Nvidia Reveals Grace Hopper Generative AI Chip (Computex 2023)
Vložit
- čas přidán 29. 05. 2023
- At Computex 2023, Nvidia CEO Jensen Huang shows off its generative AI chip, Grace Hopper. See what the CEO has to say about the future of accelerated computing and AI.
Never miss a deal again! See CNET’s browser extension 👉 bit.ly/39Ub3bv
Subscribe to our channel: / @cnethighlights - Věda a technologie
The way he was holding it and not afraid to drop it, just as brave as Linus.
he started the video by saying it's in full production. if he breaks that one it's no big deal. why would he be afraid to drop it?
@@demonz9065Well he's giving a presentation on how innovative the technology is. Dropping it on the ground in front of all those people would not be optimal.
@@oldpain7625 it would'nt be optimal but when is dropping something delicate optimal? his embarrassment would be the biggest downside to him dropping something that's just gone into full scale production. there's no real reason for him to be afraid of damaging it
Since the February 02/2023/untill now I'm trying to rebuild websiaite lol the Microsoft 365 lunch onfenrary 3
He got the power of the leather jacket
I do think Nvidia flies under the radar a little bit with the way it makes all of the advances of the other big computing companies possible. This seems like a seriously huge leap in computing power. It may prove fundamental to unlocking a real AGI.
There are other players doing pretty crazy stuff for less general purpose but extremely powerful AI training. Cerebras has their wafer scale AI processors which are very good for specific AI training models. Tesla with their Dojo computer is pretty crazy for training, it's memory bandwidth and hierarchy is suited for very high bandwidth and capacity fully addressable. It's far more powerful than their Nvidia GPU clusters. Nvidia is very general purpose, their GPUs have capabilities beyond just training instructions like Bloat 16 and low precision integer which limits the architecture, the difference between vector and matrix is decently significant. A pure matrix architecture with assistance from a simple integrated CPU in each cluster like dojo or Cerebras is quite powerful vs a GPU that must do vector and matrix.
I would also say that AGI will not be done by a large language model, it doesn't have any true intelligence, but it's very good at making connections.
@@Jaker788 but isn’t making connections the fundamental way our brains learn things as well? Neurons for connections in the brain.
@@OneDivineShot LLMs just aren't it, they're really good for certain things, but they have their limitations. Open AI has even said that GPT4 is about where the ceiling for LLMs is. We will need a different kind of model for AGI.
@@Jaker788 The crazy thing is not THAT it works, but how EASY it is to use. A pseudo-AGI architecture is more than sufficient to emulate the gamut of human proclivities. NVIDIA just handed it a body, so it can go hide anywhere.
Lol at flies under the radar. Have you seen the 300% increase in the stock recently
The crowd was soooo dead, the man just announced the worlds fastest super computer
No, it’s not a supercomputer
@@thetshadow999animates9
👉🏽 11:37
@@bengsynthmusic it’d be a stretch to call it a supercomputer, more like a marketing thing. Also, this was unscripted so you can’t really go off of what Jensen says either way.
@@thetshadow999animates9
You can see on the left side that it says AI supercomputer. 11:37 So it's not some off-script slip-up. Plus it does 1 exaflops, which would put it among the top supercomputers. It is no doubt a supercomputer.
@@thetshadow999animates9 This is the kind of chip you would find inside a supercomputer, this thing is designed for server computing racks.
If that isn't the heart of Skynet, I don't know what it is.
it is
I'm looking at that 40k pound beast of a GPU and I'm thinking .... It will be the size of a cellphone in 20 years.
Next you're gonna say, AWS will be the size of an airpod in 20 years. No, it won't.
@@user-hx5qv4kd6 doesn't mean we'll see the same change in the next 20 years. We can't shrink atoms.
@@user-hx5qv4kd6 yeah but it’s not the same. You can’t make electrons any smaller. We’re already nearing the limits when it comes to how small our transistors are. Further compaction will require some new breakthrough that we have no real concept of.
People always say the same thing, that we are already at the limit. And then a new discovery is made, and we surpass ourselves. Thinking we won’t do better isn’t a prediction I would put money on.
@@ferdinand.keller I mean I really can't see a way to make transistors smaller than a few atoms.. so he has a point. Maybe we'll achieve some great workarounds, like 3D chips, but that's it.
"I wonder if this can play Crysis"
lmao Jensen is a based CEO
cringe
I don't know why all the gamers act so mad at Nvidia
@@freemanrader75 becasue they are not really here for the gamers?
They are just not trying to lose market share and try to suck every dollar out of you, which apparently works very well.
But the big plays are happening behind Enterprise doors, that's where the real tech is.
@@knightnxk2906 Nvidia is owned by gamers.
@@freemanrader75 because they want to charge 2-3 times as much today for 20-30.. maybe 40% more performance than they did a few years ago.. MSRP not Covid Inflated prices. If were talking the 4060 then you might even get less performance than the last generation..
This guy is really a one in a million passionate CEO!
cos hes getting paid well with bonuses.
@@knightnxk2906 no sir, the guy found the company in a Denny's
@@andy68686 Sure and I am Pope.
@@knightnxk2906 Jensen is worth $33B, this guy can do everything he wants.
Money is no limitation. Do you think he wakes up every day with the idea of earning more money today?
He's 10x better than Steve Jobs. Has actually written code and cares more than anyone else about the mission.
Finally Nvidia created a GPU for LLM. LLM is memory hungry.
I thought all it’s GPUs could handle LLMs, no? Is this one chip that is tuned specifically for LLMs?
@@cidpraderas8950 They did, but LLM needs a lot of memory. So The current LLM running on current GPU has to be sharded across multiple GPUs. If the GPU has more memory, the LLM can be stored on a single GPU and the training process may be much faster due to the reduced amount of inter-GPU communications.
@@cidpraderas8950 depends on the size of the model and the optimization used. newer models usually use more though especially if they have a lot of parameters.
How long until a phone runs on an equivalent chip? 5 years? 10 years?
One thing is for sure is that he knows absolutely everything about the products NVIDIA is making. Can’t say that about all CEOs.
I like how we always go back to rooms full of computers. Despite the processing breakthroughs
More is never enough.
History is a circle they say.
More is more lolllll
10:46 At least he remembers his previous primary comsumers.
*Jensen Huang @**12:00**:* *_"DGX2000, It is one giant GPU."_*
*Voice of Morgan Freeman:* *_"Then... on Dec 12, 2023... Skynet was born"_*
crazy
I'm excited and scared at the same time. What was this GPU named again, Skynet?
I was thinking the same thing.
Yes, now LLMs can be mobile and everyone is trying to make their own AGI.. lol INCLUDING MEEE!!!
@@preddyshite6342whats an LLM?
@@dwightk.schruteiii8454 Large Language Model. That's what chatGPT is. Because it is trained on a lot of words.
Because Attention + Hype🎉
The energy of this guy is supreme!!
Jensen is stacking up hardwares like vectors, matix and tensors.
Grass Hopper
I guess mega caps don't care about price, but the cost must be astronomical
Because it is. At least 100k
This is an insane GPU, if i can even call it a GPU cos it feels like an understatement at this point.
@@knightnxk2906 LOL - you just MASSIVELY underestimated the cost, a single H100 is $40,000+ ;-)
Wow the dedication all just so we can finally play Crises the way it should be! In 40 years we’ll have one in our phones lol
In 40 years, you won't have a phone
New Internet moment: the data center is the computer. 👍
I want a big A100 cluster so I can play in my sandbox, I still don't understand the advantages of Hopper. Looks like you can link them with pods?
hmm but can it really play crysis?
The future is going to be wilder than anyone can predict
This is so insane. I love this.
Remember when people said we'd have to learn Chinese in the future?? Now you wont have to because of A.I the Black swan event.
What?
@@thetshadow999animates9 I get his point
"I wonder if this can play Crysis" hahaha a true classic
Game changer, historical inflection point comes to mind 😅
Amazing! with 1 GPU!!
What happened to the qbit quantum computers?
Will it fit in a case ...and run MSFS2000 ?
Easy: build a flat Rumba the diameter of the average stride, then put the robot on top. I'm thinking a pole with arms and a face/360 camera🫡
When AI from this thing becomes aware and takes over the world, I hope that first thing that is going to do is reduce prices od gpus for 50% so it could replicate more efficiently... 750 bucks 4090 for all
😂
When that happens that will be the least of your problems...
Even 750 bucks sounds insane for a GPU tbh... How times have changed
Unfortunatly modern machine learning AIs are actually just a glorified collage machine, so self-awareness is much like a medieval ornithopter became a modern stealth jet fighter
if that happens it will give you gpus for free, also neuralinks. In fact it will force you to take it and put you inside a tube "for your own safety" to rot in the matrix forever never remembering what reality looks like.
Wow. The next shield TV is impressive.
10:44 "I wonder if this can play crysis" lmao
Whats the MTTF of that setup?
Crazy scientist! You can't help but love this guy and his company!
Are you out of your mind ?
This man pushes greedy practices in the whole GPU industry, last gen consumer GPU are literal scams.
arent you an easy sellout.
Waht is the algorithm of your chip?
he sounds super excited.
Interesting. Dropping a comment here as an ' I TOLD YOU SO' when this piece of tech goes either south or north. Definitely a game changer. New breakthroughs coming in. A new world!?
Don't worry it will take at least 3x the VRAM for it to do anything useful.
This just makes the 3070's 8GB VRAM limitation more painful ...
that jacket is just a tradition now
AI seem to be the feature. at the same time i having a thought about skynet in terminator. computer power like this is mind blown. i still think phone computer power is powerful for its size. i still wait the days were they have contact lens small enough to have a power of a phone.
Still should have asked ibm for their optical circuits technology.
So the card is 21 times more powerful than the Geforce RTX 4090. Great.
great stuff
Looks just like the terminator AI processor 😂.
ENERGY SAVINGS? 2112 fans need less energy than millions of them
That’s a 20 Ton GPU with 144TB ram if you didn’t catch the drift.
A future version of this will be what we have to suicide attack to destroy it to stop the machines
Considering how much the products cost to make, how do you justify the MSRP?
200 billion transistors. Let that sink in for a moment...
This founder/CEO's company has market cap bigger than TSMC?
how much memory??
it's better to make super home computers based on this graphics with the ability to connect to the network via optical fiber something like bitcoin miners but working differently, buying all this costs why to buy when you can pay for use only and not for all equipment and so after a year there will be something else better
Anybody get a count on how many times he says “Grace Hopper”?
Thanks to Nvidia and his CEO, Skynet was waiting for this technology to be born !!!
a "goodbye, gamers" keynote
Such a likeable dude.
We're about to start summoning the machine gods. It's gonna sting but oh well.. here we goooooo
Jensen, if I was in the audience I would jump on stage just to make sure you are a real person or rendered by Grace Hopper?!😁
Taiwanese energy mixed with American excitement is this kind of a ceo
Me : looking at superchip on a $300 phone
Me: Reading this comment on $50 tracfone I cant unlock in my country
10:44 😄😄
Insane 👏
damn this leather jacket man is killing it
Nvidia for life 🧬 for love ❣️ for future 🤗 keep going for the great work 🤠 be happy be safe
Why does GH200 sound like T800 getting hosted on CNET
It awesome
Can't even wrap my head around the ramifications of this!
Dude we have to leave now, I just 4:12
😂
astonishing.
This is cool
This is cool and stuff im most interested in the stuff they don't tell us about...
This guy is. Skynet?
John, we found him at last
I am honestly curious if it can run crysis
can we see gpu price drops ?
When did Jackie Chan go in to tech ?
it's not enough for artificial intelligence, nvidia needs to work on apatite crystals
During the Computex 2023 event, Nvidia unveiled its highly anticipated and groundbreaking Grace Hopper Generative AI Chip. This cutting-edge technology represents a major leap forward in the field of artificial intelligence, harnessing the power of advanced algorithms and machine learning to drive innovation and transform industries. With the Grace Hopper chip, Nvidia continues to push the boundaries of what is possible in AI computing, paving the way for exciting new applications and advancements in various sectors. This remarkable achievement showcases Nvidia's commitment to shaping the future of AI and solidifies their position as a leading player in the industry. The Grace Hopper Generative AI Chip is poised to revolutionize the way we approach complex tasks and unlock new possibilities in the world of AI-driven solutions.
❤
@@Faith-kw4rl 😊
No crowd understood the crysis joke!!????
if u got a hag dentist or an imp or watever, on your back, this grace hopper chip will help you out
All we as consumers care about is what this means for GPU’s?
But is it heavy?
Can it run crisis though?
For those of u regular people like me let me explain it for u.
Computers go fast
Hopper makes computers do 100x faster
Get ready for the internet of things life’s about to get jetson like
Designed For 32K gaming Performance at 120 fps
This ceo is cool
Still don't know why they haven't developed a game engine
So thats where all of the GPU vram went to, no wonder they have to cut all of the lower tier vram
Gamers can say goodbye to normal gpu prices. Margins for these are stupidly high.
Should really talk about efficiency. This thing needs humongous power and carbon emission.
10:43 HAHAHAHAHAHA
And so it begins...
Whatever happened to nobody will ever need more than 640 kilobytes of RAM
Technology progressed
Oh Billy the Gates. How wrong he was.
Needs fewer fans as they will be a maintenance headache.
Fk***ng amazing!
The people should have been cheering the whole time.....wtf. This is awesome news people!
They r bunch of !deot, society are, they usually rush to describe scientific advancement (especially the biological & technological) with word like "Creepy "Scary "Unethical" etc.
The tech is amazing, but the presentation really wasn't.
@@sebastianjost It didn't seem very well put together but sometimes that's more organic and wonderful 😄
@@Ricolaaaaaaaaaaaaaaaaa because they know how this tech impacts the future.... its good tech dont get me wrong but anyone with a functioning brain knows we are not responsible for this to be in mass production yet
the presentation is E3's level of cringe im surprised Crowbcat havent woken up on his slumber for this lmao... they really need to get someone else to present this lol
If Nvida keeps on revolutionize their tech, they're going to be unstoppable$$$$$$$$$
Price ?
you don't want to know..
4 elephants 1 GPU 👁👄👁
reminds of 2 girls 1 cup
Hopefully someone will reverse engineer it the same way Sega and Namco reverse engineered Military hardware so it's cheaper.
You can’t reverse engineer something this advanced, you’d need machines that cost over $100,000,000 a piece plus maintenance just to manufacture the silicon it runs off of.
@@thetshadow999animates9 Sega had lots of money back in the day
@@binnieb20 while Sega was indeed worth what is equivalent to today’s companies which are worth hundreds of billions of dollars, Sega was playing on easy mode with how much simpler technology was at that time. The only technologies you could steal would either be patented or be useless or even already done by Nvidia’s competitors. An example of this is AMD copying Nvidia’s DLSS 1, 2, and at the moment trying to copy DLSS 3.
@@thetshadow999animates9 we’ll see
@@binnieb20 I think we won’t see