global Illumination in Frostbite
Diede Apers (EA | Frostbite)
Hemispherical Lighting Insights from the Call of
Duty Production Lessons
Thomas Roughton (Activision)
Achieving scalable performances for large scale
components with CBTs
Anis Benyoub (Intel), Jonathan Dupuy
(Intel)
Closing Notes for Advances in Real-Time Rendering in Games, 2024
Natalya Tatarchuk (Activision)
Working
knowledge of modern real-time graphics APIs like DirectX or Vulkan or Metal and
a solid basis in commonly used graphics algorithms. Familiarity with the
concepts of programmable shading and shading languages. Familiarity with
shipping gaming consoles hardware and software capabilities is a plus but not
required.
Technical
practitioners and developers of graphics engines for visualization, games, or
effects rendering who are interested in interactive rendering.
Natalya Tatarchuk (@mirror2mask) is a
graphics engineer and a rendering enthusiast at heart, currently serving as
Chief Technology Officer at Activision Publishing where she’s helping drive
innovative technology powering award-winning franchises like Call of Duty at Activision / Microsoft.
Prior to her current position, she drove the state-of-the-art rendering
technology, graphics performance and character content creation in her role as a
Distinguished Technical Fellow and Chief Architect, VP, Wētā Tools at
Unity. Before leading the Graphics team at Unity as VP of Graphics for the
Unity Editor and Engine, she was a AAA games developer, working on innovative
cross-platform rendering engine and game graphics for Bungie’s Destiny franchise, as well the Halo series, such as Halo: ODST and Halo: Reach, and AMD Graphics Products Group where she pushed
parallel computing boundaries investigating advanced real-time graphics
techniques, and graphics hardware design and APIs. Natalya has been encouraging
sharing in the games graphics community for several decades, largely by
organizing a popular series of courses such as Advances in Real-time Rendering,
Open Problems in
Real-Time Rendering and
Rendering Engine
Architecture, and convincing people to speak there.
It seems to be working.
|
|
Welcome and Introduction – Trends in Games
and Rendering
Abstract: This talk provides the context behind the
history of Advances in Real-Time Rendering in Games course since its inception,
as well as explaining the goals of what this session aims to achieve. The
speaker further delves into the analysis of current trends scene in gamers,
player perspectives for video games, and then explores what that means for
trends for gaming technology in rendering and related areas, as well as
platforms, and beyond.
Speaker Bio:
Natalya Tatarchuk (@mirror2mask) is a
graphics engineer and a rendering enthusiast at heart, currently serving as
Chief Technology Officer at Activision Publishing where she’s helping drive
innovative technology powering award-winning franchises like Call of Duty at Activision / Microsoft.
Prior to her current position, she drove the state-of-the-art rendering
technology, graphics performance and character content creation in her role as a
Distinguished Technical Fellow and Chief Architect, VP, Wētā Tools at
Unity. Before leading the Graphics team at Unity as VP of Graphics for the
Unity Editor and Engine, she was a AAA games developer, working on innovative
cross-platform rendering engine and game graphics for Bungie’s Destiny franchise, as well the Halo series, such as Halo: ODST and Halo: Reach, and AMD Graphics Products Group where she pushed
parallel computing boundaries investigating advanced real-time graphics
techniques, and graphics hardware design and APIs. Natalya has been encouraging
sharing in the games graphics community for several decades, largely by
organizing a popular series of courses such as Advances in Real-time Rendering,
Open Problems in
Real-Time Rendering and
Rendering Engine
Architecture, and convincing people to speak there.
It seems to be working.
Materials (Updated August 5th,
2024): PDF
(Slides + Notes, 5MB)
Neural Light
Grid: Modernizing Irradiance Volumes with Machine Learning
Abstract: Irradiance volumes and their variants
have been used in video games for over two decades. While more advanced,
real-time global illumination systems have been presented, they are only
practical on most high-end hardware. The diversity of hardware used by players,
particularly low-end mobile GPUs, still creates a need for a robust,
high-quality solution that can be universally used across all supported
platforms, without requiring to re-author the lighting setup. Despite the
popularity, the basic problems associated with irradiance volumes have not been
fully mitigated: to avoid leaking artifacts various hand-tuned heuristics are
used, but they are often insufficient and require manual fixes.
This talk will present how modern machine
learning methods can be used to eliminate these artifacts, while keeping the
performance of the algorithm acceptable even on very constrained platforms. We
will show how we took the idea of utilizing machine learning to improve
precomputed lighting through multiple experiments, reshaped it, reframed it,
reformulated it, to finally form a solution that shipped to multiple millions
of players of the Call of Duty: Warzone
and Call of Duty: Modern Warfare 3.
Speaker Bio:
Michał Iwanicki has been working
in the game industry for over 20 years. He started in CD Projekt
RED, working on the first installments of "The Witcher" series as a rendering and engine programmer. At
Lionhead he contributed to the engine technology for "Milo&Kate" project. As a
part of the rendering team at Naughty Dog, he worked on "Uncharted 3", "The Last of Us" and "Uncharted 4". Since 2014, with a
brief break to help as the CTO of CD Projekt RED,
he's been at Activision Central Technology team, focusing on precomputed
lighting, low level systems, CPU and GPU optimizations and various hardware
projects. He presented some of his previous work at GDC, SIGGRAPH, EGSR. He's
one of the co-authors of the 4th edition of the "Real-Time Rendering"
book.
Materials (Updated August 5th,
2024): PPTX
(151 MB), PDF (Slides
+ Notes, 6MB)
Seamless Rendering on Mobile: The Magic of
Adaptive LOD Pipeline
Abstract: In this talk, the authors will share a
seamless and adaptive rendering solution for GPU-driven rendering on
mobile building on previous approaches, ranging from Nanite, to other
algorithms. Their method delivers an integrated lightweight LOD generation
with low I/O overhead and high-performance rendering on mobile
platforms. The presenters’ approach gives fine granular control in all aspects
of performance and quality. The LOD techniques provided as part of this
system reduce the complexity of LOD management for game pipelines by removing
the necessity of manual LOD setup.
Speaker Bio:
Shun Cao is an expert engineer and tech lead at the
R&D center of Tencent Games. He has a broad range of research interests
with a particular focus on global illumination and animation simulation
systems. In the past few years, he has been leading the development of various
GI solutions, such as the distributed offline light baking tool Dawn and real-time mobile GI solution SmartGI, which are widely used by commercial projects at
Tencent Games.
Materials (Updated August 10th,
2024): PDF (2.6
MB)
Flexible and Extensible Shader Authoring in
Frostbite with Serac
Abstract: Frostbite, like any large game engine,
includes a lot of complex features and systems, as well as mechanisms to
customize and extend those systems. This poses a particular problem for shader
code since modern shading languages prioritize performance over flexibility and
extensibility. Traditional tools used to provide this flexibility, such as
callbacks and virtual functions, are thus not available – and would likely not
be performant if they were.
Serac is a domain-specific language
“wrapper” around HLSL that we built to address this problem without sacrificing
performance. We will give an overview of what Serac is, how it fits into the
broader shader authoring workflows in Frostbite, and what differentiates it
from other similar approaches. We will describe how the language itself is
implemented and how code is combined to form a final shader. Finally, we will
go over what we learnt rolling this system out across the engine and titles
using it, what went well and what we would do differently.
Speaker Bio:
Simon Taylor is a Senior Software
Engineer at EA Frostbite, where he currently focuses on shader authoring and
workflows. He joined Frostbite in 2016 where he has helped ship multiple
titles, including those from the Battlefield,
Battlefront, Mass Effect and FIFA franchises.
A rendering engineer since 2004, he previously worked on Enlighten at Geomerics, and APB at Real Time Worlds.
Materials (Updated August 2, 2024): PPTX
(78.3MB)
Announcing The Call of Duty
Open-Source USD Caldera Data Set
Abstract: We present
the Call of Duty Caldera Open-Source USD Data Set, a new geometric data
set release of production assets from Call of Duty: Warzone, licensed for
academic research and non-commercial use. We will discuss the details of the
data set and our hopes for how it will spur interesting solutions to novel
problems in the environmental geometry problem space. This data set is provided
in the OpenUSD format, and also
represents one of the most geometrically complex data sets available in
OpenUSD.
Speaker Bio:
Michael Vance (Mastodon) is a Senior
Vice-President and Fellow Software Engineer at Activision. He received his B.Sc
in Computer Science in 1999 from Pennsylvania State University. His first job
in the industry was at the start-up Loki Software, and he later began working
at Treyarch before its acquisition by Activision. As a technical director he
has led engineering on the Spider-Man series and contributed to many
other titles including all of the Call of Duty games from 2011’s Modern
Warfare 3 to present. He resides in Falmouth, ME, where he plays at farming
when not volunteering for the Falmouth Land Trust.
Materials (Updated August 10th,
2024):
·
Activision
Releases Call of Duty®: Warzone Caldera Data Set for Academic Use blog,
·
Download the Call of Duty Data Set
Now
Variable Rate Shading with Visibility
Buffer Rendering
Abstract: Visibility
Buffer rendering is an alternative approach to real-time rendering, with some
very different tradeoffs compared to GBuffer and
Forward rendering. But the most interesting property of Visibility Buffer
rendering is that the shading rate is completely decoupled from the native
rendering resolution. This talk discusses a method of reducing the number of
pixels shaded while still maintaining visual fidelity, as well as discussing
the advantages and disadvantages of the approach.
Speaker Bio:
John Hable is a rendering programmer who has worked at Electronic Arts, Naughty
Dog, Epic Games, and Unity.
Materials (Updated August 5th,
2024): PPTX
(35 MB)
Shipping Dynamic Global Illumination in
Frostbite
Abstract: Global Illumination Based on Surfels (GIBS)
is Frostbite's dynamic indirect diffuse lighting system, developed in
collaboration with EA SEED. We presented a thorough technical description of GIBS at Advances
in 2021. Since then, GIBS has shipped with College Football 25, and is being used in an upcoming open world
action game. This talk will cover both game’s requirements and budgets imposed
on GIBS, and discuss several practical learnings, improvements, and
optimizations that were crucial to achieve truly dynamic global illumination at
60fps on current generation consoles.
Speaker Bio:
Diede Apers is a Rendering Engineer at Electronic Arts in Stockholm. After
graduating from Breda University of Applied Sciences in 2016, he joined EA to
work on the Frostbite engine. Since the introduction of hardware-accelerated
ray tracing, he has been involved in all aspects of its adoption in Frostbite.
His work spans from the foundational infrastructure to the development of
multiple shipped ray tracing features. He co-authored an article in Ray Tracing
Gems in 2019, and presented Ray Traced Ambient Occlusion at GTC in 2021. His
contributions have directly impacted games such as: Plants vs. Zombies: Battle for Neighborville,
Battlefield 2042, Dead Space (2023), College Football 25, and more.
Materials (Updated August 10th,
2024): PDF (7 MB)
Hemispherical Lighting Insights from the Call of Duty Production Lessons
Abstract: Activision’s titles such as Call of Duty make heavy use of
precomputed lighting in order to deliver great visual
fidelity while maintaining jitter-free competitive first-person shooter
framerates. Precomputed lighting is mainly applied to surface hemispheres,
which require special reasoning to achieve the best quality for normal mapped
surfaces. In this talk, we discuss the importance of hemispherical occlusion,
and extend that to show how high-quality occlusion can be efficiently computed
from visibility cones, significantly improving the appearance of runtime AO
with normal mapped surfaces. We additionally introduce our new, efficient HHD
(Hemisphere and Highlight Direction) model for lightmaps that enables
higher-quality runtime blending than the Ambient and Highlight Direction (AHD)
model and will touch on how to correctly and generally solve for lightmaps that
preserve IrradZ, the irradiance in the hemisphere
normal.
Speaker Bio:
Thomas Roughton
is a graphics engineer at Activision
Central Tech, working on rendering techniques and performance with a particular
focus on the baked lighting pipeline.
Previously, Thomas led development of the renderer at LiveSurface;
this wide-ranging work included core engine design and development, rendering
techniques to plausibly blend many layers of custom user artwork or heightmaps
with real photos, and implementation of the content pipeline for internally
authored 2.5D or 3D scenes.
Materials (Updated August 5th,
2024): PDF
(48 MB)
Achieving scalable performances for large
scale components with CBTs
Abstract: A concurrent binary tree (CBT) is a
GPU-friendly data-structure suitable for the generation of bisection-based
terrain tessellations, i.e., adaptive triangulations over square domains. In
this talk, we will expand the benefits of this data structure in two respects.
First, we show how to bring bisection-based tessellations to arbitrary polygon
meshes rather than just square heightfields. Second, we alleviate a limitation
that restricted the triangulations to low subdivision levels. We do so by using
the CBT as a memory pool manager rather than an implicit encoding of the
triangulation as done originally. We demonstrate the benefits of our
improvements by rendering planetary scale geometry out of very coarse meshes.
Performance-wise, our triangulation method evaluates in less than 0.2ms on PlayStation®5-level hardware.
Speaker Bios:
Jonathan Dupuy is a senior
research scientist working at Intel Corporation. His research interests are primarily
oriented towards high quality real-time rendering. This encompasses a wide
range of topics including antialiasing, level-of-detail, analytic models for
both materials and lighting, and GPU/parallel programming.
Anis Benyoub is a graphics
research engineer at Intel Corporation. He is passionate about Monte Carlo
integration, physically based rendering, and real-time performance. Before
Intel, he worked at Unity Technologies in the High-Definition Render Pipeline
team. He holds an M. Sc. in Computer Science from Ecole Polytechnique de
Montréal and M. Eng degrees in Computer Science from INSA Lyon.
Materials (Updated August 5th,
2024): PDF
(12.5 MB)
Direct
contact:
Follow @mirror2mask
Mastodon @mirror2mask