All language subtitles for NVIDIA Changed Generative AI Forever (SIGGRAPH 2023 Supercut).en

it all started over 50 years ago with a

it all started over 50 years ago with a
simple question

simple question

simple question
[Music]

[Music]

[Music]
what if a computer could make pictures

what if a computer could make pictures

what if a computer could make pictures
how could we use them

how could we use them

how could we use them
[Music]

[Music]

[Music]
and what would they look like

and what would they look like

and what would they look like
Generations from now

Generations from now

Generations from now
hello folks I'm Mr computer image

[Music]

[Music]
thank you

[Music]

[Music]
[Applause]

[Applause]

[Applause]
[Music]

[Music]

[Music]
foreign

[Music]

[Music]
this

this

this
was the Showcase demo

was the Showcase demo

was the Showcase demo
it had

it had

it had
two and a half million polygons or so

two and a half million polygons or so

two and a half million polygons or so
two rays per pixel

two rays per pixel

two rays per pixel
a couple of bounces per Ray

a couple of bounces per Ray

a couple of bounces per Ray
the demonstration was frankly at the

the demonstration was frankly at the

the demonstration was frankly at the
time incredibly beautiful

time incredibly beautiful

time incredibly beautiful
that was five years ago

that was five years ago

that was five years ago
now five years later

now five years later

now five years later
racer RTX

racer RTX

racer RTX
250 million polygons 100 times more

250 million polygons 100 times more

250 million polygons 100 times more
geometry

geometry

geometry
10 Rays per pixel about 10 balances per

10 Rays per pixel about 10 balances per

10 Rays per pixel about 10 balances per
Ray

Ray

Ray
and using dlss and for like seven out of

and using dlss and for like seven out of

and using dlss and for like seven out of
eight pixels

eight pixels

eight pixels
Computing only one out of eight and as a

Computing only one out of eight and as a

Computing only one out of eight and as a
result we're able to render this at 4K

result we're able to render this at 4K

result we're able to render this at 4K
scale it up to 4K 30 Hertz hit it

not bad for real time

not bad for real time
thank you

while we were Reinventing computer

while we were Reinventing computer
graphics with artificial intelligence

graphics with artificial intelligence

graphics with artificial intelligence
we were inventing the GPU

we were inventing the GPU

we were inventing the GPU
all together for artificial intelligence

all together for artificial intelligence

all together for artificial intelligence
the GPU when I came to see you last time

the GPU when I came to see you last time

the GPU when I came to see you last time
five years ago most people would say

five years ago most people would say

five years ago most people would say
that this is what a GPU looks like and

that this is what a GPU looks like and

that this is what a GPU looks like and
in fact this is the GPU that we

in fact this is the GPU that we

in fact this is the GPU that we
announced this is the touring GPU but

announced this is the touring GPU but

announced this is the touring GPU but
this is what a GPU is today

this is what a GPU is today

this is what a GPU is today
this GPU

this GPU

this GPU
is altogether one

is altogether one

is altogether one
trillion transistors

trillion transistors

trillion transistors
this GPU has 35 000 parts

this GPU has 35 000 parts

this GPU has 35 000 parts
it's manufactured by a robot like an

it's manufactured by a robot like an

it's manufactured by a robot like an
electric car

electric car

electric car
it weighs 70 pounds

it weighs 70 pounds

it weighs 70 pounds
consumes 6000 Watts

consumes 6000 Watts

consumes 6000 Watts
and this GPU revolutionized computer

and this GPU revolutionized computer

and this GPU revolutionized computer
science altogether

science altogether

science altogether
and 12 years later after 12 years

and 12 years later after 12 years

and 12 years later after 12 years
working on artificial intelligence

working on artificial intelligence

working on artificial intelligence
something gigantic happened degenerative

something gigantic happened degenerative

something gigantic happened degenerative
AI era is upon us

AI era is upon us

AI era is upon us
the iPhone moment of AI if you will

the iPhone moment of AI if you will

the iPhone moment of AI if you will
where all of the Technologies of

where all of the Technologies of

where all of the Technologies of
artificial intelligence came together in

artificial intelligence came together in

artificial intelligence came together in
such a way that it is now possible for

such a way that it is now possible for

such a way that it is now possible for
us to enjoy AI

us to enjoy AI

us to enjoy AI
in so many different applications

in so many different applications

in so many different applications
thousands of papers in just the last

thousands of papers in just the last

thousands of papers in just the last
several years

several years

several years
have been written about this area of

have been written about this area of

have been written about this area of
large language models and generative AI

large language models and generative AI

large language models and generative AI
here are some examples of some amazing

here are some examples of some amazing

here are some examples of some amazing
things this is the Adobe Firefly Adobe

things this is the Adobe Firefly Adobe

things this is the Adobe Firefly Adobe
Firefly does owl painting imagine

Firefly does owl painting imagine

Firefly does owl painting imagine
the space around the image that we never

the space around the image that we never

the space around the image that we never
captured move AI does mocap from just

captured move AI does mocap from just

captured move AI does mocap from just
video

video

video
this is on the upper right

this is on the upper right

this is on the upper right
you could you decide which one's real

I'm going with the left

I'm going with the left
viscom does sketch to image Guided by

viscom does sketch to image Guided by

viscom does sketch to image Guided by
language prompt this one's really cool

language prompt this one's really cool

language prompt this one's really cool
there are a lot of people who know how

there are a lot of people who know how

there are a lot of people who know how
to sketch and from the sketch and some

to sketch and from the sketch and some

to sketch and from the sketch and some
guidance from your language you could

guidance from your language you could

guidance from your language you could
generate something photorealistic and

generate something photorealistic and

generate something photorealistic and
rendered

rendered

rendered
the future of computer Graphics is

the future of computer Graphics is

the future of computer Graphics is
clearly going to be revolutionized and

clearly going to be revolutionized and

clearly going to be revolutionized and
this is really cool Wonder Dynamics not

this is really cool Wonder Dynamics not

this is really cool Wonder Dynamics not
not only is the name of the company call

not only is the name of the company call

not only is the name of the company call
but they do pose and lighting detection

but they do pose and lighting detection

but they do pose and lighting detection
and replace the actor with a CG

and replace the actor with a CG

and replace the actor with a CG
character

character

character
they're just it just goes on and on and

they're just it just goes on and on and

they're just it just goes on and on and
on the generative AI era has arrived

on the generative AI era has arrived

on the generative AI era has arrived
well

well

well
what's really profound though is that

what's really profound though is that

what's really profound though is that
when you take a step back and ask

when you take a step back and ask

when you take a step back and ask
yourself what is the meaning of

yourself what is the meaning of

yourself what is the meaning of
generative AI why is this such a big

generative AI why is this such a big

generative AI why is this such a big
deal why is it changing everything

deal why is it changing everything

deal why is it changing everything
well the reason for that is

well the reason for that is

well the reason for that is
first

first

first
human is the new programming language

human is the new programming language

human is the new programming language
we've democratized computer science

we've democratized computer science

we've democratized computer science
everybody can be a programmer now

everybody can be a programmer now

everybody can be a programmer now
because human language natural language

because human language natural language

because human language natural language
is the

is the

is the
best programming language and it's the

best programming language and it's the

best programming language and it's the
reason why chat GPT has been so popular

reason why chat GPT has been so popular

reason why chat GPT has been so popular
everybody can program that computer

everybody can program that computer

everybody can program that computer
large language model is a new Computing

large language model is a new Computing

large language model is a new Computing
platform

platform

platform
because now the programming language is

because now the programming language is

because now the programming language is
human and what your program the computer

human and what your program the computer

human and what your program the computer
understands larger language models and

understands larger language models and

understands larger language models and
generative AI is the new killer app

generative AI is the new killer app

generative AI is the new killer app
these three insights has gotten

these three insights has gotten

these three insights has gotten
everybody just insanely excited and

everybody just insanely excited and

everybody just insanely excited and
because for the very first time after 15

because for the very first time after 15

because for the very first time after 15
years or so

years or so

years or so
a new Computing platform has emerged

a new Computing platform has emerged

a new Computing platform has emerged
like the PC like the internet like

like the PC like the internet like

like the PC like the internet like
mobile cloud computing a new Computing

mobile cloud computing a new Computing

mobile cloud computing a new Computing
platform has merged and this new

platform has merged and this new

platform has merged and this new
Computing platform is going to enable

Computing platform is going to enable

Computing platform is going to enable
all kinds of new applications

all kinds of new applications

all kinds of new applications
but very differently than the past

but very differently than the past

but very differently than the past
this new Computing platform benefits

this new Computing platform benefits

this new Computing platform benefits
every single Computing platform before

every single Computing platform before

every single Computing platform before
it for the very first time this new

it for the very first time this new

it for the very first time this new
Computing platform not only enables new

Computing platform not only enables new

Computing platform not only enables new
applications in this new era but helps

applications in this new era but helps

applications in this new era but helps
every application in the old era

every application in the old era

every application in the old era
this is the reason why the industry is

this is the reason why the industry is

this is the reason why the industry is
moving so fast but one particular area

moving so fast but one particular area

moving so fast but one particular area
is extremely important which is the

is extremely important which is the

is extremely important which is the
basic scale out of the cloud

basic scale out of the cloud

basic scale out of the cloud
the basic skill out of the cloud

the basic skill out of the cloud

the basic skill out of the cloud
historically was based on

historically was based on

historically was based on
off-the-shelf CPUs x86 CPUs while

off-the-shelf CPUs x86 CPUs while

off-the-shelf CPUs x86 CPUs while
general purpose Computing is a horrible

general purpose Computing is a horrible

general purpose Computing is a horrible
way of doing generative Ai and so we

way of doing generative Ai and so we

way of doing generative Ai and so we
created a brand new processor for the

created a brand new processor for the

created a brand new processor for the
era of generative Ai and this is it this

era of generative Ai and this is it this

era of generative Ai and this is it this
is the Grace Hopper we announced Grace

is the Grace Hopper we announced Grace

is the Grace Hopper we announced Grace
Hopper in fact just only recently

Hopper in fact just only recently

Hopper in fact just only recently
several months ago and today we're

several months ago and today we're

several months ago and today we're
announcing that we're going to give it a

announcing that we're going to give it a

announcing that we're going to give it a
boost we're going to give this processor

boost we're going to give this processor

boost we're going to give this processor
a boost with the world's fastest memory

a boost with the world's fastest memory

a boost with the world's fastest memory
called hbm3e the world's fast memory

called hbm3e the world's fast memory

called hbm3e the world's fast memory
fastest memory now connected to Grace

fastest memory now connected to Grace

fastest memory now connected to Grace
Hopper we're calling it gh200

Hopper we're calling it gh200

Hopper we're calling it gh200
the chips are in production it has four

the chips are in production it has four

the chips are in production it has four
petaflops of Transformer engine

petaflops of Transformer engine

petaflops of Transformer engine
processing capability

processing capability

processing capability
and

and

and
now it has five terabytes per second of

now it has five terabytes per second of

now it has five terabytes per second of
HB and 3E performance so this is the new

HB and 3E performance so this is the new

HB and 3E performance so this is the new
G

G

G
h200 based on the architecture Grace

h200 based on the architecture Grace

h200 based on the architecture Grace
Hopper and a processor for this new

Hopper and a processor for this new

Hopper and a processor for this new
Computing era the CPU now has 144 cores

Computing era the CPU now has 144 cores

Computing era the CPU now has 144 cores
the GPU has 10 terabytes per second of

the GPU has 10 terabytes per second of

the GPU has 10 terabytes per second of
frame buffer bandwidth you could take

frame buffer bandwidth you could take

frame buffer bandwidth you could take
just about any large language model you

just about any large language model you

just about any large language model you
like and put it into this and it will

like and put it into this and it will

like and put it into this and it will
inference like crazy the inference cost

inference like crazy the inference cost

inference like crazy the inference cost
of large language models will drop

of large language models will drop

of large language models will drop
significantly because look how small

significantly because look how small

significantly because look how small
this computer is and you could scale

this computer is and you could scale

this computer is and you could scale
this out in the world's data centers you

this out in the world's data centers you

this out in the world's data centers you
can connect this with ethernet you can

can connect this with ethernet you can

can connect this with ethernet you can
connect it with the finiband and of

connect it with the finiband and of

connect it with the finiband and of
course there's all kinds of different

course there's all kinds of different

course there's all kinds of different
ways that you can scale it up this is

ways that you can scale it up this is

ways that you can scale it up this is
two gpus

two gpus

two gpus
but what if we would like to scale this

but what if we would like to scale this

but what if we would like to scale this
up into a much much larger GPU run it

up into a much much larger GPU run it

up into a much much larger GPU run it
please

[Applause]

[Applause]
this is actual size by the way

this is actual size by the way

this is actual size by the way
this is actual size and it probably even

this is actual size and it probably even

this is actual size and it probably even
runs crisis

the world's largest

the world's largest
single GPU one exoflops four petaflops

single GPU one exoflops four petaflops

single GPU one exoflops four petaflops
per grasshopper 256 connected by mv-link

per grasshopper 256 connected by mv-link

per grasshopper 256 connected by mv-link
into one giant system and so this is a

into one giant system and so this is a

into one giant system and so this is a
modern GPU so next time when you order a

modern GPU so next time when you order a

modern GPU so next time when you order a
GPU on Amazon don't be surprised if this

GPU on Amazon don't be surprised if this

GPU on Amazon don't be surprised if this
shows up

okay so that's how you take race Hopper

okay so that's how you take race Hopper
and scale it up into of course a giant

and scale it up into of course a giant

and scale it up into of course a giant
system future Frontier models will be

system future Frontier models will be

system future Frontier models will be
built this way and so let me show you

built this way and so let me show you

built this way and so let me show you
this this is how you would do it and so

this this is how you would do it and so

this this is how you would do it and so
now you would have a single Grace Hopper

now you would have a single Grace Hopper

now you would have a single Grace Hopper
in each one of these nodes this is the

in each one of these nodes this is the

in each one of these nodes this is the
way Computing was done in the past for

way Computing was done in the past for

way Computing was done in the past for
the last 60 years ever since the IBM

the last 60 years ever since the IBM

the last 60 years ever since the IBM
system 360 and for the last 60 years

system 360 and for the last 60 years

system 360 and for the last 60 years
that's the way we've been doing

that's the way we've been doing

that's the way we've been doing
Computing well now

Computing well now

Computing well now
general purpose of computing is going to

general purpose of computing is going to

general purpose of computing is going to
give way to accelerated Computing and AI

give way to accelerated Computing and AI

give way to accelerated Computing and AI
Computing every single application every

Computing every single application every

Computing every single application every
single database whenever you interact

single database whenever you interact

single database whenever you interact
with an app with a computer you'll

with an app with a computer you'll

with an app with a computer you'll
likely be first engaging a large

likely be first engaging a large

likely be first engaging a large
language model that large language model

language model that large language model

language model that large language model
will figure out what is your intention

will figure out what is your intention

will figure out what is your intention
what is your desire what are you trying

what is your desire what are you trying

what is your desire what are you trying
to do given the context and present the

to do given the context and present the

to do given the context and present the
information to you in the best possible

information to you in the best possible

information to you in the best possible
way well if you were to have an ISO

way well if you were to have an ISO

way well if you were to have an ISO
budget way of processing that workload

budget way of processing that workload

budget way of processing that workload
it would take let me just choose the

it would take let me just choose the

it would take let me just choose the
number 100 million dollars and 100

number 100 million dollars and 100

number 100 million dollars and 100
million dollars would be a reasonably

million dollars would be a reasonably

million dollars would be a reasonably
small data center these days 100 million

small data center these days 100 million

small data center these days 100 million
dollars will buy you about 8 800 x86

dollars will buy you about 8 800 x86

dollars will buy you about 8 800 x86
gpus and we can take about five

gpus and we can take about five

gpus and we can take about five
megawatts to operate that and I

megawatts to operate that and I

megawatts to operate that and I
normalize the performance into 1X

normalize the performance into 1X

normalize the performance into 1X
using the exact same budget with

using the exact same budget with

using the exact same budget with
accelerated Computing Grace Hopper

accelerated Computing Grace Hopper

accelerated Computing Grace Hopper
it would consume only three megawatts

it would consume only three megawatts

it would consume only three megawatts
but your throughput goes up by an order

but your throughput goes up by an order

but your throughput goes up by an order
of magnitude

of magnitude

of magnitude
basically the Energy Efficiency the cost

basically the Energy Efficiency the cost

basically the Energy Efficiency the cost
efficiency of accelerated Computing for

efficiency of accelerated Computing for

efficiency of accelerated Computing for
generative AI applications is about 20x

generative AI applications is about 20x

generative AI applications is about 20x
20x in Moore's Law and just the current

20x in Moore's Law and just the current

20x in Moore's Law and just the current
way of scaling CPUs that would be a very

way of scaling CPUs that would be a very

way of scaling CPUs that would be a very
very long time and so this is a giant

very long time and so this is a giant

very long time and so this is a giant
Step Up in efficiency and throughput so

Step Up in efficiency and throughput so

Step Up in efficiency and throughput so
this is ISO budget let's take a look at

this is ISO budget let's take a look at

this is ISO budget let's take a look at
this now again and let's go through ISO

this now again and let's go through ISO

this now again and let's go through ISO
workload suppose your intention was to

workload suppose your intention was to

workload suppose your intention was to
provide a service and that service has

provide a service and that service has

provide a service and that service has
so many number of users and so your

so many number of users and so your

so many number of users and so your
workload is fairly well understood plus

workload is fairly well understood plus

workload is fairly well understood plus
or minus and so with ISO workload this

or minus and so with ISO workload this

or minus and so with ISO workload this
1X 100 million dollars using general

1X 100 million dollars using general

1X 100 million dollars using general
purpose computing

purpose computing

purpose computing
and using accelerated Computing Grace

and using accelerated Computing Grace

and using accelerated Computing Grace
Hopper

Hopper

Hopper
it would only cost eight million dollars

it would only cost eight million dollars

it would only cost eight million dollars
eight million dollars and only 260

eight million dollars and only 260

eight million dollars and only 260
kilowatts so 20 times less power and 12

kilowatts so 20 times less power and 12

kilowatts so 20 times less power and 12
times less cost this is the reason why

times less cost this is the reason why

times less cost this is the reason why
accelerated Computing is going to be the

accelerated Computing is going to be the

accelerated Computing is going to be the
path forward and this is the reason why

path forward and this is the reason why

path forward and this is the reason why
the world's data centers are very

the world's data centers are very

the world's data centers are very
quickly transitioning to accelerated

quickly transitioning to accelerated

quickly transitioning to accelerated
Computing and some people say and you

Computing and some people say and you

Computing and some people say and you
guys might have heard I don't know who

guys might have heard I don't know who

guys might have heard I don't know who
said it but the more you buy

said it but the more you buy

said it but the more you buy
the the more you save and and that's

the the more you save and and that's

the the more you save and and that's
that's wisdom

that's wisdom

that's wisdom
let's talk about a couple new things and

let's talk about a couple new things and

let's talk about a couple new things and
so today I want to talk about Omniverse

so today I want to talk about Omniverse

so today I want to talk about Omniverse
and generative Ai and how they come

and generative Ai and how they come

and generative Ai and how they come
together the first thing that we already

together the first thing that we already

together the first thing that we already
established is that graphics and

established is that graphics and

established is that graphics and
artificial intelligence are inseparable

artificial intelligence are inseparable

artificial intelligence are inseparable
that Graphics needs Ai and AI needs

that Graphics needs Ai and AI needs

that Graphics needs Ai and AI needs
Graphics Graphics needs Ai and AI needs

Graphics Graphics needs Ai and AI needs

Graphics Graphics needs Ai and AI needs
graphics and so the thing that we could

graphics and so the thing that we could

graphics and so the thing that we could
do is we could create a virtual world

do is we could create a virtual world

do is we could create a virtual world
that is physically simulated physics

that is physically simulated physics

that is physically simulated physics
simulator that allows an artificial

simulator that allows an artificial

simulator that allows an artificial
intelligence to learn how to perceive

intelligence to learn how to perceive

intelligence to learn how to perceive
the environment

the environment

the environment
using a vision Transformer Maybe

using a vision Transformer Maybe

using a vision Transformer Maybe
and to use reinforcement learning to

and to use reinforcement learning to

and to use reinforcement learning to
understand the consequences of its

understand the consequences of its

understand the consequences of its
physical actions and learn how to

physical actions and learn how to

physical actions and learn how to
animate and learn how to articulate to

animate and learn how to articulate to

animate and learn how to articulate to
achieve a particular goal

achieve a particular goal

achieve a particular goal
and so one mission of a connected

and so one mission of a connected

and so one mission of a connected
artificial intelligence system and a

artificial intelligence system and a

artificial intelligence system and a
virtual world system that we call

virtual world system that we call

virtual world system that we call
Omniverse is so that the future of AI

Omniverse is so that the future of AI

Omniverse is so that the future of AI
could be physically grounded the number

could be physically grounded the number

could be physically grounded the number
of applications is really quite exciting

of applications is really quite exciting

of applications is really quite exciting
because as we know the largest

because as we know the largest

because as we know the largest
Industries in the world are heavy

Industries in the world are heavy

Industries in the world are heavy
industry and those heavy Industries are

industry and those heavy Industries are

industry and those heavy Industries are
physics-based physically based and so

physics-based physically based and so

physics-based physically based and so
first application is so that AI

first application is so that AI

first application is so that AI
can learn in a virtual world

can learn in a virtual world

can learn in a virtual world
an application the second reason why AI

an application the second reason why AI

an application the second reason why AI
is and computer Graphics are inseparable

is and computer Graphics are inseparable

is and computer Graphics are inseparable
is that AI will help also to create

is that AI will help also to create

is that AI will help also to create
these Virtual Worlds

these Virtual Worlds

these Virtual Worlds
well let's take a look at what wpp the

well let's take a look at what wpp the

well let's take a look at what wpp the
world's largest Ad Agency and byd the

world's largest Ad Agency and byd the

world's largest Ad Agency and byd the
world's largest electric vehicle maker

world's largest electric vehicle maker

world's largest electric vehicle maker
how they're using Omniverse and

how they're using Omniverse and

how they're using Omniverse and
generative AI in their work play it

generative AI in their work play it

generative AI in their work play it
please

wpp is building the next generation of

wpp is building the next generation of
car configurators for automotive giant

car configurators for automotive giant

car configurators for automotive giant
byds denza luxury brand powered by

byds denza luxury brand powered by

byds denza luxury brand powered by
Omniverse cloud and generative AI open

Omniverse cloud and generative AI open

Omniverse cloud and generative AI open
USD and Omniverse Cloud allows denza to

USD and Omniverse Cloud allows denza to

USD and Omniverse Cloud allows denza to
connect High Fidelity data from

connect High Fidelity data from

connect High Fidelity data from
industry-leading CAD tools to create a

industry-leading CAD tools to create a

industry-leading CAD tools to create a
physically accurate real-time digital

physically accurate real-time digital

physically accurate real-time digital
twin of its N7

twin of its N7

twin of its N7
wpp artists can work seamlessly on this

wpp artists can work seamlessly on this

wpp artists can work seamlessly on this
model in the same Omniverse Cloud

model in the same Omniverse Cloud

model in the same Omniverse Cloud
environment with their preferred tools

environment with their preferred tools

environment with their preferred tools
from Autodesk Adobe and side effects to

from Autodesk Adobe and side effects to

from Autodesk Adobe and side effects to
deliver the next era of automotive

deliver the next era of automotive

deliver the next era of automotive
digitalization and immersive experiences

digitalization and immersive experiences

digitalization and immersive experiences
today's configurators require hundreds

today's configurators require hundreds

today's configurators require hundreds
of thousands of images to be

of thousands of images to be

of thousands of images to be
pre-rendered to represent all possible

pre-rendered to represent all possible

pre-rendered to represent all possible
options and variants

options and variants

options and variants
open USD makes it possible for wpp to

open USD makes it possible for wpp to

open USD makes it possible for wpp to
create a super digital twin of the card

create a super digital twin of the card

create a super digital twin of the card
that includes all possible variants in

that includes all possible variants in

that includes all possible variants in
one single asset deployed as a fully

one single asset deployed as a fully

one single asset deployed as a fully
interactive 3D configurator on Omniverse

interactive 3D configurator on Omniverse

interactive 3D configurator on Omniverse
Cloud gdn a network that can stream High

Cloud gdn a network that can stream High

Cloud gdn a network that can stream High
Fidelity real-time 3D experiences to

Fidelity real-time 3D experiences to

Fidelity real-time 3D experiences to
devices in over 100 regions were used to

devices in over 100 regions were used to

devices in over 100 regions were used to
generate thousands of individual pieces

generate thousands of individual pieces

generate thousands of individual pieces
of content that comprise a global

of content that comprise a global

of content that comprise a global
marketing campaign the USD model is

marketing campaign the USD model is

marketing campaign the USD model is
placed in a 3D environment that can

placed in a 3D environment that can

placed in a 3D environment that can
either be scanned from The Real World

either be scanned from The Real World

either be scanned from The Real World
using lidar and virtual production were

using lidar and virtual production were

using lidar and virtual production were
created in seconds with generative AI

created in seconds with generative AI

created in seconds with generative AI
tools from organizations such as Adobe

tools from organizations such as Adobe

tools from organizations such as Adobe
and Shutterstock this Innovative wpp

and Shutterstock this Innovative wpp

and Shutterstock this Innovative wpp
solution for byd brings generative Ai

solution for byd brings generative Ai

solution for byd brings generative Ai
and collab rendered real-time 3D

and collab rendered real-time 3D

and collab rendered real-time 3D
together for the first time powering the

together for the first time powering the

together for the first time powering the
next generation of e-commerce

next generation of e-commerce

next generation of e-commerce
kind of love that

everything everything was rendered in

everything everything was rendered in
real time nothing was pre-rendered every

real time nothing was pre-rendered every

real time nothing was pre-rendered every
single every single scene that you saw

single every single scene that you saw

single every single scene that you saw
was rendered in real time nothing was

was rendered in real time nothing was

was rendered in real time nothing was
changed you literally take the cad you

changed you literally take the cad you

changed you literally take the cad you
drag it into Omniverse you tell an AI

drag it into Omniverse you tell an AI

drag it into Omniverse you tell an AI
synthesize and generate an environment

synthesize and generate an environment

synthesize and generate an environment
and all of a sudden the car up here is

and all of a sudden the car up here is

and all of a sudden the car up here is
wherever you like it to be so this is a

wherever you like it to be so this is a

wherever you like it to be so this is a
one example of how generative Ai and

one example of how generative Ai and

one example of how generative Ai and
human designs come together to create

human designs come together to create

human designs come together to create
these incredible applications and so how

these incredible applications and so how

these incredible applications and so how
do we do this applications generative AI

do we do this applications generative AI

do we do this applications generative AI
models are making tremendous

models are making tremendous

models are making tremendous
breakthroughs in fact the hopper GPU is

breakthroughs in fact the hopper GPU is

breakthroughs in fact the hopper GPU is
impossible to design by humans we needed

impossible to design by humans we needed

impossible to design by humans we needed
AIS and generative models to help us

AIS and generative models to help us

AIS and generative models to help us
find the way to design this thing in

find the way to design this thing in

find the way to design this thing in
such a high performance way and so it

such a high performance way and so it

such a high performance way and so it
augments our design Engineers it makes

augments our design Engineers it makes

augments our design Engineers it makes
it possible for us to create some of

it possible for us to create some of

it possible for us to create some of
these amazing things at all and of

these amazing things at all and of

these amazing things at all and of
course the productivity of the teams go

course the productivity of the teams go

course the productivity of the teams go
up tremendously well we would like to do

up tremendously well we would like to do

up tremendously well we would like to do
this in just about every single industry

this in just about every single industry

this in just about every single industry
now we just need some powerful machines

now we just need some powerful machines

now we just need some powerful machines
we have powerful machines in the cloud

we have powerful machines in the cloud

we have powerful machines in the cloud
of course dgx cloud has many many uh

of course dgx cloud has many many uh

of course dgx cloud has many many uh
many footprints around the world but

many footprints around the world but

many footprints around the world but
wouldn't it be great if you had a

wouldn't it be great if you had a

wouldn't it be great if you had a
powerful machine under your desk and so

powerful machine under your desk and so

powerful machine under your desk and so
today we're announcing our latest

today we're announcing our latest

today we're announcing our latest
generation Lovelace GPU

generation Lovelace GPU

generation Lovelace GPU
the most powerful GPU we've ever put in

the most powerful GPU we've ever put in

the most powerful GPU we've ever put in
the workstation

the workstation

the workstation
is now

is now

is now
oh gosh darn it

oh gosh darn it

oh gosh darn it
I just put my fingerprints on there did

I just put my fingerprints on there did

I just put my fingerprints on there did
you guys can you guys see that

you guys can you guys see that

you guys can you guys see that
that was what that's not me could you

that was what that's not me could you

that was what that's not me could you
hey can I have this clean in the future

my bad

my bad
sorry everybody uh my bad

sorry everybody uh my bad

sorry everybody uh my bad
they work so hard it goes into these

they work so hard it goes into these

they work so hard it goes into these
amazing workstations and these amazing

amazing workstations and these amazing

amazing workstations and these amazing
workstations

workstations

workstations
packs up to four of these gpus it packs

packs up to four of these gpus it packs

packs up to four of these gpus it packs
up to four Nvidia RTX 6000 is the most

up to four Nvidia RTX 6000 is the most

up to four Nvidia RTX 6000 is the most
powerful gpus ever created and it run

powerful gpus ever created and it run

powerful gpus ever created and it run
real-time Ray tracing for Omniverse as

real-time Ray tracing for Omniverse as

real-time Ray tracing for Omniverse as
well as train fine tune and inference

well as train fine tune and inference

well as train fine tune and inference
large language models for generative AI

large language models for generative AI

large language models for generative AI
incredibly fast and critically powerful

incredibly fast and critically powerful

incredibly fast and critically powerful
and it produces answers in seconds not

and it produces answers in seconds not

and it produces answers in seconds not
minutes uh for in some of the services

minutes uh for in some of the services

minutes uh for in some of the services
that are out there okay and so another

that are out there okay and so another

that are out there okay and so another
incredible machine are the servers and

incredible machine are the servers and

incredible machine are the servers and
these servers as you know getting gpus

these servers as you know getting gpus

these servers as you know getting gpus
in the cloud these days is no easy feat

in the cloud these days is no easy feat

in the cloud these days is no easy feat
and now you can buy it okay you could

and now you can buy it okay you could

and now you can buy it okay you could
have your your company buy it for you

have your your company buy it for you

have your your company buy it for you
and put it in the data center and

and put it in the data center and

and put it in the data center and
there's a whole bunch of these servers a

there's a whole bunch of these servers a

there's a whole bunch of these servers a
whole bunch of different configurations

whole bunch of different configurations

whole bunch of different configurations
I don't know if you guys could see this

I don't know if you guys could see this

I don't know if you guys could see this
this is a server that has up to eight of

this is a server that has up to eight of

this is a server that has up to eight of
the l40s

the l40s

the l40s
Ada Lovelace gpus and of course these

Ada Lovelace gpus and of course these

Ada Lovelace gpus and of course these
are not going to be used for Frontier

are not going to be used for Frontier

are not going to be used for Frontier
models these are really used for

models these are really used for

models these are really used for
mainstream models today that you can

mainstream models today that you can

mainstream models today that you can
download from hugging face or Nvidia

download from hugging face or Nvidia

download from hugging face or Nvidia
could work with your company to create

could work with your company to create

could work with your company to create
you could use in just about all kinds of

you could use in just about all kinds of

you could use in just about all kinds of
applications around your company and you

applications around your company and you

applications around your company and you
can fine tune it with these gpus the

can fine tune it with these gpus the

can fine tune it with these gpus the
fine tuning of a gpt3 model okay so this

fine tuning of a gpt3 model okay so this

fine tuning of a gpt3 model okay so this
is GPT 340 billion parameters takes

is GPT 340 billion parameters takes

is GPT 340 billion parameters takes
about seven hours for about a billion

about seven hours for about a billion

about seven hours for about a billion
tokens and so 15 hours in a workstation

tokens and so 15 hours in a workstation

tokens and so 15 hours in a workstation
with four gpus of course takes less with

with four gpus of course takes less with

with four gpus of course takes less with
agpus and just in fine tuning this is

agpus and just in fine tuning this is

agpus and just in fine tuning this is
one and a half times faster than our

one and a half times faster than our

one and a half times faster than our
last generation

last generation

last generation
a100 and so l40s is a really terrific

a100 and so l40s is a really terrific

a100 and so l40s is a really terrific
GPU for Enterprise scale fine-tuning of

GPU for Enterprise scale fine-tuning of

GPU for Enterprise scale fine-tuning of
mainstream large language models these

mainstream large language models these

mainstream large language models these
amazing new Enterprise systems that are

amazing new Enterprise systems that are

amazing new Enterprise systems that are
in production today

in production today

in production today
all right let's change gears and talk

all right let's change gears and talk

all right let's change gears and talk
about what's going on at siggraph this

about what's going on at siggraph this

about what's going on at siggraph this
year

year

year
I'm pretty sure all of you have already

I'm pretty sure all of you have already

I'm pretty sure all of you have already
heard about open USD USD openusd is a

heard about open USD USD openusd is a

heard about open USD USD openusd is a
very big deal open USD is a framework

very big deal open USD is a framework

very big deal open USD is a framework
a universal interchange for creating 3D

a universal interchange for creating 3D

a universal interchange for creating 3D
worlds for describing for compositing

worlds for describing for compositing

worlds for describing for compositing
for simulating for collaborating on 3D

for simulating for collaborating on 3D

for simulating for collaborating on 3D
projects

open USD

open USD
is going to bring together the world

is going to bring together the world

is going to bring together the world
onto one standard 3D interchange

onto one standard 3D interchange

onto one standard 3D interchange
and has the opportunity to do for the

and has the opportunity to do for the

and has the opportunity to do for the
world and for computing what HTML did

world and for computing what HTML did

world and for computing what HTML did
for the 2D web

for the 2D web

for the 2D web
finally an industry standard

finally an industry standard

finally an industry standard
powerful and extensible 3D interchange

powerful and extensible 3D interchange

powerful and extensible 3D interchange
that brings the whole world together a

that brings the whole world together a

that brings the whole world together a
really big deal now let's take a look at

really big deal now let's take a look at

really big deal now let's take a look at
why it is such a Visionary thing that

why it is such a Visionary thing that

why it is such a Visionary thing that
Pixar did I forget exactly when they

Pixar did I forget exactly when they

Pixar did I forget exactly when they
invented it but they open sourced it in

invented it but they open sourced it in

invented it but they open sourced it in
2015 and they've been of course using

2015 and they've been of course using

2015 and they've been of course using
this framework for over a decade

this framework for over a decade

this framework for over a decade
building amazing 3D content well the 3D

building amazing 3D content well the 3D

building amazing 3D content well the 3D
pipeline is incredibly complicated you

pipeline is incredibly complicated you

pipeline is incredibly complicated you
got designers and artists and Engineers

got designers and artists and Engineers

got designers and artists and Engineers
they all specialize in some part of the

they all specialize in some part of the

they all specialize in some part of the
3D workflow it could be modeling and

3D workflow it could be modeling and

3D workflow it could be modeling and
texturing

texturing

texturing
materials

materials

materials
physics simulation

physics simulation

physics simulation
animation

animation

animation
set design scene composition

set design scene composition

set design scene composition
there are so many parts and so many

there are so many parts and so many

there are so many parts and so many
different tools and because the tools

different tools and because the tools

different tools and because the tools
are created by different companies and

are created by different companies and

are created by different companies and
largely incompatible

largely incompatible

largely incompatible
import and exporting data conversion is

import and exporting data conversion is

import and exporting data conversion is
just part of the workflow and because

just part of the workflow and because

just part of the workflow and because
they're incompatible and because there's

they're incompatible and because there's

they're incompatible and because there's
all this import and exporting

all this import and exporting

all this import and exporting
fundamentally the workflow has to be

fundamentally the workflow has to be

fundamentally the workflow has to be
serialized it's impossible to paralyze

serialized it's impossible to paralyze

serialized it's impossible to paralyze
that

that

that
and this is one of the reasons why

and this is one of the reasons why

and this is one of the reasons why
creating these incredible 3D animation

creating these incredible 3D animation

creating these incredible 3D animation
movies are so expensive and take so much

movies are so expensive and take so much

movies are so expensive and take so much
time could you imagine if every single

time could you imagine if every single

time could you imagine if every single
tool was natively compatible with USD

tool was natively compatible with USD

tool was natively compatible with USD
then as a result everybody can work in

then as a result everybody can work in

then as a result everybody can work in
parallel

parallel

parallel
The Interchange and conversion goes away

The Interchange and conversion goes away

The Interchange and conversion goes away
and instead of a serialized model you

and instead of a serialized model you

and instead of a serialized model you
have a paralyzed spoken Hub model

have a paralyzed spoken Hub model

have a paralyzed spoken Hub model
and so this way of doing work of course

and so this way of doing work of course

and so this way of doing work of course
is incredibly appealing and it's one of

is incredibly appealing and it's one of

is incredibly appealing and it's one of
the reasons why the vision of open USD

the reasons why the vision of open USD

the reasons why the vision of open USD
has taken off

has taken off

has taken off
it's being adopted in film architecture

it's being adopted in film architecture

it's being adopted in film architecture
engineering and construction

engineering and construction

engineering and construction
manufacturing and so many different

manufacturing and so many different

manufacturing and so many different
fields of Robotics well five years ago

fields of Robotics well five years ago

fields of Robotics well five years ago
we started working with Pixar and we

we started working with Pixar and we

we started working with Pixar and we
adopted USD as the foundation of

adopted USD as the foundation of

adopted USD as the foundation of
Omniverse it's not a tool in itself it's

Omniverse it's not a tool in itself it's

Omniverse it's not a tool in itself it's
a connector of tools okay so Omniverse

a connector of tools okay so Omniverse

a connector of tools okay so Omniverse
is a connector well let's take a look at

is a connector well let's take a look at

is a connector well let's take a look at
held the vision of open USD came came

held the vision of open USD came came

held the vision of open USD came came
together and this is just a fantastic

together and this is just a fantastic

together and this is just a fantastic
illustration let's starting from the

illustration let's starting from the

illustration let's starting from the
left here I think this is a Adobe Stager

left here I think this is a Adobe Stager

left here I think this is a Adobe Stager
Houdini this is a modeling system Maya

Houdini this is a modeling system Maya

Houdini this is a modeling system Maya
or animation system modeling system this

or animation system modeling system this

or animation system modeling system this
is Omniverse

is Omniverse

is Omniverse
blender

blender

blender
render man Pixar's Minuteman and Unreal

render man Pixar's Minuteman and Unreal

render man Pixar's Minuteman and Unreal
Engine from epic a game engine literally

Engine from epic a game engine literally

Engine from epic a game engine literally
all open USD

all open USD

all open USD
one data set

one data set

one data set
ingested into everybody's tools and it

ingested into everybody's tools and it

ingested into everybody's tools and it
looks basically the same

looks basically the same

looks basically the same
everybody's rendering system is a little

everybody's rendering system is a little

everybody's rendering system is a little
different and so the quality of the

different and so the quality of the

different and so the quality of the
rendering is is a little different from

rendering is is a little different from

rendering is is a little different from
tool to Tool but one data set available

tool to Tool but one data set available

tool to Tool but one data set available
and usable by every tool this is the

and usable by every tool this is the

and usable by every tool this is the
vision of open USD so incredibly

vision of open USD so incredibly

vision of open USD so incredibly
powerful

powerful

powerful
and so take a look at this this is the

and so take a look at this this is the

and so take a look at this this is the
latest of Omniverse

[Music]

[Music]
thank you

[Music]

[Music]
foreign

foreign

foreign
[Music]

foreign

foreign
[Music]

no art all physics now Omniverse as I

no art all physics now Omniverse as I
mentioned before is not a tool it's a

mentioned before is not a tool it's a

mentioned before is not a tool it's a
platform for tools it's not a tool is a

platform for tools it's not a tool is a

platform for tools it's not a tool is a
platform for tools okay and so we would

platform for tools okay and so we would

platform for tools okay and so we would
like to put Omniverse in as many places

like to put Omniverse in as many places

like to put Omniverse in as many places
as possible there's a whole bunch of

as possible there's a whole bunch of

as possible there's a whole bunch of
different applications and we showed you

different applications and we showed you

different applications and we showed you
earlier how we used AI workbench to

earlier how we used AI workbench to

earlier how we used AI workbench to
train this model to fine-tune this model

train this model to fine-tune this model

train this model to fine-tune this model
we started with llama 2 and we taught it

we started with llama 2 and we taught it

we started with llama 2 and we taught it
we taught it we fine-tuned it for USD

we taught it we fine-tuned it for USD

we taught it we fine-tuned it for USD
and so let's take a look at the video

and so let's take a look at the video

and so let's take a look at the video
for USD developers building profiling

for USD developers building profiling

for USD developers building profiling
and optimizing large 3D scenes can be a

and optimizing large 3D scenes can be a

and optimizing large 3D scenes can be a
very complex process

very complex process

very complex process
chat USD is an llm that's fine-tuned

chat USD is an llm that's fine-tuned

chat USD is an llm that's fine-tuned
with USD functions and python USD code

with USD functions and python USD code

with USD functions and python USD code
Snippets using Nvidia AI workbench and

Snippets using Nvidia AI workbench and

Snippets using Nvidia AI workbench and
the Nemo framework

the Nemo framework

the Nemo framework
this generative AI copilot is easily

this generative AI copilot is easily

this generative AI copilot is easily
accessed as an Omniverse Cloud API

accessed as an Omniverse Cloud API

accessed as an Omniverse Cloud API
simplifying your USD development tasks

simplifying your USD development tasks

simplifying your USD development tasks
directly in Omniverse

directly in Omniverse

directly in Omniverse
use chat USD for general knowledge like

use chat USD for general knowledge like

use chat USD for general knowledge like
to understand the geometry properties of

to understand the geometry properties of

to understand the geometry properties of
your USD schema or complete previously

your USD schema or complete previously

your USD schema or complete previously
tedious repetitive tasks

tedious repetitive tasks

tedious repetitive tasks
like generating code to find and replace

like generating code to find and replace

like generating code to find and replace
materials on specific objects

or to instantly expose all variants of a

or to instantly expose all variants of a
USD print

USD print

USD print
[Music]

chat USD can also help you build complex

chat USD can also help you build complex
scenes such as scaling a scene and

scenes such as scaling a scene and

scenes such as scaling a scene and
organizing it in a certain way in your

organizing it in a certain way in your

organizing it in a certain way in your
USD stage

build bigger more complex Virtual Worlds

build bigger more complex Virtual Worlds
faster than ever with chat USB

faster than ever with chat USB

faster than ever with chat USB
generative AI for USD workflows

generative AI for USD workflows

generative AI for USD workflows
chat USD now everybody can speak USD and

chat USD now everybody can speak USD and

chat USD now everybody can speak USD and
chat USB USD could be a USD teacher it

chat USB USD could be a USD teacher it

chat USB USD could be a USD teacher it
could be a USD copilot and help you

could be a USD copilot and help you

could be a USD copilot and help you
create your virtual world okay enhance

create your virtual world okay enhance

create your virtual world okay enhance
your productivity incredibly we're super

your productivity incredibly we're super

your productivity incredibly we're super
excited about the work that we're doing

excited about the work that we're doing

excited about the work that we're doing
here we're so happy that we chose

here we're so happy that we chose

here we're so happy that we chose
openusd as the foundation of Omniverse

openusd as the foundation of Omniverse

openusd as the foundation of Omniverse
and all of the work that we've done to

and all of the work that we've done to

and all of the work that we've done to
extend it into real time into

extend it into real time into

extend it into real time into
physics-based applications and this is

physics-based applications and this is

physics-based applications and this is
the beginning of a journey that we will

the beginning of a journey that we will

the beginning of a journey that we will
finally be able to digitalize

finally be able to digitalize

finally be able to digitalize
to bring software-driven artificial

to bring software-driven artificial

to bring software-driven artificial
intelligence powered workflows into the

intelligence powered workflows into the

intelligence powered workflows into the
world's heavy Industries the 50 trillion

world's heavy Industries the 50 trillion

world's heavy Industries the 50 trillion
dollars worth of industries that are

dollars worth of industries that are

dollars worth of industries that are
wasting enormous amounts of energy and

wasting enormous amounts of energy and

wasting enormous amounts of energy and
money and time all the time because it

money and time all the time because it

money and time all the time because it
was simply based built on technology

was simply based built on technology

was simply based built on technology
that wasn't available at the time and so

that wasn't available at the time and so

that wasn't available at the time and so
Omniverse for industrial digitalization

Omniverse for industrial digitalization

Omniverse for industrial digitalization
well all of this momentum that we've

well all of this momentum that we've

well all of this momentum that we've
already seen with openusd is about to

already seen with openusd is about to

already seen with openusd is about to
get turbocharged

get turbocharged

get turbocharged
Alliance for open USD was announced with

Alliance for open USD was announced with

Alliance for open USD was announced with
Pixar Apple Adobe Autodesk Nvidia as the

Pixar Apple Adobe Autodesk Nvidia as the

Pixar Apple Adobe Autodesk Nvidia as the
founding members the alliance's mission

founding members the alliance's mission

founding members the alliance's mission
is to Foster development and

is to Foster development and

is to Foster development and
standardization of open USD and

standardization of open USD and

standardization of open USD and
accelerate this adoption so whatever

accelerate this adoption so whatever

accelerate this adoption so whatever
momentum we've already enjoyed the

momentum we've already enjoyed the

momentum we've already enjoyed the
vision that we've already enjoyed it's

vision that we've already enjoyed it's

vision that we've already enjoyed it's
about get kicked into Turbo Charge

about get kicked into Turbo Charge

about get kicked into Turbo Charge
well I want to thank all of you for

well I want to thank all of you for

well I want to thank all of you for
coming today and remember remember

coming today and remember remember

coming today and remember remember
accelerated Computing and generative AI

accelerated Computing and generative AI

accelerated Computing and generative AI
the more you buy the more you save

the more you buy the more you save

the more you buy the more you save
[Applause]

[Applause]

[Applause]
[Music]