GET IN EARLY! I'm Investing In This HUGE AI Chip Breakthrough
Summary
TLDRThe video discusses the potential of Nvidia's new Blackwell GPUs, which are set to revolutionize the computing industry with their advanced performance and efficiency. The presenter argues that Nvidia's end-to-end AI computing platform, highlighted at the GTC conference, positions the company to surpass even Apple and Microsoft in market value. The Blackwell GPUs, with their dual-die design and high-speed connectivity, promise to transform data centers and supercomputers, making significant strides in AI training and inference. The presenter's enthusiasm for Nvidia's technology leads them to invest in the company's stock, anticipating its long-term dominance in the generative AI era.
Takeaways
- π Nvidia's Blackwell GPUs represent a significant leap in computing capabilities, with the potential to outperform and outscale existing technologies.
- π‘ The Blackwell architecture introduces a dual-die design, allowing for unprecedented data transfer rates and performance improvements over its predecessor, Hopper.
- π§ Blackwell's design costs are higher due to its complexity, which may initially result in lower profit margins, but its performance could offset this with strong market demand.
- π Nvidia is positioning itself as an end-to-end AI computing platform, not just an AI hardware company, showcasing a wide range of software solutions at GTC.
- π The Blackwell GPUs, along with Nvidia's networking technologies, enable the creation of massive, scalable data centers capable of handling the most demanding AI workloads.
- πΌ The speaker believes Nvidia's data center hardware is at the center of a major technology transformation, comparing the potential impact of Blackwell to the iPhone's influence on mobile computing.
- π The speaker is bullish on Nvidia's stock, moving it up on their list of stocks to invest in, based on the company's strong product lineup and the anticipated impact of Blackwell.
- π Blackwell's performance advancements could allow for more efficient training of AI models, reducing the cost and energy consumption associated with computing.
- π’ Nvidia's Blackwell GPUs and data center solutions are expected to benefit a variety of industries, including self-driving cars, smart factories, and even humanoid robots.
- π The speaker's investment strategy emphasizes understanding the science and technology behind companies like Nvidia, as a key to long-term wealth generation.
Q & A
What is the main assertion about Nvidia's potential in the transcript?
-The main assertion is that Nvidia has the potential to become the biggest company on Earth, surpassing Apple and Microsoft, due to its innovative products and technology, particularly in the realm of AI and data center hardware.
What are some key takeaways from the GTC conference regarding Nvidia's business?
-Nvidia showcased its end-to-end AI computing platform with a focus on software, interviewed executives, attended expert panels, and got hands-on with new products and services. The company is involved in various sectors from self-driving cars to supercomputers and humanoid robots.
What are the two variants of Nvidia's Blackwell GPUs?
-The two variants of Nvidia's Blackwell GPUs are the B100 and the B200. The B100 is a drop-in replacement for the H100, while the B200 is a new, state-of-the-art chip with improved performance.
How does the Blackwell GPU architecture differ from its predecessor, Hopper?
-Blackwell architecture scales better than Hopper in data centers and supercomputers, offering an 80% increase in speed while using the same power. It also features a unique design that combines two dies into a single chip connected by a 10 terabyte per second link.
What are the implications of the Blackwell GPU's higher unit costs?
-The higher unit costs of the Blackwell GPU mean lower margins for Nvidia. However, the significant performance improvements give the company pricing power and the ability to attract more businesses to the CUDA platform.
How does Nvidia's GB200 super chip connect the Blackwell GPUs?
-The GB200 super chip pairs two B200 GPUs with one Grace server CPU, connected via an NV link, which provides a chip-to-chip connection with 900 GB per second of bandwidth.
What is the significance of Nvidia's Envy link switch chip?
-The Envy link switch chip, with 50 billion transistors, enables every GPU to communicate with every other GPU at full speed simultaneously, thanks to its four MV links, each capable of 1.8 terabytes per second.
How does the GB200 NVL 72 system scale to meet market demands?
-The GB200 NVL 72 system scales by combining 18 compute trays with two B200 GPUs each, connected via NV link, and nine Envy link switch trays, supporting all-to-all GPU communications at 130 terabytes per second.
What is the impact of Blackwell GPUs on training large AI models?
-Blackwell GPUs can train an 1.8 trillion parameter GPT model in 90 days with only 2,000 GPUs and 4 megawatts of power, compared to 8,000 GPUs and 15 megawatts with previous technology.
What is the speaker's conclusion about investing in Nvidia stock?
-The speaker concludes that due to the advancements and potential applications of Blackwell GPUs, Nvidia stock is a strong investment that will likely perform well in the long term, leading to the decision to move Nvidia up one spot on their list of stocks to invest in.
Outlines
π Introduction to Nvidia's Potential and Blackwell GPUs
The speaker returns from GTC with a strong conviction that Wall Street undervalues Nvidia. They emphasize the importance of understanding a company's products over just its profits and predict Nvidia's growth to surpass Apple and Microsoft. The speaker plans to discuss Nvidia's new Blackwell GPUs, their design, costs, margins, and Nvidia's networking technologies. They also mention their personal investment strategy and share some insights from GTC, positioning Nvidia at the center of a major technology transformation.
π‘ Blackwell GPUs: Performance, Design, and Market Strategy
The speaker delves into the specifics of Nvidia's Blackwell GPUs, highlighting their two variants, the B100 and the B200. They discuss the B100's compatibility with existing infrastructure and its significant performance improvement over the H100. The B200 is introduced as a cutting-edge chip with superior performance at a higher power draw. The speaker also addresses the manufacturing costs and margins associated with Blackwell GPUs and suggests that future improvements in chip fabrication processes could lead to more cost-effective solutions. They emphasize the scalability of Blackwell systems in creating comprehensive data center solutions.
π Nvidia's Networking Innovations and Data Center Scaling
The speaker explains the role of Nvidia's networking technologies, including the Envy link switch and Bluefield DPUs, in scaling Blackwell GPUs to create powerful data center and supercomputer solutions. They describe the capabilities of the GB200 super chip and its integration with the Grace CPU, emphasizing the high bandwidth and computational efficiency. The speaker also details the structure of the GB200 NVL 72 system, which combines multiple GPUs and networking technologies to achieve unprecedented computational and communication speeds within a single rack.
π Nvidia's Stock Prospects and the Future of Computing
The speaker concludes by discussing the implications of Blackwell GPUs for Nvidia's stock and the broader impact on the future of computing. They express a belief that Blackwell will significantly advance Nvidia's position in the market, with applications extending beyond data centers to autonomous vehicles, smart factories, and humanoid robots. The speaker shares their personal decision to move Nvidia up on their list of stocks to invest in, based on the transformative potential of Blackwell and its alignment with the ongoing technology revolution.
Mindmap
Keywords
π‘GTC
π‘Nvidia
π‘Blackwell GPUs
π‘AI Computing
π‘Data Centers
π‘Investing
π‘Stocks
π‘Performance
π‘Costs and Margins
π‘Networking Technologies
π‘Mental Health
Highlights
The speaker is convinced that Wall Street does not understand Nvidia's potential, predicting it will become the biggest company on Earth.
The key to finding great stocks is understanding a company's products rather than just their profits.
Nvidia is not just an AI hardware company; they are an end-to-end AI computing platform.
Nvidia showcased a lot of software at GTC, indicating a comprehensive approach to AI solutions.
The speaker learned about every part of Nvidia's business, including interviewing executives and attending expert panels.
Nvidia's data center hardware is at the center of a major technology transformation.
Blackwell and Hopper are predicted to be the platforms of the generative AI era, similar to how the iPhone made Apple the biggest company in the world.
The Blackwell GPU is Nvidia's successor to the Hopper GPUs, offering significant advancements in computing power.
Blackwell GPUs have two variants, B100 and B200, each offering different improvements and use cases.
The B100 variant is an 80% faster drop-in replacement for the H100, which is beneficial for ramping up sales.
The B200 variant is a state-of-the-art chip offering 10-25% faster performance than the B100 at a higher power draw.
Blackwell GPUs perform four times better at training AI models and an impressive 30 times better for AI inference.
The Blackwell GPU combines two dies into a single chip with a 10 terabyte per second link, showing innovative chip design.
Nvidia's chips are made at TSMC on a custom process node called N4P, indicating a specialized manufacturing process.
Blackwell systems scale to create full data center and supercomputer-sized solutions, demonstrating Nvidia's commitment to large-scale computing.
The GB200 super chip pairs two B200 GPUs with a Grace server CPU, showcasing a compact yet powerful computing solution.
Nvidia's Envy link switch chip enables every GPU to communicate at full speed simultaneously, highlighting their advanced networking technology.
The GB200 NVL 72 system is a single rack with 1.4 exaflops of AI training performance, positioning it as a significant technological achievement.
Nvidia's goal is to reduce the cost and energy associated with computing, allowing for more efficient and sustainable advancements in AI.
The speaker is moving Nvidia up one spot on their list of stocks to get rich without getting lucky, showing a strong confidence in the company's future.
Transcripts
I just got back from GTC and I'm
convinced that Wall Street does not
understand Nvidia that's because the key
to finding great stocks is understanding
a company's products not just their
profits and I'm convinced that Nvidia
will be bigger than Apple bigger than
Microsoft and Nvidia stock will sit on
top of the indexes as the biggest
company on Earth let me show you why
your time is valuable so let's get right
into it first things first I'm not here
to hold you hostage here's everything
I'm going to talk about in this video
I'll cover nvidia's insane new Blackwell
gpus I'll talk about a few things that
investors need to know about Blackwell's
design costs and margins the networking
technologies that Nvidia built to scale
Blackwell data centers to mindboggling
sizes and of course whether I'm buying
Nvidia stock as a result and where I'd
put it on the list that I made at the
start of the year to get rich without
getting lucky but let me point out a few
quick things before I dive into nvidia's
insane new Blackwell gpus first Nvidia
is not just an AI hardware company
they're an endtoend AI Computing
platform and Nvidia showcased a lot of
software at GTC I spent my time at the
conference learning as much as I can
about every part of this business I
interviewed nvidia's Executives I
attended close to a dozen expert panels
and sessions I got my hands on new
products and service demos everything
that I think the mainstream media and
Wall Street analysts are going to ignore
until it's too late and as you can
imagine there's way too much to cram
into a single video from self-driving
cars and supercomputers to humanoid
robots and digital twins for heavy
Industries and even foundational AI
models that other businesses can build
on top of so I'm going to spend the next
few videos breaking it all down for you
why because the world is going through a
major technology transformation right
now that's my second Point big
businesses are changing everything from
the hardware they run on to the kinds of
tools and processes they build on top of
it to the services and products that
they sell on top of that and like I've
been saying for a couple of years now
nvidia's data center Hardware is at the
center of this transformation and third
that's why I think Nvidia will become
the biggest company on the planet just
like the iPhone becoming the platform of
the mobile Computing era made Apple the
biggest company in the world Blackwell
and Hopper will be the platforms of the
generative AI era and if you still think
that big companies can't get bigger
remember this Warren Buffett didn't
invest in Apple until 2016 10 years
after the iPhone came out Apple was the
biggest company in the world by market
cap for 5 years when Warren Buffett
bought it and Apple stock is still 50%
of his portfolio today I don't want to
wait 10 years to find the next iPhone or
hold the next Apple stock that's why I'm
making these videos now all right one
last point and then I'll dive right into
nvidia's Hardware I get asked all the
time how I make these videos and I won't
lie it can be a real Mental Health
Battle I like to do everything myself so
my main challenge is burnout I get this
brain fog that drains my energy and
crushes my productivity for days and I
didn't think that anyone could relate so
instead of talking to someone I would
just ignore it or try to power through
it which would just make me fall even
further behind that's why I'm so excited
to tell you about betterhelp the sponsor
of this video whether you're stressed
out about balancing work and family or
you're dealing with depression or
anxiety you are not alone and getting
help is nothing to be ashamed of and
honestly it's never been easier it can
be fully remote and online and after
answering a few questions better help
can usually match you with a
professional therapist within 48 hours
that's huge because they make it easy to
switch to a new therapist at no
additional cost without worrying about
insurance or who's in your network that
way you can focus on finding the right
therapist for you between running my own
business and preparing for my wedding
next year my mental health is something
I'm investing in more and more so if you
want better help with your mental health
check out the link in the description or
go to betterhelp.com slts yoou to save
10% off your first month and a big thank
you to betterhelp and to you for
supporting the Channel all right I can't
wait anymore let's talk about Blackwell
the rate at which we're advancing
Computing is insane and it's still not
fast enough so we built another
chip Hopper is fantastic
but we need bigger
gpus and so ladies and gentlemen I would
like to introduce you to a very very big
GPU this is Blackwell the successor to
nvidia's Hopper gpus and just like
Hopper Blackwell has two variants the
B100 and the b200 the B100 is a dropin
replacement for the h100 which means
that data centers and supercomputers can
swap out an h100 tray and L Al put a
B100 tray in its place that's important
for investors because it means that
Nvidia won't have a hard time ramping up
Blackwell sales since the infrastructure
to support the b100s is already deployed
anywhere that there's an h100 today the
B100 is a whopping 80% faster than the
h100 while drawing the same 700 watts of
power which means the entire Blackwell
architecture scales much better than
hopper in data centers and
supercomputers that's why they're going
to want to phase in the B1 00s over time
on the other hand the b200 is nvidia's
new state-of-the-art chip it's anywhere
from 10 to 25% faster than the b00
depending on the workload while drawing
around 1,000 Watts instead of 700
compared to Hopper these new Blackwell
chips perform a whopping four times
better at training AI models and an
insane 30 times better for AI inference
but how Nvidia made it happened is
really important for investors to
understand this is the most advanced GPU
in the world in production today
this is
Hopper this is hopper Hopper changed the
world this is
Blackwell it's okay
Hopper 28 billion transistors and so so
you could see you I can see that there's
a small line between two dieses this is
the first time two dies have abutted
like this together in such a way that
the two CH the two guys think it's one
chip there's 10 terabytes of data
between it 10 terabytes per second so
that these two these two sides of the
Blackwell Chip have no clue which side
they're on there's no memory locality
issues no cash issues it's just one
giant chip when we were told that
Blackwell's Ambitions were beyond the
limits of physics uh the engineer said
so what and so this is what what
happened so there are two important
points here first this blackwall GPU
actually combines two dieses into a
single chip and connects them with a 10
terb per second link so that they act as
one GPU this is a really clever chip
design but two dieses an ultra
high-speed link and a bigger total area
for advanced packaging means that
Blackwell costs over twice as much to
make as Hopper and higher unit costs
means lower margins personally I think
the trade-off is worth it in the short
term because Blackwell's insane
performance gives Nvidia a lot of
pricing power compared to other GPU
makers and that flexibility is what lets
them get even more businesses onto the
Cuda platform at a time where every Data
Center and cloud service provider is
scrambling to get as many accelerators
as they can during a gold rush getting
everyone on your pick and shovels is
what matters most second if you've been
watching this channel for a while you
know that nvidia's chips are made at
tsmc on a process node called n4p which
was custom made for NVIDIA so as tsmc's
chip fabrication process keeps improving
shareholders should want Nvidia to offer
a single die version of these chips or
find another packaging Solution that's
higher margin than the current
architecture not necessarily higher
performance just lower manufacturing
costs while still meeting Market demands
and speaking of Market demands let's see
how Blackwell systems scale to create
full data center and even super computer
sized Solutions because that's where the
magic really happens this is a fully
functional board and I just be careful
here this right here is I don't know $10
billion two Blackwell chips and four
Blackwell dyes connected to a Grace CPU
the grce CPU has a super fast chipto
chip link what's amazing is this
computer is the first of its kind where
this much computation fits into this
small of a place so the gb200 super chip
pairs two b200 gpus with one Grace
server CPU they're connected via NV link
which is a chipto chip connection with
900 GB per second of bandwidth that's
fast enough to move 150 featurelength 4K
movies between chips every single second
this gb200 super chip is what Jensen is
holding in his left hand while he
casually holds the10 billion initial
prototype board in his right this right
here is I don't know
10
billion yeah I'm never going to be that
cool two of those gb200 Grace Blackwell
Super Chips go into a Blackwell compute
node which is one tray in an AI data
center rack the 2GB 200 Super Chips in
each tray are also connected via nvlink
at a high level nvlink is for chipto
chip connections while infin band fiber
optics connect different compute nodes
and Trays at the server level Nvidia
also makes these blue field dpus or data
processing units here's an easy way to
think about dpus CPUs have a low amount
of cores that run one operation at a
time gpus have more cores and can run
more operations in parallel dpus have
tons of cores and can handle highly
parallel workloads like coordinating
securing analyzing and transferring data
between many different chips in a data
center so dpus end up handling most of
the networking work workloads so that
CPUs and gpus don't have to all right so
18 of these compute trays go into one
gb200 nvl 72 system remember the gb200
super chip has two Blackwell gpus on it
connected via NV link there are two
Super Chips per compute tray which are
also connected via NV link and there are
18 trays per system 2 * 2 * 18 is 72
Blackwell gpus connected by NV link
that's why it's called the gb200 nvl 72
all right so we already have a good
handle on the gb200 but what about this
NV link well it turns out that NV link
is way more than just a chipto chip
connection the rate at which we're
advancing Computing is insane and it's
still not fast enough so we built
another
chip this chip is just an incredible
chip we call it the Envy link switch
it's 50 bill transistors it's almost the
size of Hopper all by itself this switch
chip has four MV links in
it each 1.8 terabytes per second what is
this chip
for if we were to build such a chip we
can have every single GPU talk to every
other GPU at full speed at the same time
that's
insane by the way and vlink infiniband
and these Bluefield dpus all came out of
nvidia's $7 billion acquisition of
melanox back in 2019 which will probably
go down as one of the best Acquisitions
in Silicon Valley history all right so
sandwiched between the compute notes
that we just talked about are nine EnV
link switch trays Each of which have two
of these EnV link chips in them and like
we saw in the compute trays each Envy
link chip connects four Blackwell gpus
four ports per nvlink chip times two
chips per tray times nine trays is the
72 ports supporting all to all GPU
Communications at a mindboggling
bandwidth of 130 terabytes per second
that's the back that's the that's the
back the dgx MV link spine 130 terabytes
per
second goes through the back of that
chassis that is more than the aggregate
bandwidth of the internet we could
basically send everything to everybody
within a second more bandwidth than the
entire internet don't worry I'll give
you a second to let your brain stop
melting all right the 18 compute trays
and the nine Envy link switch trays come
together to form this gb200 nvl 72
system which is just one Rack in an AI
data center this one rack has 1.4 xoff
flops of AI training performance in it
just to be clear there are only a few
exaflop supercomputers on Earth today
and Nvidia just built one in a single
120 KW rack that they sell off the shelf
and at the top of each of these single
rack supercomputers exists a single
infin band switch tray to connect them
to even more single rack supercomputers
and the resulting system can be
connected together to make one 32,000
GPU 645 exaflop AI Factory for the new
Industrial Revolution one Computing
platform for generative AI one giant GPU
and while your brain finishes melting
again let me ask you a simple question
why would Nvidia build something like
this and does Humanity really need gpus
this powerful 600,000 Parts somebody
used to say you know you guys make gpus
and we do but this is what a GPU looks
like to me when somebody says GPU I see
this now let's see what it looks like in
operation if you were to train a GPT
model 1.8 trillion parameter model it
took about 3 to 5 months or so uh with
25,000 amp uh if we were to do it with
hopper it would probably take something
like 8,000 gpus and it would consume 15
megawatts 8,000 gpus and 15 megawatts it
would take 90 days about three months if
you were to use Blackwell to do this it
would only take 2,000
gpus 2000 gpus same 90 days but this is
the amazing part only 4 megawatt of
power power so from 15 yeah that's
right and that's and that's our goal our
goal is to continuously drive down the
cost and the energy they're directly
proportional to each other cost and
energy associated with the Computing so
that we can continue to expand and scale
up the computation that we have to do to
train the Next Generation models so you
can do the same amount of work with one
quarter of the gpus in one quarter of
the data center footprint using one quar
of the power for Big Data Centers cloud
service providers and future
supercomputers Blackwell lets them do a
lot more in the same amount of space and
power and on the other side startups and
smaller companies can now buy a lot more
compute for the same amount of money and
power budget that's why Blackwell is
such a big deal all right I know this
video is a little long but I want it to
be thorough since Blackwell is such an
important piece of technology to wrap
your head around including its double
die design and what that means for unit
costs and margins as well as how
nvidia's networking Solutions can scale
Blackwell data centers to do anything
Humanity can throw at it now with all of
that context let's talk about whether I
would buy Nvidia stock at today's prices
and if you feel I've earned it consider
hitting the like button and subscribing
to the channel that really helps me out
and it lets me know to put out more deep
Dives like this thanks and with that out
of the way let's talk about Nvidia stock
I think that Blackwell is about to move
the needle in a big way for NVIDIA black
Blackwell is not just a new data center
GPU it's an entire compute platform
chips based on the Blackwell
architecture will also go into
self-driving Cars Smart factories and
even humanoid robots all of which I'll
cover over the next few videos because I
really believe that Nvidia stock will
sit at the top of the indexes and
Pension funds and everyone's portfolios
for years to come so with everything
I've learned from Nvidia GTC I'm moving
Nvidia up one spot on my list of stocks
to get rich with without getting lucky
that's a pretty big move considering how
high it was already and the fact that I
just can't afford to buy as much at $900
per share I also ended up moving Amazon
stock up one spot since AWS is the
biggest cloud service provider on the
planet and I think that their Bedrock
platform for generative AI is about to
benefit a lot from nvidia's Blackwell
gpus this is why it's so important to
understand the science behind the stocks
and if you want to see why I pick these
stocks here's the video where I walk
through it step by step either way
thanks for watching and until next time
this is ticker symol U my name is Alex
reminding you that the best investment
you can make is in
you
Browse More Related Video
NOT TOO LATE! I'm Buying Nvidia Stock (NVDA) After Earnings!
Could Intel Stock become the next NVIDIA?! π
Is Nvidia (Still) Overvalued?
NVIDIA'S HUGE AI Chip Breakthroughs Change Everything (Supercut)
NVidia is launching a NEW type of Accelerator... and it could end AMD and Intel
Which nVidia GPU is BEST for Local Generative AI and LLMs in 2024?
5.0 / 5 (0 votes)