Original subtitles

Now this particular system here is 40

Now this particular system here is 40

Now this particular system here is 40 pedaflops which is approximately the

pedaflops which is approximately the

pedaflops which is approximately the performance of

performance of

performance of the Sierra supercomput in

the Sierra supercomput in

the Sierra supercomput in 2018. The Sierra supercomput has 18,000

2018. The Sierra supercomput has 18,000

2018. The Sierra supercomput has 18,000 GPUs volta GPUs. This one node here

GPUs volta GPUs. This one node here

GPUs volta GPUs. This one node here replaces that entire supercomput.

4,000 times increase in performance in

4,000 times increase in performance in

4,000 times increase in performance in six

six

six years. That is extreme Moore's

years. That is extreme Moore's

years. That is extreme Moore's law. Remember, I've said before that AI,

law. Remember, I've said before that AI,

law. Remember, I've said before that AI, Nvidia has been scaling computing by

Nvidia has been scaling computing by

Nvidia has been scaling computing by about a million times every 10 years,

about a million times every 10 years,

about a million times every 10 years, and we're still on that track. And you

and we're still on that track. And you

and we're still on that track. And you heard me say

heard me say

heard me say before, the modern computer is an entire

before, the modern computer is an entire

before, the modern computer is an entire data center. The data center is a unit

data center. The data center is a unit

data center. The data center is a unit of computing. No longer just a PC, no

of computing. No longer just a PC, no

of computing. No longer just a PC, no longer just a server. The entire data

longer just a server. The entire data

longer just a server. The entire data center is running one job. And the

center is running one job. And the

center is running one job. And the operating system would change. The

operating system would change. The

operating system would change. The revolutionary computer called Hopper

revolutionary computer called Hopper

revolutionary computer called Hopper came into the world about three years

came into the world about three years

came into the world about three years ago and it revolutionized AI as we know

ago and it revolutionized AI as we know

ago and it revolutionized AI as we know it. It became probably the most popular,

it. It became probably the most popular,

it. It became probably the most popular, most well-known computer in the world.

most well-known computer in the world.

most well-known computer in the world. In the last several years, we've been

In the last several years, we've been

In the last several years, we've been working on a new computer to make it

working on a new computer to make it

working on a new computer to make it possible for us to do inference time

possible for us to do inference time

possible for us to do inference time scaling or basically thinking incredibly

scaling or basically thinking incredibly

scaling or basically thinking incredibly fast because when you think you're

fast because when you think you're

fast because when you think you're generating a lot of tokens in your head,

generating a lot of tokens in your head,

generating a lot of tokens in your head, if you will, you're generating a lot of

if you will, you're generating a lot of

if you will, you're generating a lot of thoughts and you iterate in your brain

thoughts and you iterate in your brain

thoughts and you iterate in your brain before you produce the answer. So what

before you produce the answer. So what

before you produce the answer. So what used to be oneshot AI is now going to be

used to be oneshot AI is now going to be

used to be oneshot AI is now going to be thinking AI, reasoning AI, inference

thinking AI, reasoning AI, inference

thinking AI, reasoning AI, inference time scaling AI and that's going to take

time scaling AI and that's going to take

time scaling AI and that's going to take a lot more computation. And so we

a lot more computation. And so we

a lot more computation. And so we created a new system called Grace

created a new system called Grace

created a new system called Grace Blackwell. Grace Blackwell does several

Blackwell. Grace Blackwell does several

Blackwell. Grace Blackwell does several things. It has the ability to scale up.

things. It has the ability to scale up.

things. It has the ability to scale up. Scale up means to turn what is a

Scale up means to turn what is a

Scale up means to turn what is a computer into a giant computer. Scale

computer into a giant computer. Scale

computer into a giant computer. Scale out is to take a computer and connect

out is to take a computer and connect

out is to take a computer and connect many of them together and let the work

many of them together and let the work

many of them together and let the work be done in many different computers.

be done in many different computers.

be done in many different computers. Scaling out is easy. Scaling up is

Scaling out is easy. Scaling up is

Scaling out is easy. Scaling up is incredibly

incredibly

incredibly hard. Building larger

hard. Building larger

hard. Building larger computers that is beyond the limits of

computers that is beyond the limits of

computers that is beyond the limits of semiconductor physics is insanely hard.

semiconductor physics is insanely hard.

semiconductor physics is insanely hard. And that's what Grace Blackwell does.

And that's what Grace Blackwell does.

And that's what Grace Blackwell does. It's available in coreweave now for

It's available in coreweave now for

It's available in coreweave now for several weeks. It's already being used

several weeks. It's already being used

several weeks. It's already being used by many CSPs. And now you're starting to

by many CSPs. And now you're starting to

by many CSPs. And now you're starting to see it coming up from everywhere.

see it coming up from everywhere.

see it coming up from everywhere. Everybody started to tweet out that

Everybody started to tweet out that

Everybody started to tweet out that Grace Blackwell is in for production. In

Grace Blackwell is in for production. In

Grace Blackwell is in for production. In Q3 of this year, just as I promised,

Q3 of this year, just as I promised,

Q3 of this year, just as I promised, every single year, we will increase the

every single year, we will increase the

every single year, we will increase the performance of our platform every single

performance of our platform every single

performance of our platform every single year like Rhythm. And this this year in

year like Rhythm. And this this year in

year like Rhythm. And this this year in Q3, we'll upgrade to Grace Blackwell

Q3, we'll upgrade to Grace Blackwell

Q3, we'll upgrade to Grace Blackwell GB300. Same architecture, same physical

GB300. Same architecture, same physical

GB300. Same architecture, same physical footprint, same electrical, mechanicals,

footprint, same electrical, mechanicals,

footprint, same electrical, mechanicals, but the chips inside have been upgraded.

but the chips inside have been upgraded.

but the chips inside have been upgraded. It has upgraded with a new black wall

It has upgraded with a new black wall

It has upgraded with a new black wall chip is now one and a half times more

chip is now one and a half times more

chip is now one and a half times more inference performance, has one and a

inference performance, has one and a

inference performance, has one and a half times more HBM memory and it has

half times more HBM memory and it has

half times more HBM memory and it has two times more networking. And so the

two times more networking. And so the

two times more networking. And so the overall system performance is higher.

overall system performance is higher.

overall system performance is higher. Well, let's take a look at what's inside

Well, let's take a look at what's inside

Well, let's take a look at what's inside Grace Blackwell. Grace Blackwell starts

Grace Blackwell. Grace Blackwell starts

Grace Blackwell. Grace Blackwell starts with this compute node. This compute

with this compute node. This compute

with this compute node. This compute node right here, this is one of the

node right here, this is one of the

node right here, this is one of the compute nodes. This is um what the last

compute nodes. This is um what the last

compute nodes. This is um what the last generation looks like. the B200. This is

generation looks like. the B200. This is

generation looks like. the B200. This is what B300 looks like. Notice right here

what B300 looks like. Notice right here

what B300 looks like. Notice right here in the center, it's 100% liquid cooled

in the center, it's 100% liquid cooled

in the center, it's 100% liquid cooled now, but otherwise externally it's the

now, but otherwise externally it's the

now, but otherwise externally it's the same. You could plug it into the same

same. You could plug it into the same

same. You could plug it into the same systems and same chassis. And so this is

systems and same chassis. And so this is

systems and same chassis. And so this is the Grace Blackwell GB300 system. It's

the Grace Blackwell GB300 system. It's

the Grace Blackwell GB300 system. It's one and a half times more inference

one and a half times more inference

one and a half times more inference performance. The training performance is

performance. The training performance is

performance. The training performance is about the same, but the inference

about the same, but the inference

about the same, but the inference performance is one and a half times

performance is one and a half times

performance is one and a half times more. Remember, I've said before that

more. Remember, I've said before that

more. Remember, I've said before that AI, Nvidia has been scaling computing by

AI, Nvidia has been scaling computing by

AI, Nvidia has been scaling computing by about a million times every 10 years,

about a million times every 10 years,

about a million times every 10 years, and we're still on that track. But the

and we're still on that track. But the

and we're still on that track. But the way to do that is not just to make the

way to do that is not just to make the

way to do that is not just to make the chips faster. There's only a limit to

chips faster. There's only a limit to

chips faster. There's only a limit to how fast you can make chips and how big

how fast you can make chips and how big

how fast you can make chips and how big you can make chips. In the case of

you can make chips. In the case of

you can make chips. In the case of Blackwell, we even connected two chips

Blackwell, we even connected two chips

Blackwell, we even connected two chips together to make it possible. TSMC

together to make it possible. TSMC

together to make it possible. TSMC worked with us to invent a new co-as

worked with us to invent a new co-as

worked with us to invent a new co-as process called cos l that made it

process called cos l that made it

process called cos l that made it possible for us to create these giant

possible for us to create these giant

possible for us to create these giant chips but still we want chips way bigger

chips but still we want chips way bigger

chips but still we want chips way bigger than that and so we had to create what

than that and so we had to create what

than that and so we had to create what is called mvlink this is the world's

is called mvlink this is the world's

is called mvlink this is the world's fastest switch this MVL link here right

fastest switch this MVL link here right

fastest switch this MVL link here right here is 7.2 2 terabytes per second. Nine

here is 7.2 2 terabytes per second. Nine

here is 7.2 2 terabytes per second. Nine of these go into that rack. And that

of these go into that rack. And that

of these go into that rack. And that nine, those nine switches are connected

nine, those nine switches are connected

nine, those nine switches are connected by this miracle. This is um quite

by this miracle. This is um quite

by this miracle. This is um quite heavy. That's because I'm quite strong.

heavy. That's because I'm quite strong.

heavy. That's because I'm quite strong. I made it I made it look so light, but

I made it I made it look so light, but

I made it I made it look so light, but this is almost this is 70

this is almost this is 70

this is almost this is 70 lbs. And so this is the MVLength spine.

lbs. And so this is the MVLength spine.

lbs. And so this is the MVLength spine. Two miles of

Two miles of

Two miles of cables, 5,000 cables

cables, 5,000 cables

cables, 5,000 cables structured, all

structured, all

structured, all coaxed impen matched and it connects all

coaxed impen matched and it connects all

coaxed impen matched and it connects all 72 GPUs to all of the other 72 GPUs

72 GPUs to all of the other 72 GPUs

72 GPUs to all of the other 72 GPUs across this network called MVLink

across this network called MVLink

across this network called MVLink switch. 130 terabytes per second of

switch. 130 terabytes per second of

switch. 130 terabytes per second of bandwidth across the MVLink spine. So

bandwidth across the MVLink spine. So

bandwidth across the MVLink spine. So just put in

just put in

just put in perspective the

perspective the

perspective the peak traffic of the entire

peak traffic of the entire

peak traffic of the entire internet the peak traffic of the entire

internet the peak traffic of the entire

internet the peak traffic of the entire internet is

internet is

internet is 900

900

900 terabits per

terabits per

terabits per second. Divide that by

second. Divide that by

second. Divide that by eight.

eight.

eight. This moves more traffic than the entire

internet. one MVLink spine across this

internet. one MVLink spine across this

internet. one MVLink spine across this MV nine of these MVLink switches so that

MV nine of these MVLink switches so that

MV nine of these MVLink switches so that every single GPU can talk to every other

every single GPU can talk to every other

every single GPU can talk to every other GPU at exactly the same time. This is

GPU at exactly the same time. This is

GPU at exactly the same time. This is the miracle

of

of

of GB200 and because there's a limit to how

GB200 and because there's a limit to how

GB200 and because there's a limit to how far you can drive Certis. This is as far

far you can drive Certis. This is as far

far you can drive Certis. This is as far as any Certis has ever driven from CH.

as any Certis has ever driven from CH.

as any Certis has ever driven from CH. This goes chip to the switch out to the

This goes chip to the switch out to the

This goes chip to the switch out to the spine to any other any other any other

spine to any other any other any other

spine to any other any other any other switch any other chip all electrical.

switch any other chip all electrical.

switch any other chip all electrical. And so that limit caused us to put

And so that limit caused us to put

And so that limit caused us to put everything in one rack. That one rack is

everything in one rack. That one rack is

everything in one rack. That one rack is 120 kilowatts which is the reason why

120 kilowatts which is the reason why

120 kilowatts which is the reason why everything has to be liquid cooled. We

everything has to be liquid cooled. We

everything has to be liquid cooled. We now have the ability to disagregate the

now have the ability to disagregate the

now have the ability to disagregate the GPUs out of one motherboard essentially

GPUs out of one motherboard essentially

GPUs out of one motherboard essentially across an entire rack. And so that

across an entire rack. And so that

across an entire rack. And so that entire rack is one motherboard. That's

entire rack is one motherboard. That's

entire rack is one motherboard. That's the miracle completely disagregated. And

the miracle completely disagregated. And

the miracle completely disagregated. And now the GPU performance is incredible.

now the GPU performance is incredible.

now the GPU performance is incredible. The amount of memory is incredible. The

The amount of memory is incredible. The

The amount of memory is incredible. The networking bandwidth is incredible. And

networking bandwidth is incredible. And

networking bandwidth is incredible. And now we can really scale these out. Once

now we can really scale these out. Once

now we can really scale these out. Once we scale it up, then we can scale it out

we scale it up, then we can scale it out

we scale it up, then we can scale it out into large systems. And notice almost

into large systems. And notice almost

into large systems. And notice almost everything Nvidia builds are gigantic.

everything Nvidia builds are gigantic.

everything Nvidia builds are gigantic. And the reason for that is because we're

And the reason for that is because we're

And the reason for that is because we're not building data centers and servers.

not building data centers and servers.

not building data centers and servers. We're building AI factories. This is

We're building AI factories. This is

We're building AI factories. This is Coreweave. This is Oracle Cloud. And

Coreweave. This is Oracle Cloud. And

Coreweave. This is Oracle Cloud. And this is the XAI Colossus factory. This

this is the XAI Colossus factory. This

this is the XAI Colossus factory. This is

is

is Stargate. 4 million square feet. 4

Stargate. 4 million square feet. 4

Stargate. 4 million square feet. 4 million square feet. One gigawatt. And

million square feet. One gigawatt. And

million square feet. One gigawatt. And so just think about this factory here.

so just think about this factory here.

so just think about this factory here. This one gawatt factory. This one gawatt

This one gawatt factory. This one gawatt

This one gawatt factory. This one gawatt factory is probably going to be about,

factory is probably going to be about,

factory is probably going to be about, you know, 60 to80 billion. out of that

you know, 60 to80 billion. out of that

you know, 60 to80 billion. out of that 60 to80 billion the electronics the

60 to80 billion the electronics the

60 to80 billion the electronics the computing part of it these systems are

computing part of it these systems are

computing part of it these systems are 4050 billion dollars of it and so these

4050 billion dollars of it and so these

4050 billion dollars of it and so these are gigantic factory

are gigantic factory

are gigantic factory investments the reason why people build

investments the reason why people build

investments the reason why people build factories is

factories is

factories is because you know you know the answer the

because you know you know the answer the

because you know you know the answer the more you the more you

buy say it with me the more you buy the

buy say it with me the more you buy the

buy say it with me the more you buy the more you

more you

more you make. That's what factories

make. That's what factories

make. That's what factories do.

do.

do. Okay. And so today we're announcing

Okay. And so today we're announcing

Okay. And so today we're announcing something very

something very

something very special. We're announcing NVIDIA MVLink

special. We're announcing NVIDIA MVLink

special. We're announcing NVIDIA MVLink Fusion.

Fusion.

Fusion. Envy Link

Envy Link

Envy Link Fusion is so that you can

Fusion is so that you can

Fusion is so that you can build

build

build semicustom AI

semicustom AI

semicustom AI infrastructure, not just semi-custom

infrastructure, not just semi-custom

infrastructure, not just semi-custom chips, because those are the good old

chips, because those are the good old

chips, because those are the good old days. You want to build AI

days. You want to build AI

days. You want to build AI infrastructure. And everybody's AI

infrastructure. And everybody's AI

infrastructure. And everybody's AI infrastructure could be a little

infrastructure could be a little

infrastructure could be a little different. Some of you could have a lot

different. Some of you could have a lot

different. Some of you could have a lot more CPUs and some of it could have a

more CPUs and some of it could have a

more CPUs and some of it could have a lot more Nvidia GPUs and some of it

lot more Nvidia GPUs and some of it

lot more Nvidia GPUs and some of it could be somebody's semi-custom AS6. And

could be somebody's semi-custom AS6. And

could be somebody's semi-custom AS6. And those systems are so insanely hard to

those systems are so insanely hard to

those systems are so insanely hard to build. And they're all missing this one

build. And they're all missing this one

build. And they're all missing this one incredible ingredient. This incredible

incredible ingredient. This incredible

incredible ingredient. This incredible ingredient called MVLink. MVLink so that

ingredient called MVLink. MVLink so that

ingredient called MVLink. MVLink so that you could scale up these semi-custom

you could scale up these semi-custom

you could scale up these semi-custom systems and build really powerful

systems and build really powerful

systems and build really powerful computers. And so today we're announcing

computers. And so today we're announcing

computers. And so today we're announcing the MVLink Fusion. MVLink Fusion kind of

the MVLink Fusion. MVLink Fusion kind of

the MVLink Fusion. MVLink Fusion kind of works like this. This is the Nvidia

works like this. This is the Nvidia

works like this. This is the Nvidia platform, 100% Nvidia. You got Nvidia

platform, 100% Nvidia. You got Nvidia

platform, 100% Nvidia. You got Nvidia CPU, Nvidia GPU, the MVLink

CPU, Nvidia GPU, the MVLink

CPU, Nvidia GPU, the MVLink switches, the networking from Nvidia

switches, the networking from Nvidia

switches, the networking from Nvidia called Spectrum X or

called Spectrum X or

called Spectrum X or Infiniband,

Infiniband,

Infiniband, Nyx, network

Nyx, network

Nyx, network interconnects,

interconnects,

interconnects, switches and all of the entire system,

switches and all of the entire system,

switches and all of the entire system, the entire infrastructure built end to

the entire infrastructure built end to

the entire infrastructure built end to end. And we now today make it possible

end. And we now today make it possible

end. And we now today make it possible for you to mix and match it even at the

for you to mix and match it even at the

for you to mix and match it even at the compute level. This would be what you

compute level. This would be what you

compute level. This would be what you would do using your custom ASIC. And it

would do using your custom ASIC. And it

would do using your custom ASIC. And it doesn't have to be just a transformer

doesn't have to be just a transformer

doesn't have to be just a transformer accelerator. It could be an accelerator

accelerator. It could be an accelerator

accelerator. It could be an accelerator of any kind that you would like to

of any kind that you would like to

of any kind that you would like to integrate into a large scaleup

integrate into a large scaleup

integrate into a large scaleup system. We create an MVLink chiplet.

system. We create an MVLink chiplet.

system. We create an MVLink chiplet. It's basically a switch that a butts

It's basically a switch that a butts

It's basically a switch that a butts right up to your chip.

right up to your chip.

right up to your chip. There's IP that will be available to

There's IP that will be available to

There's IP that will be available to integrate into your semi-custom ASIC.

integrate into your semi-custom ASIC.

integrate into your semi-custom ASIC. You've been building your own CPU for

You've been building your own CPU for

You've been building your own CPU for some time and maybe your CPU has built a

some time and maybe your CPU has built a

some time and maybe your CPU has built a very large ecosystem and you would like

very large ecosystem and you would like

very large ecosystem and you would like to integrate Nvidia into your

to integrate Nvidia into your

to integrate Nvidia into your ecosystem and now we can make it

ecosystem and now we can make it

ecosystem and now we can make it possible for you to do that. You could

possible for you to do that. You could

possible for you to do that. You could do that by building a your custom CPU.

do that by building a your custom CPU.

do that by building a your custom CPU. We provide you with our MVLink chip to

We provide you with our MVLink chip to

We provide you with our MVLink chip to chip interface into your ASIC. We

chip interface into your ASIC. We

chip interface into your ASIC. We connect it with MVLink chiplets and now

connect it with MVLink chiplets and now

connect it with MVLink chiplets and now it connects and directly abuts

it connects and directly abuts

it connects and directly abuts into the Blackwell chips and our next

into the Blackwell chips and our next

into the Blackwell chips and our next generation Reuben chips. And again, it

generation Reuben chips. And again, it

generation Reuben chips. And again, it fits right into this ecosystem. This

fits right into this ecosystem. This

fits right into this ecosystem. This incredible body of work now becomes

incredible body of work now becomes

incredible body of work now becomes flexible and open for everybody to

flexible and open for everybody to

flexible and open for everybody to integrate into. you instantly get

integrate into. you instantly get

integrate into. you instantly get integrated into the entire larger Nvidia

integrated into the entire larger Nvidia

integrated into the entire larger Nvidia ecosystem that makes it possible for you

ecosystem that makes it possible for you

ecosystem that makes it possible for you to scale up into these AI

to scale up into these AI

to scale up into these AI supercomputers. Now, let me talk to you

supercomputers. Now, let me talk to you

supercomputers. Now, let me talk to you about some new product categories. As

about some new product categories. As

about some new product categories. As you know, I've shown you a couple of

you know, I've shown you a couple of

you know, I've shown you a couple of different

different

different computers.

computers.

computers. However, in order to serve the vast

However, in order to serve the vast

However, in order to serve the vast majority of the world, there are still

majority of the world, there are still

majority of the world, there are still some computers that are missing. And so,

some computers that are missing. And so,

some computers that are missing. And so, I'm going to talk about them. This new

I'm going to talk about them. This new

I'm going to talk about them. This new computer we call DGX Spark is in full

computer we call DGX Spark is in full

computer we call DGX Spark is in full production. DJX

Spark will be ready, will be available

Spark will be ready, will be available

Spark will be ready, will be available shortly, probably in a few weeks. We

shortly, probably in a few weeks. We

shortly, probably in a few weeks. We have tremendous partners working with

have tremendous partners working with

have tremendous partners working with us. Dell,

us. Dell,

us. Dell, HPI, Asus, MSI,

HPI, Asus, MSI,

HPI, Asus, MSI, Gigabyte,

Gigabyte,

Gigabyte, Lenovo, incredible partners with working

Lenovo, incredible partners with working

Lenovo, incredible partners with working with us. And this is the DJX Spark. This

with us. And this is the DJX Spark. This

with us. And this is the DJX Spark. This is actually a production unit. This is

is actually a production unit. This is

is actually a production unit. This is our version. This is our version.

our version. This is our version.

our version. This is our version. However, our partners are building a

However, our partners are building a

However, our partners are building a whole bunch of different

whole bunch of different

whole bunch of different versions. This is designed for AI native

versions. This is designed for AI native

versions. This is designed for AI native developers. If you're a developer,

developers. If you're a developer,

developers. If you're a developer, you're a student, you're a researcher,

you're a student, you're a researcher,

you're a student, you're a researcher, and you don't want to keep opening up

and you don't want to keep opening up

and you don't want to keep opening up the cloud and getting it prepared and

the cloud and getting it prepared and

the cloud and getting it prepared and then when you're done scrubbing it,

then when you're done scrubbing it,

then when you're done scrubbing it, okay, but you would just like to have

okay, but you would just like to have

okay, but you would just like to have your own basically your own AI cloud

your own basically your own AI cloud

your own basically your own AI cloud sitting right next to you and it's

sitting right next to you and it's

sitting right next to you and it's always on, always waiting for you. It

always on, always waiting for you. It

always on, always waiting for you. It allows you to do your prototyping early

allows you to do your prototyping early

allows you to do your prototyping early development. And this is what's amazing.

development. And this is what's amazing.

development. And this is what's amazing. This is um DGX Spark. It's one

This is um DGX Spark. It's one

This is um DGX Spark. It's one pedlops and 128

pedlops and 128

pedlops and 128 gigabytes. In

gigabytes. In

gigabytes. In 2016, when I delivered DGX1, this is

2016, when I delivered DGX1, this is

2016, when I delivered DGX1, this is just the bezel. I can't lift a whole

just the bezel. I can't lift a whole

just the bezel. I can't lift a whole computer. It's 300 lb. This is

computer. It's 300 lb. This is

computer. It's 300 lb. This is DGX1. This is one

DGX1. This is one

DGX1. This is one pedlops and 128 gigabytes. Of course,

pedlops and 128 gigabytes. Of course,

pedlops and 128 gigabytes. Of course, this is 128 gigabytes of HBM memory and

this is 128 gigabytes of HBM memory and

this is 128 gigabytes of HBM memory and this is 128 GB of

this is 128 GB of

this is 128 GB of LPDDR5X. The performance is in fact

LPDDR5X. The performance is in fact

LPDDR5X. The performance is in fact quite similar. But what's most important

quite similar. But what's most important

quite similar. But what's most important is that the work that you could do you

is that the work that you could do you

is that the work that you could do you could work on this is the same work you

could work on this is the same work you

could work on this is the same work you could do here. It's an incredible

could do here. It's an incredible

could do here. It's an incredible achievement over just the course of

achievement over just the course of

achievement over just the course of about 10 years. Okay, so this is DGX

about 10 years. Okay, so this is DGX

about 10 years. Okay, so this is DGX Spark for anybody who would like to have

Spark for anybody who would like to have

Spark for anybody who would like to have their

their

their own AI supercomputer. If that one isn't

own AI supercomputer. If that one isn't

own AI supercomputer. If that one isn't big enough for you, here's one. This is

big enough for you, here's one. This is

big enough for you, here's one. This is another desk side. This is also going to

another desk side. This is also going to

another desk side. This is also going to be available from Dell and HPI, Asus,

be available from Dell and HPI, Asus,

be available from Dell and HPI, Asus, Gigabyte, MSI, Lenovo, amazing

Gigabyte, MSI, Lenovo, amazing

Gigabyte, MSI, Lenovo, amazing workstation companies. And this is going

workstation companies. And this is going

workstation companies. And this is going to be your own personal

to be your own personal

to be your own personal DGX

DGX

DGX supercomput. This computer is the most

supercomput. This computer is the most

supercomput. This computer is the most performance you can possibly get out of

performance you can possibly get out of

performance you can possibly get out of a wall socket. You could put this in

a wall socket. You could put this in

a wall socket. You could put this in your

your

your kitchen, but just

kitchen, but just

kitchen, but just barely. If you put this in your kitchen

barely. If you put this in your kitchen

barely. If you put this in your kitchen and then somebody runs the microwave, I

and then somebody runs the microwave, I

and then somebody runs the microwave, I think that's the limit. And so this is

think that's the limit. And so this is

think that's the limit. And so this is the limit. This is the limit of what you

the limit. This is the limit of what you

the limit. This is the limit of what you can get out of a wall outlet. And this

can get out of a wall outlet. And this

can get out of a wall outlet. And this is a DGX station. The programming model

is a DGX station. The programming model

is a DGX station. The programming model of this and the giant systems that I

of this and the giant systems that I

of this and the giant systems that I showed you are the same. That's the

showed you are the same. That's the

showed you are the same. That's the amazing thing. one architecture, one

amazing thing. one architecture, one

amazing thing. one architecture, one architecture. And this has the ability,

architecture. And this has the ability,

architecture. And this has the ability, enough capacity and performance to run a

enough capacity and performance to run a

enough capacity and performance to run a one trillion parameter AI model.

one trillion parameter AI model.

one trillion parameter AI model. Remember Llama is Llama 70B. A one

Remember Llama is Llama 70B. A one

Remember Llama is Llama 70B. A one trillion parameter model is going to run

trillion parameter model is going to run

trillion parameter model is going to run wonderfully on this machine. Okay, so

wonderfully on this machine. Okay, so

wonderfully on this machine. Okay, so that's the DGX station. But in order for

that's the DGX station. But in order for

that's the DGX station. But in order for us to bring AI into a new world and this

us to bring AI into a new world and this

us to bring AI into a new world and this new world is enterprise IT, we have to

new world is enterprise IT, we have to

new world is enterprise IT, we have to go back to our roots and we have to

go back to our roots and we have to

go back to our roots and we have to reinvent computing and bring AI into

reinvent computing and bring AI into

reinvent computing and bring AI into traditional enterprise computing. And so

traditional enterprise computing. And so

traditional enterprise computing. And so let's take a look at that.

Okay, this is

This is the brand new RTX Pro. RTX Pro

This is the brand new RTX Pro. RTX Pro

This is the brand new RTX Pro. RTX Pro Enterprise and Omniverse server. This

Enterprise and Omniverse server. This

Enterprise and Omniverse server. This server can run everything. Everything

server can run everything. Everything

server can run everything. Everything that runs in the world today should run

that runs in the world today should run

that runs in the world today should run here. Omniverse runs on here perfectly.

here. Omniverse runs on here perfectly.

here. Omniverse runs on here perfectly. But in addition to that, in addition to

But in addition to that, in addition to

But in addition to that, in addition to that, this is the

that, this is the

that, this is the computer for enterprise AI

computer for enterprise AI

computer for enterprise AI agents. Those AI agents could be only

agents. Those AI agents could be only

agents. Those AI agents could be only text. Those AI agents could also be

text. Those AI agents could also be

text. Those AI agents could also be computer graphics, could be in video

computer graphics, could be in video

computer graphics, could be in video form. All of those workloads work on

form. All of those workloads work on

form. All of those workloads work on this system. No matter the modality,

this system. No matter the modality,

this system. No matter the modality, every single model that we know of in

every single model that we know of in

every single model that we know of in the world, every application that we

the world, every application that we

the world, every application that we know of should run on this. In fact,

know of should run on this. In fact,

know of should run on this. In fact, even Crisis works on here. Okay? So,

even Crisis works on here. Okay? So,

even Crisis works on here. Okay? So, anybody who's a GeForce

anybody who's a GeForce

anybody who's a GeForce gamer, there are no GeForce gamer in the

gamer, there are no GeForce gamer in the

gamer, there are no GeForce gamer in the room. What connects these eight GPUs,

room. What connects these eight GPUs,

room. What connects these eight GPUs, the Blackwell, new Blackwell RTX, RTX

the Blackwell, new Blackwell RTX, RTX

the Blackwell, new Blackwell RTX, RTX Pro 6000s, is this new motherboard. This

Pro 6000s, is this new motherboard. This

Pro 6000s, is this new motherboard. This new motherboard is actually a switched

new motherboard is actually a switched

new motherboard is actually a switched network. CX8 is a new category of chips.

network. CX8 is a new category of chips.

network. CX8 is a new category of chips. It's a switch first, networking chip

It's a switch first, networking chip

It's a switch first, networking chip second. It's also the most advanced

second. It's also the most advanced

second. It's also the most advanced networking chip in the world. So each

networking chip in the world. So each

networking chip in the world. So each one of these GPUs have their own

one of these GPUs have their own

one of these GPUs have their own networking interface. All of the GPUs

networking interface. All of the GPUs

networking interface. All of the GPUs are now communicating to all of the

are now communicating to all of the

are now communicating to all of the other GPUs on east west traffic.

other GPUs on east west traffic.

other GPUs on east west traffic. Incredible performance. Okay, let's talk

Incredible performance. Okay, let's talk

Incredible performance. Okay, let's talk about robots.

about robots.

about robots. So agent AIs, agentic AIs, AI agents, a

So agent AIs, agentic AIs, AI agents, a

So agent AIs, agentic AIs, AI agents, a lot of different ways to say it. Agents

lot of different ways to say it. Agents

lot of different ways to say it. Agents are essentially digital robots. Reason

are essentially digital robots. Reason

are essentially digital robots. Reason for that is because a robot perceives,

for that is because a robot perceives,

for that is because a robot perceives, understands, and plans. And that's

understands, and plans. And that's

understands, and plans. And that's essentially what agents do. But we would

essentially what agents do. But we would

essentially what agents do. But we would like to build also physical robots. And

like to build also physical robots. And

like to build also physical robots. And these physical robots first it starts

these physical robots first it starts

these physical robots first it starts with the ability to learn to be a robot.

with the ability to learn to be a robot.

with the ability to learn to be a robot. Now you know that we've been working in

Now you know that we've been working in

Now you know that we've been working in in autonomous systems for some time. Our

in autonomous systems for some time. Our

in autonomous systems for some time. Our self-driving car basically has three

self-driving car basically has three

self-driving car basically has three systems. There's the system for creating

systems. There's the system for creating

systems. There's the system for creating the AI model and that's GB200, GB300.

the AI model and that's GB200, GB300.

the AI model and that's GB200, GB300. It's going to be used for that, training

It's going to be used for that, training

It's going to be used for that, training the AI model. Then you have Omniverse

the AI model. Then you have Omniverse

the AI model. Then you have Omniverse for simulating the AI model. And then

for simulating the AI model. And then

for simulating the AI model. And then when you're done with that AI model, you

when you're done with that AI model, you

when you're done with that AI model, you put that model, the AI into the

put that model, the AI into the

put that model, the AI into the self-driving car. So we're doing exactly

self-driving car. So we're doing exactly

self-driving car. So we're doing exactly the same thing in robotic systems just

the same thing in robotic systems just

the same thing in robotic systems just like cars. And so this is our Isaac

like cars. And so this is our Isaac

like cars. And so this is our Isaac Groot platform. The simulation is

Groot platform. The simulation is

Groot platform. The simulation is exactly the same. It's omniverse. The

exactly the same. It's omniverse. The

exactly the same. It's omniverse. The compute, the training system is the

compute, the training system is the

compute, the training system is the same. When you're done with the model,

same. When you're done with the model,

same. When you're done with the model, you put it into inside this Isaac group

you put it into inside this Isaac group

you put it into inside this Isaac group platform. It Isaac Group platform starts

platform. It Isaac Group platform starts

platform. It Isaac Group platform starts with a brand new computer called Jetson

with a brand new computer called Jetson

with a brand new computer called Jetson Thor. This is just started in

Thor. This is just started in

Thor. This is just started in production. It is an incredible

production. It is an incredible

production. It is an incredible processor. Basically, a robotic

processor. Basically, a robotic

processor. Basically, a robotic processor goes to self-driving cars and

processor goes to self-driving cars and

processor goes to self-driving cars and it goes into a human or robotic system.

it goes into a human or robotic system.

it goes into a human or robotic system. On top is an operating system we called

On top is an operating system we called

On top is an operating system we called Isaac Nvidia Isaac. The Nvidia Isaac

Isaac Nvidia Isaac. The Nvidia Isaac

Isaac Nvidia Isaac. The Nvidia Isaac operating system is the runtime. It does

operating system is the runtime. It does

operating system is the runtime. It does all of the neuronet network processing,

all of the neuronet network processing,

all of the neuronet network processing, the sensor processing, pipelines all of

the sensor processing, pipelines all of

the sensor processing, pipelines all of it and deliver actuated results. The

it and deliver actuated results. The

it and deliver actuated results. The biggest challenge in robotics is and

biggest challenge in robotics is and

biggest challenge in robotics is and well the biggest challenge in AI overall

well the biggest challenge in AI overall

well the biggest challenge in AI overall is what is your data

is what is your data

is what is your data strategy and your data strategy has to

strategy and your data strategy has to

strategy and your data strategy has to be that's where great deal of research

be that's where great deal of research

be that's where great deal of research and a great deal of technology goes into

and a great deal of technology goes into

and a great deal of technology goes into in the case of robotics human

in the case of robotics human

in the case of robotics human demonstration just like we demonstrate

demonstration just like we demonstrate

demonstration just like we demonstrate to our children or a a coach

to our children or a a coach

to our children or a a coach demonstrates to an athlete you

demonstrates to an athlete you

demonstrates to an athlete you demonstrate using

demonstrate using

demonstrate using teleoperations you demonstrate to the

teleoperations you demonstrate to the

teleoperations you demonstrate to the robot how to perform the task and the

robot how to perform the task and the

robot how to perform the task and the robot can generalize ize from that

robot can generalize ize from that

robot can generalize ize from that demonstration because AI can generalize

demonstration because AI can generalize

demonstration because AI can generalize and we have technology for

and we have technology for

and we have technology for generalization. You can generalize from

generalization. You can generalize from

generalization. You can generalize from that one demonstration other techniques.

that one demonstration other techniques.

that one demonstration other techniques. Okay. And so what

Okay. And so what

Okay. And so what if what if you want to teach this robot

if what if you want to teach this robot

if what if you want to teach this robot a whole bunch of skills? How many

a whole bunch of skills? How many

a whole bunch of skills? How many different teleoperation people do you

different teleoperation people do you

different teleoperation people do you need? Well, it turns out to be a lot.

need? Well, it turns out to be a lot.

need? Well, it turns out to be a lot. And so what we decided to do was use AI

And so what we decided to do was use AI

And so what we decided to do was use AI to

to

to amplify the human demonstration systems.

amplify the human demonstration systems.

amplify the human demonstration systems. And so this is essentially going from

And so this is essentially going from

And so this is essentially going from real to real and using an AI to help

real to real and using an AI to help

real to real and using an AI to help us expand amplify the amount of data

us expand amplify the amount of data

us expand amplify the amount of data that was collected during human

that was collected during human

that was collected during human demonstration to train an AI

demonstration to train an AI

demonstration to train an AI model. Is that correct?

model. Is that correct?

model. Is that correct? So in order for robotics to happen, you

So in order for robotics to happen, you

So in order for robotics to happen, you need you need AI. But in order to teach

need you need AI. But in order to teach

need you need AI. But in order to teach the AI, you need AI. And so this is

the AI, you need AI. And so this is

the AI, you need AI. And so this is really the great thing

really the great thing

really the great thing about the era of agents where we need a

about the era of agents where we need a

about the era of agents where we need a a large amount of synthetic data

a large amount of synthetic data

a large amount of synthetic data generation, robotics, a large amount of

generation, robotics, a large amount of

generation, robotics, a large amount of synthetic data generation and skill

synthetic data generation and skill

synthetic data generation and skill learning called fine-tuning, which is a

learning called fine-tuning, which is a

learning called fine-tuning, which is a lot of reinforcement learning and

lot of reinforcement learning and

lot of reinforcement learning and enormous amount of compute. And so this

enormous amount of compute. And so this

enormous amount of compute. And so this is an er this is a whole era where the

is an er this is a whole era where the

is an er this is a whole era where the training of these AI the development of

training of these AI the development of

training of these AI the development of these AI as well as the running of the

these AI as well as the running of the

these AI as well as the running of the AI needs an enormous amount of compute.

AI needs an enormous amount of compute.

AI needs an enormous amount of compute. Well, as I mentioned earlier, the world

Well, as I mentioned earlier, the world

Well, as I mentioned earlier, the world has a severe shortage of labor. And the

has a severe shortage of labor. And the

has a severe shortage of labor. And the reason why humano robotics is so

reason why humano robotics is so

reason why humano robotics is so important is because it is the only form

important is because it is the only form

important is because it is the only form of robot that can be deployed almost

of robot that can be deployed almost

of robot that can be deployed almost anywhere, brownfield. It doesn't have to

anywhere, brownfield. It doesn't have to

anywhere, brownfield. It doesn't have to be green field. It could fit into the

be green field. It could fit into the

be green field. It could fit into the world we created. It could do the task

world we created. It could do the task

world we created. It could do the task that we made for ourselves. We

that we made for ourselves. We

that we made for ourselves. We engineered the world for ourselves and

engineered the world for ourselves and

engineered the world for ourselves and now we could create a robot that fit

now we could create a robot that fit

now we could create a robot that fit into that world to help us. Now the

into that world to help us. Now the

into that world to help us. Now the amazing thing about human robotics is

amazing thing about human robotics is

amazing thing about human robotics is not just the fact that if it worked it

not just the fact that if it worked it

not just the fact that if it worked it could be quite versatile. It is likely

could be quite versatile. It is likely

could be quite versatile. It is likely the only robot that is likely to work

the only robot that is likely to work

the only robot that is likely to work and the reason for that is because

and the reason for that is because

and the reason for that is because technology needs scale.

technology needs scale.

technology needs scale. Most of the robotic systems we've had so

Most of the robotic systems we've had so

Most of the robotic systems we've had so far are too low volume. And those low

far are too low volume. And those low

far are too low volume. And those low volume systems will never achieve the

volume systems will never achieve the

volume systems will never achieve the technology scale to get the flywheel

technology scale to get the flywheel

technology scale to get the flywheel going far enough fast enough so that

going far enough fast enough so that

going far enough fast enough so that we're willing to dedicate enough

we're willing to dedicate enough

we're willing to dedicate enough technology into it to make it better.

technology into it to make it better.

technology into it to make it better. But human or robot, it is likely to be

But human or robot, it is likely to be

But human or robot, it is likely to be the next multi-trillion dollar industry.

the next multi-trillion dollar industry.

the next multi-trillion dollar industry. And the technology innovation is

And the technology innovation is

And the technology innovation is incredibly fast and the consumption of

incredibly fast and the consumption of

incredibly fast and the consumption of computing and data centers enormous. But

computing and data centers enormous. But

computing and data centers enormous. But this is one of those applications that

this is one of those applications that

this is one of those applications that needs three computers. One computer is

needs three computers. One computer is

needs three computers. One computer is an AI for learning. One computer is a

an AI for learning. One computer is a

an AI for learning. One computer is a simulation engine where the AI could

simulation engine where the AI could

simulation engine where the AI could learn how to be a robot in a uh in a

learn how to be a robot in a uh in a

learn how to be a robot in a uh in a virtual environment. And then also the

virtual environment. And then also the

virtual environment. And then also the deployment of it. Everything that moves

deployment of it. Everything that moves

deployment of it. Everything that moves will be

will be

will be robotic. As we put these robots into the

robotic. As we put these robots into the

robotic. As we put these robots into the factories, remember the factories are

factories, remember the factories are

factories, remember the factories are also robotic.

also robotic.

also robotic. Today's factories are so incredibly

Today's factories are so incredibly

Today's factories are so incredibly complex. This is

complex. This is

complex. This is Delta's manufacturing line and they're

Delta's manufacturing line and they're

Delta's manufacturing line and they're getting it ready for a robotic future.

getting it ready for a robotic future.

getting it ready for a robotic future. It is already robotics and software

It is already robotics and software

It is already robotics and software defined and now in the future there will

defined and now in the future there will

defined and now in the future there will be robots working in it. In order for us

be robots working in it. In order for us

be robots working in it. In order for us to create robots and design robots that

to create robots and design robots that

to create robots and design robots that operate in and as a fleet, as a team,

operate in and as a fleet, as a team,

operate in and as a fleet, as a team, working together in a factory that is

working together in a factory that is

working together in a factory that is also robotic, we have to give it

also robotic, we have to give it

also robotic, we have to give it omniverse to learn how

omniverse to learn how

omniverse to learn how to work together. And that digital twin,

to work together. And that digital twin,

to work together. And that digital twin, you now have a digital twin of the

you now have a digital twin of the

you now have a digital twin of the robot. You have a digital twin of all of

robot. You have a digital twin of all of

robot. You have a digital twin of all of the equipment. You're going to have

the equipment. You're going to have

the equipment. You're going to have digital twin of the factory. Those

digital twin of the factory. Those

digital twin of the factory. Those nested digital twins are going to be

nested digital twins are going to be

nested digital twins are going to be part of what Omniverse is able to do.

part of what Omniverse is able to do.

part of what Omniverse is able to do. These are all digital twins. They're all

These are all digital twins. They're all

These are all digital twins. They're all simulations. They just look beautiful.

simulations. They just look beautiful.

simulations. They just look beautiful. The image just looks beautiful, but it's

The image just looks beautiful, but it's

The image just looks beautiful, but it's they're all digital twins. TSMC is

they're all digital twins. TSMC is

they're all digital twins. TSMC is building a dig digital twin of their

building a dig digital twin of their

building a dig digital twin of their next fab.

As we

As we

As we speak, there are five trillion dollars

speak, there are five trillion dollars

speak, there are five trillion dollars of plants being planned around the world

of plants being planned around the world

of plants being planned around the world over the next three years. Five trillion

over the next three years. Five trillion

over the next three years. Five trillion dollars of new plants because the world

dollars of new plants because the world

dollars of new plants because the world is reshaping because

is reshaping because

is reshaping because re-industrialization moving around the

re-industrialization moving around the

re-industrialization moving around the world. New plants are being built

world. New plants are being built

world. New plants are being built everywhere. This is an enormous

everywhere. This is an enormous

everywhere. This is an enormous opportunity for us to make sure that

opportunity for us to make sure that

opportunity for us to make sure that they build it well and cost effectively

they build it well and cost effectively

they build it well and cost effectively and on time. And so putting everything

and on time. And so putting everything

and on time. And so putting everything into a digital twin is really a great

into a digital twin is really a great

into a digital twin is really a great first step and preparing it for a

first step and preparing it for a

first step and preparing it for a robotic future. In fact, building that

robotic future. In fact, building that

robotic future. In fact, building that $5 trillion doesn't include a new type

$5 trillion doesn't include a new type

$5 trillion doesn't include a new type of factory that we're building. And even

of factory that we're building. And even

of factory that we're building. And even our own factories, we put in a digital

our own factories, we put in a digital

our own factories, we put in a digital twin. This is the Nvidia AI factory in a

twin. This is the Nvidia AI factory in a

twin. This is the Nvidia AI factory in a digital

digital

digital twin. Gong is a digital twin. They made

twin. Gong is a digital twin. They made

twin. Gong is a digital twin. They made Gausong a digital twin.

Gausong a digital twin.

Gausong a digital twin. There are already hundreds of thousands

There are already hundreds of thousands

There are already hundreds of thousands of buildings, millions of miles of

of buildings, millions of miles of

of buildings, millions of miles of roads. And so, yes, Gaus is a digital

roads. And so, yes, Gaus is a digital

roads. And so, yes, Gaus is a digital twin. We are at a once- ina-lifetime

twin. We are at a once- ina-lifetime

twin. We are at a once- ina-lifetime opportunity. It is not it is not an

opportunity. It is not it is not an

opportunity. It is not it is not an understatement to say that the

understatement to say that the

understatement to say that the opportunity ahead of us is

opportunity ahead of us is

opportunity ahead of us is extraordinary. For the very first time

extraordinary. For the very first time

extraordinary. For the very first time in all of our time together, not only

in all of our time together, not only

in all of our time together, not only are we

are we

are we creating the next generation of IT,

creating the next generation of IT,

creating the next generation of IT, we've done that several times from PC to

we've done that several times from PC to

we've done that several times from PC to internet to cloud to mobile cloud. We've

internet to cloud to mobile cloud. We've

internet to cloud to mobile cloud. We've done that several times. But this time,

done that several times. But this time,

done that several times. But this time, not only are we creating the next

not only are we creating the next

not only are we creating the next generation of IT, we are in fact

generation of IT, we are in fact

generation of IT, we are in fact creating a whole new industry. This

creating a whole new industry. This

creating a whole new industry. This whole new

whole new

whole new industry is going to expose us to giant

industry is going to expose us to giant

industry is going to expose us to giant opportunities ahead. I look forward to

opportunities ahead. I look forward to

opportunities ahead. I look forward to partnering with all of you on building

partnering with all of you on building

partnering with all of you on building AI factories, agents for enterprises,

AI factories, agents for enterprises,

AI factories, agents for enterprises, robots, all of you amazing partners,

robots, all of you amazing partners,

robots, all of you amazing partners, building the ecosystem with us around

building the ecosystem with us around

building the ecosystem with us around one architecture. And so I want to thank

one architecture. And so I want to thank

one architecture. And so I want to thank all of you for coming today. Have a

all of you for coming today. Have a

all of you for coming today. Have a great Computex everybody.

great Computex everybody.

great Computex everybody. [Applause]