14. A high-resolution canopy height model of the Earth
Nico Lang et al, Eco Vision Lab, ETHZ
https://arxiv.org/pdf/2204.08322.pdf
https://nlang.users.earthengine.app/view/global-canopy-height-2020
15. "..establish a practical, repeatable,
documented, community-driven and
vetted step away from existing,
disconnected platforms and towards
integration-ready services."
18. A SIMPLE IDEA
TO PHYSICALLY ACCURATELY RENDER X
TO TRAIN AN AI-BASED CONTROL SYSTEM TO DO Y
19. 19
MANUALLY GENERATED TRAINING DATA LIMITS AI CAPABILITIES AND PERFORMANCE
Long Tail Anomalies
Non-Visual Sensors
(Radar, Lidar)
Occlusions
Indirect Features
(speed, direction)
Real dataset that addresses all such challenges in a scalable way doesn’t exist today.
20. 20
UNIVERSAL SCENE DESCRIPTION
Developed by Pixar
Foundation for NVIDIA Omniverse
Open-sourced API and file framework for complex scene
graphs
Easily extensible, simplifies interchange of assets between
industry software
Introduces novel concept of layering
Enables simultaneous collaboration for large teams in
different department working on the same scene
Originated in M&E, now becoming a standard across industries
including AEC, Manufacturing, Product Design, Robotics
The “HTML” of 3D Virtual Worlds
21. 21
ADVANCED TOOLS AND TECHNOLOGIES
Foundational Platform Components
docs.omniverse.nvidia.com
22. • Fourier Forecasting with
adaptive FNO lowers
barrier for uncertainty
quantification.
• Prediction of extreme
weather events.
• 145,000*x faster at 30km
resolution vs. IFS &
25,000x lower carbon/
compute footprint (single run).
• Allows 1000 member
ensembles improving
probability accuracy.
*100 ensembles with lower #variables & vertical levels than IFS
DGX Station (4x A100) vs. 3060 cluster of CPU
Hypothetical 18km res 45,000x speedup.
25. 25
seen unseen
FNO CAPTURES THE ENERGY SPECTRUM
One snapshot
Zero shot super-resolution: train on 64x64, test on 256x256.
Extrapolate to unseen wavenumbers (32->128).
26. 26
OMNIVERSE FOR AN INTUITIVE INTERFACE
Interactive, Intuitive, Real-time, Flexible, Collaborative
27. Stage 1 FourCastNet
Stage 2 Tasking
Stage 3 Coupled GPU+DPU
with NIC for rapid edge compute
onboard plane [S41679]
Geospatial processing [SE1840]
All talks are available offline with free registration
nvidia.com/gtc/sessions/
28. GTC 2022 TALKS
Geospatial image processing [S41754]
High Perf sensors at the Edge with LMCO [S41679]
Digital twins for weather, climate & energy [S41823]
FourcastNet paper: arxiv.org/pdf/2202.11214.pdf
Modulus: developer.nvidia.com/modulus
FNO & Transformers for extreme weather prediction [S41936]
All talks are available offline with free registration
nvidia.com/gtc/sessions/
30. 30
EXPLORE OMNIVERSE ENTERPRISE
SEE YOU IN OMNIVERSE
GET ACCESS TO A FREE TRIAL DEVELOP ON OMNIVERSE
DOCUMENTATION
docs.omniverse.nvidia.com
FORUMS
omniverse.nvidia.com/forums
TUTORIALS AND WEBINARS
omniverse.nvidia.com/tutorials
DISCORD
discord.gg/nvidiaomniverse
31. ANOTHER WAY TO THINK OF DEVELOPMENT
GETTING STARTED GOING FASTER PRODUCTION AI INTEGRATION
Cupy
Numba
cuSpatial
cuGraph
DALI
MatX
nvJPEG
nvJPEG2K
MatX
OpenACC
OpenMP
TensorRT
TAO
Triton
34. 34
APPLIED RESEARCH ACCELERATOR PROGRAM
Use case with deployed
GPU-accelerated
application
Development of GPU-
accelerated
application
Basic Research
conducted by University
Applied Research
project(s)
Program focus
Supports research projects that have the potential to make a real-world impact through
deployment into GPU-accelerated applications adopted by commercial and government
organizations.
Program Benefits
Hardware and funding grants
Technical guidance and support
Grant application support
Hands-on training with the NVIDIA Deep
Learning Institute
Networking and marketing opportunities
Robotics and AI for Automation