Elon Musk spoke about many breakthroughs at a Tesla All Hands meeting yesterday.
Elon said that the Tesla Dojo 2 chip will be in volume production in a few months. The Tesla Dojo 2 chip has ten times the performance of the Dojo 1.
If Tesla Dojo 2 is ten times better than the Dojo 1 training chip then it would be 3.6 petaflops for fp8. This would be several time better than the Nvidia H100 and maybe 40% of the B200. What needs to be known is how does the Dojo 2 compare for power and cost. It could compare with B200 when factoring in cost. The B200 is sold for about 75% margin. If Nvidia sells the B200 for $40,000 but it costs $10,000 to Nvidia. If it costs Tesla $10,000 for Dojo 2 with the power of 40% of a B200, then it would be worth it. $25,000 worth of Dojo 2 would be equal to a $40,000 B200.
Dojo 1 per-chip performance: 0.362 PFLOPS (362 TFLOPS) at BF16/CFP8.
Dojo 2 projected per-chip performance*: 0.362 PFLOPS x 10 = 3.62 PFLOPS.
Nvidia B200 (Blackwell Architecture)
Process: TSMC 4nm (custom).
Compute:
– FP4 (with sparsity): 9,000 TFLOPS (9 PFLOPS), per Nvidia’s GTC 2024 keynote.
– FP8 (with sparsity): ~4,500 TFLOPS (4.5 PFLOPS, estimated based on scaling trends).
Power: ~1,000W TDP.
Use case: Next-gen AI training, targeting extreme performance.
Nvidia H100
Process: TSMC 4nm.
Compute:
– FP16 (with sparsity): 989 TFLOPS (~0.99 PFLOPS).
– FP8 (with sparsity): 1,979 TFLOPS (~1.98 PFLOPS).
Power: ~700W TDP.
Use case: General-purpose AI training, widely adopted across industries.
Tesla Highlights from the Video and Cern Basher Compilation
– Already produced over 7 million vehicles globally.
– will surpass 10 million next year.
– at times there are rocky times, a little bit of stormy weather but the future is bright.
– Tesla’s work related injury rate has declined over time.
– where does AI and robots fit into the sustainability picture?
– it’s about sustainable abundance for all.
– on track to complete the Semi factory – will make millions of the Tesla Semi – will also have the ability to go autonomous down the road.
– the future is autonomous.
– five years from now, autonomous Teslas will be everywhere – regulatory globally.
– almost the entire fleet – which will surpass 10 million vehicles next year – is capable of autonomy.
– an autonomous car could have the usefulness that’s 5x to 10x a regular car.
– with a software update, we could turn 10 million cars into the usefulness of 50 to 100 million overnight.
Megapack can more than double the total power output of a given grid and excellent for stablizing the grid.
– Long-term more than 90% of all power on earth will be solar + batteries.
– Cell manufacturing – we’re making the most efficient cell in the world – lowest cost per kWh.
– built the first Optimus on the production line in Fremont.
– Should make 5,000 to 6000 Optimus Teslabots this year and parts for twice as many. Ten times as many in 2026.
– Cortex 1 – used for AI training. Over 50,000 active GPUs – soon to be 100,000 GPUs – making it top five in the world.
– Have Dojo 1 active now – handling 5% to 10% of the training load. Dojo 2 – will be 10x better than Dojo 1. Optimistic about Dojo – “have a real shot at a breakthrough.”
– have the cars doing useful work for the first time with no one in them – driving from end of line and park themselves in Frement and now in Austin.
– Tesla self-driving will be 10x better than human drivers.
– Optimus is going into production this year – the new 22 DOF hand is now in production.
– “most sophisticated humanoid robot on earth”
– “our robot has a real brain”
– “Tesla is the leader in real-world AI – what we learned from the cars we transferred to the bot.”
– Tesla is the only company that can make intelligent humanoid robots at scale.
– “Optimus will be the biggest product of all time by far – nothing will be even close – I think it will be 10x bigger than any other product ever made.”
– the Cybercab production line looks like a high speed consumer electronics line – moving so fast that people can’t get close to it – cars coming off the line in 5 seconds.
– we need even bigger casting machines – 50,000 tons – “how big can a casting machine be – what are the limits of physics? – let’s find out!”
– hope make about 5,000 Optimus (parts for 10k to 12k) this year.
An employee asked: is a robot going to steal my job? Elon: people will manage a flock of robots. Same will be true for self-driving cars – people will manage a fleet of cars.
– interesting opportunity to make an electric VTOL jet – “maybe at some point we’ll do that.”
– Hyperloop – vacuum tunnels – high speed autonomous pods from city center to city center.
– Master Plan 3 + abundance for all = Master Plan 4
– Optimus production in Fremont + an even bigger production line in Austin. “We’ll be making tens of millions of robots a year – serious volume – maybe one hundred million robots a year.”
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 technology and trends including Space, Robotics, Artificial Intelligence, Medicine, Anti-aging Biotechnology, and Nanotechnology.
Known for identifying cutting edge technologies, he is currently a Co-Founder of a startup and fundraiser for high potential early-stage companies. He is the Head of Research for Allocations for deep technology investments and an Angel Investor at Space Angels.
A frequent speaker at corporations, he has been a TEDx speaker, a Singularity University speaker and guest at numerous interviews for radio and podcasts. He is open to public speaking and advising engagements.
Long-term more than 90% of all power on earth will be solar + batteries. I dont believe that we are trending in that direction. An interesting post would be about making that case.
MSNBC’s Rachel Madow saw that meeting too. Her report on it reminds people of Musk’s over-confident predictions in the past: https://youtu.be/2e_o5irr-ss?si=8U_pKHW8SseUT1pS
Of course, Thunderfoot has almost made his career reminding people of Musk’s failed predictions for a decade in his videos.
I have no doubt autonomous driving won’t be any more real this summer than it was last summer when it was promised then too. It won’t be real until an independent insurance company insures truly hands off unsupervised autonomous driving for a competitive price people can afford.
It’s one thing for a technologist to exaggerate his companies prospects, but when he’s in the government, millions of lives are at stake, maybe even the future of the world with the power Musk has been granted, which is without precedent. And DOGE is flailing even worse than Musk’s companies.
Rachel Maddow saw that meeting too. Her report on it reminds people of Musk’s over-confident predictions in the past:
She’s (unintentionally) hilarious at 5:30 into the video: “Mars isn’t nearly as hot as I thought it would be.”
No, duh, Rachel!
Even *Elton John* knew, 53 years ago: “Mars ain’t the kind of place to raise your kids. In fact, it’s cold as hell…”
“Cell manufacturing – we’re making the most efficient cell in the world – lowest cost per kWh”
This was the big surprise for me. The rest are largely projections of the future etc.
So Tesla batteries are being made cheaper than what CATL or BYD can do sitting in China? Unbelievably fantastic!
BYD claims it can charge its cars to 0-80% in 5 minutes now, for a 400km range, making it almost comparable to a gas-powered car refueling in 5 minutes, though the latter still has a bit of range advantage.
Eventually, America will be forced to drop protectionist blocks against Chinese EVs, which are far cheaper and often better than American EVs. Tesla is already losing to them in any market where they are allowed without tariffs – China (obviously), Australia, much of South America, SE Asia, even parts of Europe that don’t have their own automakers to protect. It’s only a matter of time. Elon needs to refocus on Tesla now, with newer cheaper models with more range.