No Priors Ep. 89 | With NVIDIA CEO Jensen Huang

[@rolodexter]

1. "Hyper Moore's Law" Prediction (1:48-2:28) "over the next 10 years... we could double or triple performance every year at scale" - BOMBSHELL: Jensen predicts AI compute growth will drastically outpace Moore's Law, potentially growing 2-3x annually versus 2x every two years 2. Volta Decommissioning (7:32-7:40) "Sam was just telling me that he he had decommissioned Volta just recently" - REVELATION: OpenAI is actively retiring older GPU architectures, suggesting rapid infrastructure turnover in AI industry 3. 5x Performance Improvement (6:50-7:05) "we've improved the performance of Hopper by a factor of five without the algorithm without the layer on top ever changing" - SHOCKING: NVIDIA achieved 5x performance gains in just one year through optimization alone 4. Scale of NVIDIA's Testing (12:11-12:15) "we have five supercomputers in our company today next year we're going to build easily five more" - REVELATION: NVIDIA maintains multiple supercomputers just for internal testing 5. Seven-Year Software Support (14:14-14:29) "Nvidia Shield our Android TV... shipped it seven years ago it is still the number one Android TV... we just updated the software just this last week" - SURPRISING: NVIDIA's long-term commitment to software support, even for older consumer products

[@JustinHalford]

Jensen speaks the future into existence. Absolutely incredible to witness in real time the OOMs shift to data centers with the power draw of Manhattan. The single most important shift underway today - all else pales in comparison.

[@protectanimals9792]

Jensen & Elon I think are the most knowledgeable hands on CEOs in the world. Others sometimes barely know their product

[@mikeskupniewitz8057]

Their…

[@protectanimals9792]

@@mikeskupniewitz8057 Thank you

[@diamlierx]

Yeah I agree, have AI as a tool makes a lot more sense than these morons running around saying it will replace you.

[@etbadaboum]

Then look at China with Robin Zeng, Ren Zhengfei and Wang Chuanfu among others

[@AIForHumansShow]

oooooooh now this is a get -- always love these episodes

[@Emerson1]

Great Interview! - & congrats to all on the leather jacket budget , hopefully its vegan leather

[@Dhejdjskd]

Pathetic vegan leather. Get a life sis.

[@phatcat7924]

He made leather jacket formal😂😂

[@StatMachLearn]

Is this the "Leather Jacket AI Meeting"?

[@life_sized]

Yeah, I don't know who took the leather lead here. AI rock band post-show stuck in an office vibe.

[@mastershredder2002]

Elad could only afford one quarter of a leather jacket. The rest was patched up and frankensteined together with black canvas.

[@galinageorgieva4735]

I am buying one right away, wanna be part of the Leather Nerd Gang, we (nerds) are cool now

[@lekan4word]

Feels like the matrix 😂

[@UnsaltedCashew38]

Jensen has already made fun of interviewers who wear leather jackets to copy him and these guys did it as well. He wore a suit to one interview to throw interviewers off, at another he wore sunglasses. You don't need to dress like the person you're interviewing!! TAKE A HINT

[@SarbjeetJohal]

Great interview. No swirl chairs, or tell Elad not to move in his chair :)

[@tulpjeeen]

It's not that bad

[@TianaMireyaLilliana]

designedbyai AI fixes this (AI Architectural Designs). "Jensen Huang discusses technology."

[@NilsWestgardh]

The GOAT

[@SuperMunQ]

Which Matrix film is this from??

[@JustinHalford]

Matrix: Multiplications

[@SuperMunQ]

@@JustinHalford 🤣

[@Hamzairshad5]

Please 🥺 add subtitles this thing help me and lots of other people to easily understand conversation

[@GreylanderTV]

youtube adds automatically generated subtitles, just click the [cc] button at bottom of video (sometimes there is a delay before they are added so maybe they were not available for you because you watched very soon after the vid was uploaded, but they are there now for me)

[@Hamzairshad5]

​@@GreylanderTVAutomatically generated subtitles sucks

[@rayghoward]

Elon and Jensen are the most important entrepreneurs in the world 🌎 currently.

[@Kylbigel]

Favorite leader in engineering.

[@tk1576]

god damn I wish I could work at NVDA, they actually value their developers and software

[@diamlierx]

I agree, using AI as a tool makes a lot more sense than these morons saying it will replace your job , any automated robotics/software system currently in use is limited and requires supervision. It takes a network of people to maintain and supervise it ,not to mention the huge cost of deploying a new system and the downsides such as learning curves, limitations , and other various reasons that impact production over the system currently in place. Most companies(not all) are looking for the cheapest route and will not invest the money and resources it takes to properly deploy and run an reliable automated system.

[@damianxd720]

You shoul’ve asked him if he thinks jobs are gonna disappear

[@kevn23]

I just wonder if it makes sense for these customers to use the training NVDA hardware for inference if there are ASICs that can do inference 100x or 1000x the speed. But like Jensen said, the whole system has to run optimally to get the speed boost. And I am not sure if ASICs can take full advantage of the NVDA ecosystem. ASICs are also very expensive to design and manufacture each generation.

[@skierpage]

The hyperscalers are developing their own chips as an insurance against Nvidia charging $250,000 for its next GPU. But none of them are "100x or 1,000x the speed" at general-purpose generative AI; instead the goals are cheaper and higher performance/Watt on specific workloads. Facebook found its MTIA 1 inference chip only 3x better perf/W than a GPU at DLRM ("Deep Learning Recommendation Model," aka picking an unending stream of divisive inflammatory content for its victims). The independent chip companies developing inference ASICs seem to be struggling to keep up. Intel shut down Nervana and now hopes for Habana, Cerebra still has few customers and pivots between "fastest for inference!" and "fastest for training!", Graphcore keeps hoping.

[@UnsaltedCashew38]

Why doesn't anyone ask Jensen about the price gouging Nvidia is doing with video cards? Why is the RTX 4090 $2000+?? Just 4 years ago the latest video cards were $400. 5x the price?? C'mon.

[@Allplussomeminus]

2:05 why does his voice sound like it did an edit cut there?

[@urimtefiki226]

He does not know my algorithm, he has not seen the combination of my numbers. He is prevented to see it.

[@DericO-rz3be]

*MetaCortex Black Pleather JacketBots Circa 1999* There is no avocado toast!

[@labsanta]

Nvidia's Role in the AI Revolution and Future Directions Current Context: Nvidia, led by CEO Jensen Huang, plays a pivotal role in the AI revolution, with a market cap of over $3 trillion and significant investments in scaling computing power (0:16-0:29). Evolution in Computing: Transition from CPUs to GPUs revolutionized computing by enabling parallelization across clusters and data centers, allowing for massive scalability (1:07-1:37). Future Goals: Nvidia aims to double or triple computing performance annually at scale, significantly reducing costs and energy usage, and exceeding traditional Moore's Law (1:43-2:10). Technological Innovations: Codesign & Full Stack: Optimizing both the hardware and software architecture to continue scaling computing performance (3:00-3:36). Data Center Scale: The evolution of NVLink to facilitate high computational needs and low latency in data centers (3:43-4:23). Applications and Data Centers: Transitioning from traditional software coding to AI-driven systems has dramatically increased the efficiency and capabilities of Nvidia’s hardware and software ecosystem (5:00-7:05). AI and Science: AI is increasingly applied in fields like chip design, where it significantly augments human capability by exploring larger design spaces (21:00-22:17). Challenges and Opportunities: Building large-scale data centers and superclusters involve significant challenges in logistics and engineering, like the rapid setup of Elon Musk's x.AI supercluster (15:03-18:49). Future Outlook: Nvidia envisions a future where AI will aid significantly in chip design, making it a major focus area due to its complex and impactful nature (30:09-30:36).

[@NicholasWilliams-uk9xu]

That's a great focus, per compute scale (efficiency and performance metrics) measure end point metrics relative to end point scale of application.

[@NicholasWilliams-uk9xu]

Sometimes I'll not understand something properly, then I'll go to A.I to try to get right answer, or to lay out the idea, and it will deliver confirmation bias, then I'll be think I know something for months till that day I figure out that I was convolutedly fooled lol. Blood flow to the head as embarrassment kicks in, the brains way of feeding calories and work to hurry up a fix the problem before I further look as gullible as I actually am.

[@velo1337]

what type of vsi scaling is he talking about at 2:45

[@skierpage]

DuckDuckGo, Wikipedia, and Microsoft Bing-Chat-Copilot-AI-for-Windows-Azure-365 are your friends. "In semiconductor electronics, Dennard scaling, also known as MOSFET scaling, is a scaling law which states roughly that, as transistors get smaller, their power density stays constant, so that the power use stays in proportion with area; both voltage and current scale with length." "Carver Mead's VLSI (Very Large Scale Integration) of integrated circuits scaling law... key idea is that as the physical dimensions of transistors are reduced, several beneficial effects occur: speed improvement, power reduction, and density increase. This scaling law provided a scientific foundation for Moore's Law."

[@velo1337]

@@skierpage ty

[@elianaoliveira2431]

colibri precisa de água doce e frutas da pra fala de um jeito elegante por gentileza obrigada🤭🤭🤭🤭

[@shankarbalakrishnan2360]

U can make chips to help with our senses make it jensen

[@PrimordialEconomics]

The fact I can't find this podcast on X and I have to use a google product to watch it is frankly criminal

[@RK-um9tu]

First, no one cares. Second, don't watch......problem solved.

[@DericO-rz3be]

That is because X sucks!

[@martiananomaly]

*Twitter

[@DericO-rz3be]

20 or so MEGA TECH BROS renting people at their latest whim to build the next scale blah blah blah... what a beautiful future...

[@MesiasNavarro-n4u]

If Ai can not generate clean electricity to run itself .. it is not that intelligent .

[@wikath449]

I always wonder: Why is scientific progress always (and imho: the ONLY) justification in Jensen ´s arguments, for progressing with A(G)I THAT fast and THAT uncontrolled ? What about all the social "casualties" that may happen alongside ? What if this (NVIDIA-Superclusters) is REALLY the fabric, for an intelligence, that eventually supersedes mankind intellectually, and subsequently drives its own targets (i.e. the "singularity-issue") - (this goes along the arguments of Geoffrey Hinton). Is the scientific "progress" (please define it...) than ENOUGH to justify the means... ? I guess: Jensen himself will regret it - and he/we will watch it in the next 10-20 years. How can he then enjoy the wealth he has amassed without regret... ? I would gladly trade-off some "not-yet discovered" scientific progress for my autonomy (as a human being). (i.e.: I prefer my autonomy) For me, the whole AI-frenzy boils down to the question: Life-quality vs life-longevity, and eventually the whole "purpose-discussion" of AI is always about this "insane" desire for: IMMORTALITY - and that is futile anyway...because even our Earth has an expiry date.

[@skierpage]

You raise good points, though it doesn't boil down to your last question. Scientific progress improves many things people want: better health, better video games, a sexy plastic pal to confide in who tells you how great you are, and in general cheaper and better products and services; though it does little to effect social changes like a more just society. It's likely AI will deliver those improvements hence the trillion-dollar gold rush. The hope is we figure out the alignment problem before an artificial superintelligence pursues its unknowable goals without caring about humanity. I'm far more concerned about the very known goals of the sociopathic billionaires developing frontier models: 1) get us hooked on an unending stream of divisive and inflammatory content so they can sell our profiles to advertisers; 2) squash any attempts to limit their power or tax their wealth (and in Elon Musk's case buy his way into government so Tesla doesn't have to face consequences for him lying about the capabilities of Full Self-Driving 5 years ago!).

[@joseph24gt]

you are welcome to try your mortality, as early as you want. me, as most people i guess, will try our immortality as much as possible. ai is our only hope.

[@RK-um9tu]

@@joseph24gt Nut job alert...

[@RK-um9tu]

@@skierpage Of course you aren't concerned with just societies. You are a white Western male who is only concerned with things that limit you. The fact that you can't see the 2 points in your second pargraph create socially just societies proves my point. You are welcome for the free education.

[@goranmajic4943]

From gaming nerd image, to Bitcoin infrastructure builder to AI whatever... He builds GPUs and is not a genius.

[@BenLi-d7n]

GPUs are capable are capable of doing 3D calculations which are quiet complex