Esme Es Mejor
Member
I really hope not. The FWDs are super cool & bring lots attention, so I hope they stay on future versions of the X, but the Y really needs to be simple to ramp & highly reliable. The X3/Rogue market is begging to be conquered.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
- Status
I really hope not. The FWDs are super cool & bring lots attention, so I hope they stay on future versions of the X, but the Y really needs to be simple to ramp & highly reliable. The X3/Rogue market is begging to be conquered.
The timing of Model Y unveil is tricky. Doing it too early may reduce the demand for current models. But doing early can help to lock in ICE SUV demands. I think 70% chance Model Y has regular doors.
I agree that it will have regular doors. EM admitted that his executive team talked him out of a completely new platform for the Y and instead will use the '3' platform. EM starting to show signs he is maturing away from hubris and instead making something efficiently at less expense.
dc_h
Active Member
Could go either way, but platform will be shared. Hopefully after closing in on 10,000 weekly rate, they’ll build the Y on the same line. That will keep demand stretched out for months and make funding GF 2&3 (or is it 3&4) much easier.
MitchJi
Trying to learn kindness, patience & forgiveness
So vertical integration continuous. What are the pros and cons?
Owning the value chain is a clear pro. The list of industries Tesla competes now against is getting longer and longer....
Tesla is developing its own custom AI chip, says Elon Musk
Does anyone have a good idea based on when the team was formed and the length of time development of a new chip typically takes when they might be ready to roll it out?
Does anyone have a good idea based on when the team was formed and the length of time development of a new chip typically takes when they might be ready to roll it out?[/QUOTE]
Tesla appears to have had key members of the chip team in place by April 2016. Tesla poached a team of chip architects and execs from AMD to develop the next gen Autopilot
Don't know how long the development process takes.
MitchJi
Trying to learn kindness, patience & forgiveness
I believe that they believe that they can produce excellent MY’s that are reliable and that they intend to do that. Elon initially said that the MX would sell at a small premium compared to the MS and they finally hit that a while ago when they reduced the MX price. I believe that they know that they can do this now and that they also intend to do that now.
If so doing that won’t stop them from competing with the X3 etc.
Of course I could be incorrect.
Recently a company said they produced the first batch of sample chips (Tesla's AI chip design) for Tesla. I think it's in advantage stage. Otherwise Tesla wouldn't have the confidence to say their AI chip will be the best in the world with 1/10 the cost.
Although Tesla denied the news at that time, we know those sample chips most likely were delivered. That CEO can't make a wild lie like that in a conference call.
I don't know how much change they have to make to the samples and the production yield. It's very difficult to estimate from outside. I think Tesla's two top priorities are Model 3 ramp and autonomous driving. I imagine Tesla's AI chip gets top priority.
He's just plain wrong here, if he was quoted correctly. We don't even have a definition of general AI.
He's correct about the extreme danger of "AI" projects. The threat, however, is idiot savant AIs, which are *not* general AI. We are well into the development of many idiot savant AIs, and very far from general AIs. If idiot humans start treating the idiot savant AIs as if they are general AIs, and giving them work outside their capabilities...
...well, let's just say that the idiot savant physicists who developed nuclear bombs never bothered to talk to biologists about the long-term effect of radioactive fallout. We don't need automated versions of those idiots.
Idiots, indeed.
P.S. I can give a much more specific example. I am not aware of any current succesful effort to mate up logic-and-reasoning-based "AIs" with pattern-matching "AIs" in more than a very primitive fashion. Human brains have a lot of specialized subunits which cooperate and argue with each other, of which some are logic-and-reasoning, some are pattern-matching, and others are... different things which we haven't replicated yet. The current AI efforts are essentially about replicating the subunits, not about developing the general, integrative, or "oversight" intelligence faculty. This is actually a problem. A true "general AI" should be able to say "Well, I *could* give you a self-driving car, but it'll just be stuck in traffic -- wouldn't you rather I moved your home closer to your work?" Or "Well, I could invade Iraq, but based on my study of history, that would create a geopolitical backlash and hurt the US, plus according to my study of law, it's illegal, and so I'm not going to do that." We aren't anywhere close to that. Maybe it will happen eventually, but the actual serious danger is *before* we get that.
Last edited:
One factor in Tesla's favor for timing regarding rolling it out is that the GPU/ AI chip is on a daughtercard in the autopilot unit. So, assuming they use the pre-existing connector pinout, they can test it right away and once it is proven, switch to it on new vehicles and retrofit older vehicles if needed.
I was following AMD for a while (always a fan of the underdog
) and remember they re-hired Keller a while back to start working on the new "Zen" architecture, which rolled out as "Ryzen" for desktop and mobile and "Epyc" for servers about a year ago and put AMD back on the map.According to Wikipedia, Keller & Co started working on Zen in August 2012 and he left AMD again in September 2015. This roughly coincides with the time when Zen was reported to have taped out and confirmed working properly - so I would say 3 years from start to finish of the design. If you are Intel you can throw more money at it and maybe do it sooner, but I don't think Tesla can dedicate more funds to this than AMD could.
Now this is only the "design complete" phase, mind you. So whatever was reported as first samples shipped could have been the so called "tape-out" when they make the design into silicon and see how it actually performs. They may do several iterations ("spins") which, as far as I remember from AMD's difficult past, can take months as things are fixed for thermal issues, insufficient clock speeds, manufacturability, bugs, etc.
I'm not an expert, and certainly don't know how long it takes to design an AI chip vs a general purpose CPU, also, if using NVIDIA's chips is any indication, an AI chip may be closer to a GPU design. In any case, I don't think 2 years is enough to get from zero to mass production, but this is Tesla and they do crazy things like that all the time.
3Victoria
Active Member
It will depend on the complexity of the chip and how many subsections it has. A CPU has many components. The Tesla chip may be much more regular: mpu+cache+bus-matrix; with multiple copies. I can see them having at least a preliminary kick-at-the-can to check out their ideas. Keller has great experience.
I feel the Tesla AI chip is above normal sense of vertical integration. This has big implication down the road.
Tesla said they want to become the best AI company both in Software and Hardware. This shows they are not just talking, they have been seriously working toward it. Maybe more people will realize, after all, Tesla is an AI company, not a car company. The valuation of an AI company is totally different from a car manufacture. If you have the best AI system, on the spot, you probably can command a $200B market cap. Because you could modify your AI system and use it in other fields too.
If Tesla gets the system to work, with 1/10 the cost, and consumes much less energy, they could sell the system to the world, bring in $100~$200B profit each year. (Regarding selling Autonomous system to other companies, Tesla said "Never say never". They might do it.)
Autonomous driving is the center piece to achieve the Master Plan 2. If the hardware is controlled by other companies, that hardware company would try to develop software and supply autonomous system to all the car manufactories. It would be difficult for Tesla to stay in the leading position in ride sharing. When Tesla controls both AI hardware and software, they can move so much faster, and makes it difficult for other companies to copy.
A working autonomous driving system will help Tesla to sell as many cars as they can produce with healthy margin.
Anyway, I think the Tesla AI news is huge. I remember long time ago we had a discussion and we all agree it's certain Tesla will develop their own AI chip. It's difficult and costly, but makes perfect sense. When they do it, they will disrupt it.
That is our problem too. Everything bad that has happened to me is a direct consequence of my stupidity, including health problems, because a subunit of the brain was dominant at the wrong time. Some of that "cloud" has been lifted with impotence. I'll be "cured" soon.
Drumheller
Active Member
Tesla is going to be competing against NVidia, Intel, AMD and others to develop a general AI chip. It may be that they are, instead, looking to develop a niche AI chip specifically for transportation to be used in vehicles only. Either way, this is a long project and not something I would consider relevant to the stock price for the next three years, at least.
Only if they plan to sell the ICs to others. If not, they are transferring all of NVIDIA's current profit of sales to Tesla to their bottom line (along with gains due to chip cost optimization).
DNN capability is widely applicable, autonomous driving is only one application. I predict they will have it (new chip) in vehicles within the next 2 years.
Also predict it will be used in SpaceX's initial Mars payload as part of the brains of the site preparation robots.
An AI chip will mostly need a lot of multiply/add units, which are fairly simple circuits (simple enough to be in the entry level course digital electronics I followed 25 year ago). There is porbably a lot of control logic around it, but still this should be a lot simpler than building a top of the line x86/x64 chip (what Keller did in the mentioned 3 year period at AMD). It probably looks more like a DSP or GPU than a general purpose CPU. I don’t think it would need a lot of of cache (at least not in the sense of L1/L2/L3 cache on a CPU), maybe just some on chip memory. If Jim Keller can make a tape-out of a top of the line x64 chip in 3 years, he can get this AI thing to the tapeout stage a lot faster. I wouldn’t be surprised if Tesla is testing their first silicon right now.
Wow, that is shocking. So the bottom line is that in order to save the uneconomical coal power plants, the government will pay for all their running costs while they can sell the energy at whatever cost making 100% profit. This sounds much more like communism than capitalism...
- Status
Similar threads
- Sticky
