Tesla's Hidden Compute Power: Bigger Than Expected! w/ Brian Wang
HTML-код
- Опубликовано: 15 май 2024
- Big things are changing for Tesla and they are changing fast.
Let's cover the growing compute power that Tesla and Xai already have and how much they plan to have.
Will Tesla have the advantage for the massive data centers that will require massive energy?
Brian Wang is a futurist Thought Leader and a popular Science blogger with 1 million readers per month. His blog Nextbigfuture.com is ranked #1 Science News Blog. It covers many disruptive technologies and trends
Level 4 self driving is happening before our eyes.
• Tesla is rapidly creat...
Tesla Robotaxi Unveil on Aug 8 and the FSD Ramp. What is known and unknown?
• Tesla Robotaxi Unveil ...
AI Takeover: Are We Ready for Computers to Surpass Human Intelligence? AGI Timeline • AI Takeover: Are We Re... Follow Brian on nextbigfuture.substack.com
Follow Brian on / nextbigfuture
Follow Brian Wang on NextBigFuture.com Follow Brian Wang on X.com/@nextbigfuture
My website: www.TeslaInvestor.com
Get Free TESLA Milestone Tables
Check out 15+ modules of resources for the $TSLA Investor
Join this channel or Patreon to get access to perks:
Get free access to 15+ modules of TSLA investor resources
Become a Member: / @brighterwithherbert
Become a Patreon: / brighterwithherbert
Use my referral link to buy a Tesla product:
ts.la/herbert23392
❤️ Thank you to everyone who supports me and this channel! I really do appreciate your likes, subscribes and comments.
Let's get brighter!
Website: www.herbertong.com
X: @herbertong
Disclaimer: All content on this channel should not be construed as professional financial advice or recommendation to buy or sell any securities.
Author is invested in multiple Humanoid companies and is long TSLA and other stocks at time of original video publish date.
#Tesla #TSLA #stock #ElonMusk #investing #teslabot #technology #ai #TeslaNews #TeslaFSD #teslastockanalysis #TeslaUpdate #ElonMuskNews #EV #FSD 
Наука
This is one of the best conversation i have heard in a while. Fantastic work Herbert !
The combination of sustainable energy + compute as a mission for Tesla would make perfect sense.
The most interesting thing I have seen on youtube for a very long time - you have my subscription - many thanks.
Yes Tesla Energy as I said yesterday. Is going to be 🎉🎉🎉Thanks for this extremely informative video Herbert and Brian❤
Brian's description of the economics of how a gpt cost per query difference of 1 penny making the difference between success and failure, just hit me.
Tesla is in a position to soon have +1MM people paying it $100/month AND these customers pay for all of the needed power and edge compute for inference, themselves.
It just struck me this is a huge advantage over other AI companies.
+1MM morons feeding into Elmo's lies.
And you just alerted me. 👍
100$/month? Where did you get that price?
@@dominicstlaurent I mentally always round 99 up to 100. It bugs me that everything is priced with trailing 9s because it makes the price look smaller than it is.
FSD is $99/mo not $100/month.
Tesla needs to release a Kitt voice and xAI inference OTA upgrade when they fully release FSD.
It's a slump with no sugar coating it. I hope we come out of this rough time ASAP.
This is one of the most informative videos i ve seen recently. You guys are awesome. Thanks!
Hopefully at some point the AI will become smart enough to reveal a smarter and more energy efficient way to move forward rather than just continue with brute force.
Exactly, thought the same and commented similarly before reading your comment.
I was just thinking the same thing that’s crazy lol. Maybe ai can come up with better materials for more efficient solar panels
great show!
17:25 - There is the column called „Energy“ but it contains values for power - thatˋs confusing 🤔
Marty we need 1.21 gigawatts to produce AGI.
Cheers guys
Good news 📰 very interesting and informative
Thank you again for another amazing episode! Wonder if you could talk a bit about Groq, an inference chip way ahead of Nvidia. And if maybe Telsa could partner with them in anyway if needed. Nvdia vs Dojo vs Groq. Each may be differentiated by their training capabilities, inference, power usage, use case, cost, software, pro and cons
No mention of the power needed for cooling. As I understand its roughly 1:1 cpu consumption:cooling consumption. The close placement to an abundant source of coolant AND power could be very advantageous, e.g. a Dam. If there is local population, hot water could also be provided as a by-product.
sounds amazing ... what kind of safe guards will you need for all this ... physical target all in one place, fire, power outages, hackers ?
Any safeguard for nuclear plant?
This is the best Podcast ever 🥇. But I need to repeat some section to grab it in better the detail what Wang said. It is mind-blowing info. It explains better why nVidia is/was? going to the moon. And probability for Tesla Energy growth. But Man!!! the money involved to finance this, I can't grab how all those companies will do this. The energy delivery in the short term is probably the limiting factor, beside money. Herbert, greaaaat job !!!
Good show. Very informative.
Tesla should already be in the nuclear energy business comparable to Last Energy's SMR business. The compute & data centers are going to be energy limited. Nuclear has THREE MILLION times the energy density of batteries, solar, wind, fossil fuels. The NRC needs to be abolished, and have the safety mandate discontinued and a commercialization mandate added, if NRC is going to exist.
Would love a review of GSI’s Gemini II APU. It’s now in production and requires way less energy to operate related to GPUs.
The need for energy will no doubt slow down as chips get way more efficient and the compute more streamlined. Perilous journey for the players for sure, easy to blow a ton of money on obsolescence.
Exiting times for sure!
Let's hope Optimus can build the solar power supply, the thought of China doing it on Coal is a nightmare
Inconceivable!
All Hail Jeff!
This is, the best program
I'm not sure the starlink network is currently good backhaul for distributed training. I would think that a single shipping container with enough juice inside to need a megapack would need more than 20Mbit upload.
Energy requirements at these future scales suggests that space solar power may be necessary to supply the demand. in fact, rather than beaming the energy down to earth from orbit, maybe put future compute centers up in orbit with the solar arrays.
Elon has calculated how much energy the usa needs and put solar panels over 100 x 100 miles would be mire than enough.
the answer is usually you have your source of energy locally as well as your compute for best latency. Solar power on the ground.
putting compute into space probably would have heat problems as there's no easy way to remove the heat, vacuum is not a good conductor. This will be science-fiction which is fun. They can put data centers on the moon but then there's latency. Might have to invent a new way to transfer heat through some electromagnetic beam which might look kinda like a heat ray... (insert sci-fi sounds)
@@geniegogo True, vacuum is not a good conductor, but all that vacuum is at about 5 degrees Kelvin (or something like that), makes for a terrific radiative heat sink. In space, like in a geosynchronous orbit, the incident solar radiation is for all intents and purposes constant (makes for a good base load supply) and considerably stronger than here on the surface of the Earth. One thing that hasn't been talked about much so far is the possibility of using the Moon as a source of materials for solar arrays orbiting Earth. Much of what is needed can be found on the Moon, and the energy requirements for Moon-to-geosynchronous Earth orbit is way less than that for Earth surface-to-geosynchronous earth orbit, A much easier trip. A lot of the hassle and expense associated with doing stuff in space revolves around keeping people alive and healthy up there, but with the rapidly growing capabilities of robots, a lot of fabrication and assembly could potentially be done with them instead of people.
I asked ChatGPT4 how much more solar energy was available per unit area on the moon vs on the surface of the earth - there is not that much difference - on the earth about 1kw/m2 - on the moon about 1.366kw/m2, on Mars (further from the Sun) 0.59 kw/m2 ... I'm not sure it is worth the effort of setting up all your panels on the moon for only 40% more energy - until such a time as you have a load of self-replicating Optimi up there to do it for you - but you still have to get the energy (prob in form of a laser) back through the atmosphere to ground to use it, there will be losses involved in doing this...
The Tesla team are funking great
@herbert. How about just send data center modules into space with solar attached and connect them to Starlink. Like mini planets. It’s cold up there and lots of sun.
Need higher orbits (not LEO) to ensure 24 hr sunlight. (Or something like polar, but precessional??) Or batteries cycling every orbit.
Not that much Sun, actually, without much, much, much bigger solar panels than what anybody has ever done before. And then there is cooling. Cooling is a huge problem in space because there is no atmosphere you can dump the heat into and if you want to do evaporative cooling, you need to bring all the cooling liquid up with you.
It is a really, really dumb idea.
Evening guys
19:00 Not sure why I'm the only one out here talking about X(Twitter) being the best LLM data input for Grok. I believe it's the real reason Elon bought it. The data!
The real reason he bought it was because he was forced to, after signing a contract and then trying to back out after realizing he was paying twice as much as it was worth. Quite the genius.
You're not the only one. I was saying the same thing from the beginning.
@@jerrykohut278 was Twitter too expensive - yes; will it be worth a multiple, after Musk realized his plans with it - yes. Mark my words and never bet against Elon Musk....
Brian, will tesla make a tesla bot skill store and Optimus training kits? I think they can rent out the training services for skill developers!
Yes that's the future much like apple has their eco system the same will be said of the bots
Not certain yet, but it appears that the only place where you have unlimited and uninterrupted power along with ultra low temperature for cooling is in space. Therefore, a modular data center in space could be launched and provisioned within a day, covering several geographic regions. Here Tesla and SpaceX have a competitive advantage. The only problem I see is ensuring that DOJO and Nvidia chips are “space ready” given the risk of solar events.
I think there is Grok and Groq which are two different companies. But they know each other 😉
Doesn't Iceland have, essentially unlimited, geothermal power? Rather than try to knit together data centers in different states, couldn't you build a single data center in Iceland?
Except you still have to get your initial data to Iceland, which might not be too bad if your input data is relatively static. And Iceland has some pretty big undersea data pipes, I think..
Iceland is also geo-thermally very unstable with 130 volcanoes including stratovolcano's, the classic cone-shaped peak with explosive eruptions & shield volcanoes with low-profile, wide-spreading lava flows - essentially fissures that open up over significant distances with lava gushing out. Maybe not a great place for massive data centres.
@@gdok6088 yeah looks like only about 200MW of capacity at the 3 largest data centers in Iceland. Not as large as is thought. At least not AI size.
would future exponential energy consumption and computing giga data center affect the heating of the planet?
No exponential is forever.
@@peterfireflylundSo then it has potential to overheat the sun.
What about MI300x? Elon said Tesla is buying AMD chip, don't know.if they're referring to MI300x or not.
Tesla is a huge tech company!
That was a great show with great explanations
ABSOLUTELY OUTSTANDING EPISODE!!!
how many acres of solar panels and how many megapacks in say Arizona does it take to equal the power generation capacity of a nuclear power plant?
Same or less. And the fuel is free. For next 6b years. Costs 3 times less and built in 2 years not 18. And waste doesn't last for 100,000 years.
Oh, and you don't have to buy enriched U from Putin.
Apart from these minor differences 😂
It's more about end of life of a solar farm vs a nuclear power plant. You can recycle and bury a solar farm. But you spend BILLIONS of dollars to deal with Nuclear power plant that is no longer producing power.
@@HughButler35
1. You need a lot more land for solar. A lot lot more.
2. In about 100 thousand years Earth will be uninhabitable, as the Sun is getting hotter.
3. Nuclear could be very cheap without all that unnecessary red tape and ancient tech. Nuclear power today is like rockets before SpaceX, except here you are not allowed to innovate. Drastically better reactors were developed already in the 60's.
4. Nuclear waste is the only one that decays, all other waste lives forever. And fast breeder reactors could reduce that time to around 300 years, while also making 100x more energy from the same fuel.
5. Uranium is super abundant, you don't have to buy from Putin. And there's Thorium too, which is even more abundant.
That said, solar has is the easier short term solution.
About 15,000 acres and 15 million PV panels. Roughly.
Say circle 27 miles radius.
Plus at least 16-25 GWh of batteries (~$2 bill), and some form of backup.
Better be on good terms with Xi. ;-)
See SunCable (Australia), it’s almost identical. 2x as big.
@@iandavies4853
I'm pretty sure those are way too big numbers. Don't have time to do the math now, but did years ago and remember smaller numbers.
Tesla should buy a 4th gen Nuclear Power startup like Copenhagen Atomics.
Seem perfect for these Data Warehouses
I just started watching and stopped at the first table - Grok-7 through 10 will increase compute 5x each year while only doubling the power consumption. WOW. Though 3 GW is insane. No wonder why Elon is saying more electricity is going to needed and so valuable. 3 GW is the total production of a good size power plant. The largest power plant in North America is just under 7 GW and it’s hydro, can’t build those everywhere.
Sounds like a great opportunity to locate datacenters in the west adjacent to the best geothermal sites.
Robotaxi.
Any chance inference chip will be massive, much bigger than HW4?
This’d allow much improved performance with limited (?) modelling / training GPUs.
Leave customer cars with second rate FSD.
How about PHOTONIC NEUROMORPHIC QUANTUM LIGHT COMPUTING that used 10x less power ! ?
Anastasi?
How’s about a deep dive into Tesla stock price going down hill unstoppable
Re: Tesla's Processing Lead. Despite mentioning his major clients of the Nvidia Blackwell B200 Chip at its GTC unveiling, which he touted as 4X Faster than the H100, could Jensen Huang have deferred in mentioning Tesla, at the time, due to an NDA to mute speculation on the latter's plans?
Also, recall Iceland's not only green energy, but virtually unlimited refrigeration, where most of the energy loss exists. To what extents will Groq chips moderate such energy requirements vs Blackwell B200s?
Use the heat generated to heat homes.
MUSK-OLOGY 101
Elon class for World
LLMs are trained to write text. Great for research or generative output where you don't care about the exact output. They can write a few paragraphs about how to do your taxes but they can't actually reliably complete complex the task such that you'd pay good money for it. They just convince you that they can through english prose.
FSD is directly trained on data to the end job of driving like a human given visual input and is easily worth $100/month.
The more breakthroughs Tesla and Elon does the more attacks come they’re shaking their boots out there😂😂
Electricity is the AI cost driver. Solar, wind, battery is the cheapest add. Nuke is most expensive. Heat from these centers needs to be captured and sold.
Nuclear is very cheap long term. If done right. And with newer tech from the 60's it could cost like 100x less. The only reason we are not full nuclear today is politics.
Re: Tesla's Processing Lead. Despite mentioning his major clients of the Nvidia Blackwell B200 Chip at its GTC unveiling, which he touted as 4X Faster than the H100, could Jensen Huang have deferred in mentioning Tesla, at the time, due to an NDA to mute speculation on the latter's plans?
Also, recall Iceland's not only green energy, but virtually unlimited refrigeration, where most of the energy loss exists. To what extents will Groq chips moderate such energy requirements vs Blackwell B200s?
I've been a fan of Tesla for five years now. It's been a while since I've learned something this profound.
The notion that AI will need lots of chips is nothing new but powering it? That's a new level for me.
Google, ChatGPT and other AI companies will be hamstrung by the power requirements.
Only Tesla will be able to power such a system *and* feed it the data it needs. All paths lead to Tesla.😉
LMAO, GQP domestic terrorists thinking Tesla is a leader in anything.
Compute is the future currency.
Put AI data centers in space with its solar panels.
It should be built and maintained by Optimus.
Solar panels in space are more effective and has no interruption of the power from bad wether. Except for solar wether.
? Will All Humans need a Tesla NEURA LINK Chip in Brain just to keep up with AI ?
It is mind boggling how efficient the human brain is. In my opinion there is no way around finding ways to make compute more energy efficient, maybe ai can help us with that once it passed human IQ......
Right now I'm a few cents above water but anyhow I don't plan on selling if anything buy the dip.
I guess I'm a bull.
Data centers need to sell their heat.
Lost in my other comment.
How? It’d be so low temperature as to be useless. 40°C?
Norway is storing district heating at 150°C.
Ok, so the AI compute has to become more power efficient!
Seems that dojo has been lapped by a long way by Nvidia
Will we turn the world into Cybertron?
How many people actually watching right now have solved Rubik’s cube…..?….we know it can be done but this is what it takes😮😅
Man, I hope Tesla energy gets into nuclear in a massive way very soon.
Tesla must begin making its own chips instead of being a chip customer.
Maybe it's time to buy a used aircraft carrier that has a nuclear power onboard. Use this to power a data center???
Those are like 2x100MW, too small. Technically the solution is trivial, we could mass produce small modular reactors. Like how SpaceX mass produces rockets at a fraction of the cost of legacy ones. But with all that political baggage, it's not going to happen. Currently solar is faster to deploy and easier to approve, by a lot.
With Starship space solar could be economical too.
!!!
What will Elon do with the heat all this energy consumption will produce? He for sure will no waste it in the current methods of cooling! Tesla cars don’t waste available heat!
FSD EV computer will be the only computer the ower will need.
Starlink will instantly know what every EV FSD vehicle needs.
Selfplug-in V2G EVs parked 23hrs daily and driving building to building and trading electricity with the grid for money and having computing power with the parked vehicle 😳 👌 👏 👍 😀 😊😊😊
Just a big idea for a big thinker.
Remember horse meat was cheap when the Ford model T production line started.
How about hearing from smart people with facts and data ! How about we stop the high school gossiping and have mature debates?
34:45 - orrrr... Offshore containerized data centers powered by wind tubine farms and pulling data from undersea fiber optic, cooled by limitless desalinated water piped through. "Staff" each container with a bot or three for rack maintenance services and these could operate near-autonomously for a few years before returning to port for renewal of the obsolete. All out beyond the reach of the tax man and pesky regulators. @elonmusk should start buying up supermax cargo ships and oil platforms. Just sayin'...
The conversation’s remind me of what tesla and X’s biggest advantage is, Elon makes fast decisions he OK spending as soon as it appears to be needed. Open AI and Microsoft are not doing that.
Nuclear electricity is grid electricity.
Electricity is dirt dirt cheap.
The grid is insanely expensive.
New grid capacity to millions and millions and millions and millions of customers is the economic road block.
The elephant in the room.
Grid is $1million /km and and millions of km.
Rooftop solar PV and EV big batteries means the grid is UNLOADED and nuclear electricity is available for computing.
So forget cars it's going to be all about energy and feeding the AI beast(s)!!
With Neurolink he will attempt to harness human intelligence as well as AI and make it one thing. We will merge with the AI.
They don’t have money to buy Nvidia chips
Maybe the vast amount of power required by these new supercomputers will spur the deployment of Liquid Flouride Thorium Reactors. LFTR's are much safer and cheaper than current LWRs with almost zero radio active waste products. There is such a thing as "good nukes". Only China is currently working on this design.
The problem with LFTR is not technical, but political. It was developed and thoroughly tested in the 60's, there's not much left to develop.
And there are simpler designs with 100% off the shelf parts that could be mass produced like tomorrow, and would be much safer too.
and build it at the moon?
I’m hearing enough from the bulls. Who else wanna hear from the bears so we get opinions from the two sides equally. $TSLA tanking sigh
You can get perma bear stuff thunderf00t. Chemist that loves to hate all that is Musk.
The problem with bears they are too miopic in their understanding
The Fallacy of Both Sides. There is never unanimity on an issue but arguing with a persons hate is not a winning argument and comity won’t be reached nor is preaching to the choir. While I am up for contrarian views to mine, to be honest they have not swayed me.
I agree
Go listen to Gordon Johnson
How long before AI can design the new Tesla?
AI maybe can't do that by itself,... yet. But it will be an important engineering too,l long before that.
So, not only Optimus solving the last few steps in automating manufacturing. But the nexxt to next Tesla might be designed by an AI who simultaneously teaches Optimus its jobs to manufacture it. The entire factory could be run virtually, before being built.
Very quick iteratiinns!
The future is near!