NVIDIA at Breakthrough Discuss for Space Exploration

A visual journey through the state-of-the-art capability for AI-enabled space exploration. Today & tomorrow.

1. Open the pod bay doors, HAL…
ALISON B. LOWNDES | SENIOR SCIENTIST | GLOBAL AI

3. NVIDIA
Turbocharges
Science
From weather forecasting and energy exploration,
to computational fluid dynamics and life sciences,
researchers are fusing traditional simulations and
AI to solve the next grand challenges.

4. 4
JOHAN ROCKSTRÖM,
MATTHIAS KLUM
STOCKHOLM

5. “
“
—Me, 2022
“..we already have HAL9000.”

6. INTRODUCING
JETSON AGX ORIN SERIES
▪Up to 275 INT8 TOPS powered by Ampere GPU +DLA
▪Up to 12x A78 ARM CPU
▪Up to 64 GB memory, 204 GB/s
▪Production Modules will be available NOW
▪Get started developing for all Jetson Orin Modules with the
Jetson AGX Orin Dev Kit
Server Class AI Performance at the Edge

7. 7

8. Announcing NVIDIA Blackwell
The Engine of the New Industrial Revolution
DECOMPRESSION ENGINE
800 GB/s
2nd GEN TRANSFORMER ENGINE
FP4/FP6 Tensor Core
AI SUPERCHIP
208B Transistors
5
th
GENERATION NVLINK
Scales to 576 GPUs
RAS ENGINE
100% In-System
Self-Test
SECURE AI
Full Performance
Encryption & TEE
Built to Democratize Trillion-Parameter AI
20 PetaFLOPS of AI performance on a single GPU
4X Training | 30X Inference | 25X Energy Efficiency & TCO
Expanding AI Datacenter Scale to beyond100K GPUs

10. With NVIDIA DRIVE Sim features such as road
elevation, road markings, islands, traffic
signals, signs, and vertical posts are replicated
at centimeter-level accuracy. Autonomous
vehicles can drive millions of miles in a wide
range of simulated scenarios so they hit the
road running, safely.
NVIDIA DRIVE SimTurbocharges
Developer Productivity to Get Self-
Driving Cars on the Road

12. More than 4000 Software Packages in the AI Development Stack
NVIDIA AI Enterprise provides enterprise-grade security and acceleration
DATA PROC
(RAPIDS)
TRAIN
(PyTorch)
OPTIMIZE
(TensorRT)
DEPLOY
(Triton)
NVIDIA CUDA
NVIDIA AI Enterprise
NVIDIA Dependency3rd party & opensource
Security (CVE) Patching
CVE Patching
API Stability Enterprise Support with SLAs

13. TRAINDATA PROC
OPTIMIZE
DEPLOY
Multi-Cloud to On-Prem -Available Anywhere
NVIDIA AI Enterprise
NVIDIA AI Enterprise
Multi Cloud
NVIDIA Certified Server
Coming soon on OCI Marketplace

15. Getting Million-X Speedups to Power AI and Scientific Computing
Accelerated Computing + AI Provides the Compute Required
1980 1990 2000 2010 2020
Single-threaded perf
1.5X per year
1.1X per year
10
2
10
3
10
4
10
5
10
6
10
7
10
9
10
8
10
1
AI
SCALE
UP &OUT
ACCELERATED
COMPUTING
CP
U
GPU
Compute-Intensive
(Parallel) Functions
Application Code
Rest of Sequential
CPU Code
CPU
+

16. Grace CPU

17. Grace CPU Superchip

18. FP8
https://arxiv.org/pdf/2209.05433.pdf

19. A Blending of Physical and Virtual Worlds
THE METAVERSE

22. 23

23. SCENE RECONSTRUCTION: APPEARANCE
●Realism, viewpoint manipulation
NeRF
Mildenhall et al., 2020; Müller et al., 2022

24. 25

25. AUSTRALIA ACCELERATES AI
WITH NVIDIA PARTNERSHIP
In 2022, Australia’s national science and
research organization, CSIRO, and its advisory
arm, Data 61, teamed with NVIDIA to advance
the nation’s AI program on multiple fronts
including climate action, space exploration,
quantum computing, and AI education.
frontierdevelopmentlab.org

26. AI PORTFOLIO
Forecast of magnetic field perturbation
30 mins in advance for the first time…

27. INDUSTRIAL DIGITALIZATIONGENERATIVE AISUSTAINABILITY

28. World Record Route Optimization
NVIDIA cuOpt

29. 30
1-6
•Goal 1: End poverty in all its formsdonate what you don’t use
•Goal 2: Zero Hungerwaste less food & support local farmers
•Goal 3: Healthvaccinate your family
•Goal 4: Educationhelp educate the children in your community
•Goal 5: Gender equality and women’s empowermentempower women & girls and ensure their equal rights
•Goal 6: Water and Sanitationavoid wasting water
7-12
•Goal 7: Energyuse only energy-efficient appliances & light-bulbs
•Goal 8: Economic Growthcreate job opportunities for youth
•Goal 9: Infrastructure, industrializationfund projects that provide basic infrastructure
•Goal 10: Inequalitysupport the marginalized and disadvantaged
•Goal 11: Citiesbike, walk or use public transport
•Goal 12: Sustainable consumption and productionrecycle paper, plastic, glass and aluminium
13-17
•Goal 13: Climate ChangeACT NOW to reduce CO2 emissions
•Goal 14: Oceansreduce/avoid plastics and clean the oceans
•Goal 15: Biodiversity, forests, desertificationplant trees and help protect the environment
•Goal 16: Peace, justice and strong institutionsstand up for human rights
•Goal 17: Partnershipslobby your government to boost development financing

30. 31
SAILDRONE
Startup Charts Autonomous Oceanic Monitoring | NVIDIA Blog

31. 32

33. Orbital Sidekick’s GHOSt2
fleet now in orbit

34. Transmitter 6G Channel Receiver
SIONNA
SDK for 6G Link-Level Simulations
developer.nvidia.com/sionna

36. Earth-2 Platform
Connecting Complex Simulation, Data and AI Workflows
Renderer
Modulus
Omniverse
Training Inference
Interactive
Visualisation
GPU
acceleratd
Simulation
AI Enterprise

37. Production Deployment through NVIDIA Inference Microservices
Deploy Customised Models leveraging Optimised Inference Engines
Click this linkto check out the browser demo!
Prebuilt container
and Helm chart
Industry standard
API
Optimised inference
engines
Support for custom
models
NVIDIA AI Enterprise
Domain specific
code

38. Resources
•More resources
•Earth-2 Solutions page
•Modulus product page
•Call to action
•For developers building Climate tech applications, please
reach out here for early access to cloud services.
•For Enterprises interested in Earth-2 platform stack, please
reach out to your Nvidia account rep
•For researchers in research and science collaboration –
Engage with us on our open-source repository

39. Niklas
Roeber
(Jan 2023)

40. OPTION 2
WORLD SIMULATION IS THE ULTIMATE COMPUTER SCIENCE CHALLENGE
VIRTUAL WORLD
Rendering

41. OPTION 2
WORLD SIMULATION IS THE ULTIMATE COMPUTER SCIENCE CHALLENGE
VIRTUAL WORLD
Physics
Rendering

42. WORLD SIMULATION IS THE ULTIMATE COMPUTER SCIENCE CHALLENGE
VIRTUAL WORLD
Rendering
Physics
AI

43. SUPERCOMPUTING BRINGS WORLD SIMULATION TO AN INFLECTION POINT
VIRTUAL WORLD
Rendering
Physic
s
AI

44. Sensors
Controllers
Humans
PHYSICAL WORLD
Real-Time Data
Synchronization
Control Signal
DIGITAL TWINS ARE A NEW CLASS OF WORLD SIMULATION
VIRTUAL WORLD
Rendering
Physic
s
AI

45. 46
Watch
Here

47. 48
48

48. 49
roboCasa.ai

49. 50
MIMICGEN
https://mimicgen.github.io/

50. 51
NVIDIA Isaac
Development platform for building AI robots
PLATFORM
NVIDIA Isaac on Jetson
Isaac Sim in Omniverse
NVIDIA Isaac AI
For Autonomous Mobile Robots
For Industrial Manipulation Arms
For Humanoids

51. 52
NVIDIA ROBOTICS
RESEARCH LABS
Drive breakthrough robotics
research to enable the next-
generation of robots that safely
work alongside humans,
transforming industries such as
manufacturing, logistics, healthcare,
and more

52. 53
EXPLORE OMNIVERSE ENTERPRISE
SEE YOU IN OMNIVERSE
DOWNLOAD OMNIVERSE
nvidia.com/omniverse
DEVELOP ON OMNIVERSE
DOCUMENTATION
docs.omniverse.nvidia.com
FORUMS
omniverse.nvidia.com/forums
TUTORIALS AND
WEBINARS
omniverse.nvidia.com/tutorials
DISCORD
discord.gg/nvidiaomniverse

53. Industrial Metaverse Opportunities Exist at Every Scale
Every physical process will be software-defined
Product
Factory
City
Nation
Planetary
Galaxy

54. Anatomy of a Sensor Processing Workflow
A Vision of Towards Self Driving Instruments and a Science Programmable Edge
Edge GPU (xN)
Sensor
Data aggregation
and reduction,
domain specific
transforms
Pre-Processing /
Data Filtering
Apply pre-trained AI
model to data
stream to drive
real-time analytics
and experiment
AI Inference
Apply domain
specific data
transformations
Post-Processing
Show AI-augmented
results to
operator/user
Visualization
Real Time Insights and Automatic Sensor Control
Analysis &
Storage
Digital Twin /
Simulation
AI Training
Edge
Datacentre
Cloud
Scale, Resiliency, and Sensor
Fusion

55. NVIDIA Holoscan
SDK for Building AI-Enabled Sensor Processing Applications
▪C++ and Python APIs for
domain agnostic sensor data
processing workflows
▪Multi-Node and Multi-GPU
support with advanced
pipeline scheduling options
and network-aware data
movement
▪AI Inference with pluggable
backends such as ONNX,
Torchscript, and TensorRT
▪Scalable from Jetson Orin
Nano (ARM + GPU) to DGX
(x86 + H100)
▪Apache 2 Licensed and
Available on GitHub
Features
▪Simplifies sensor I/O to
GPU
▪Simplifies the performant
deployment of an AImodel
in a streaming pipeline
▪Provides customizable,
reusable, and flexible
components to build and
deploy GPU-accelerated
algorithms
▪Scale workloads with
Holoscan’s distributed
computing features
▪Deploy to the Cloud with
Holoscan App Packager and
Kubernetes
Benefits
GitHub

56. Holohub
Sample Holoscan Applications, Tutorials,
and Helpful Operators
IO:UDP Ethernet, Lidar, High Speed Cameras, SDR, SAR, 3D Volumes
AI:Segmentation, Tracking, AR/VR, LLMs
Tools/Tutorial: MATLAB, MONAI, Playground on AWS, Scaling Apps
GitHub

57. Deep Space Demonstration
Allen Telescope Array Holoscan Pipelinewith CyberEther
CyberEthe
rGitHub

58. Allen Telescope Array
Hat Creek, CA

59. Getting Started with Holoscan
Holoscan References
https://github.com/nvidia-holoscan/holoscan-sdk
docker pull nvcr.io/nvidia/clara-holoscan/holoscan:v2.2.0-dgpu
pip install holoscan
https://docs.nvidia.com/clara-holoscan/sdk-user-guide/index.html
Debian Packages available on NGC

60. Holoscan Bootcamp and Open Hackathons
Half Day Free Tutorial –September 24
th
, 2024 –1pm –5pm Eastern
Register Here

61. Free AI Technical Training
A Few Clicks Away
Elevate your skills with free AI technical training. For a limited
time, we are offering a complimentary self-paced training course
to new members of our NVIDIA Developer Program.
You can select one of the 16 courses like:
•Getting Started with Deep Learning
•Deploying a Model for Inference at Production Scale
•Fundamentals of Accelerated Computing with CUDA Python
•Synthetic Data Generation for Training Computer Vision Models
•Get Started with Highly Accurate Custom ASR for Speech AI
•Optimizing CUDA Machine Learning Codes with Nsight Profiling
Tools
Seize this free opportunity to embrace the future of technology
at your own pace. Join the Developer Programnow and claim your
course today.
https://nvda.ws/3ugFbbo
Join The NVIDIA Developer Program

62. FREE TO
ALL
https://nvda.ws/3MGcObb

63. NGC CATALOG
The Hub of GPU-Optimized Software
ACCELERATES DEVELOPMENT
Focus on building, not setup
FULLYTRANSPARENT
Quickly identify and deploy the right software
PERFORMANCE OPTIMIZED
Tested across GPU-accelerated Platforms
SOTA models Model resumes Develop once. Deploy anywhere w/ NVIDIA VMI
Detailed security scan reports One click deploy from NGCMonthly swcontainer updates
0.9
1.4
1.9
Jun '20Nov '20Nov '21
Training Speedup
Faster Training on the Same
Stack*
*8x NVIDIA A100 40GB. NVIDIA DGX. Resnet-50. Mixed Precision. 256 Batch size.
https://catalog.ngc.nvidia.com/resources

64. 65

65. 66
Any questions?
[email protected]