Why are Tesla’s self driving chips so much better than the competition’s?!
Vložit
- čas přidán 13. 01. 2021
- If you wish to support us in creating fun, informative content, please consider giving at our Patreon site here: / drknowitallknows . Thank you!
Wherein Dr. Know-it-all looks into the details of why Tesla's "inference engine" autonomous driving chips are so far ahead of the embedded chips used by other auto manufacturers. How do embedded chips function, traditionally? How do Tesla's chips do it differently? Why can they operate so much faster and more efficiently than other embedded chips? And what the heck does all this have to do with the space shuttle?! Watch and find out! Gone are the long waits at charging stations: Chinese electric-vehicle startup NIO is pioneering battery-swap systems, challenging Tesla and other rival car makers. Here’s how NIO and Tesla are racing for the world’s largest EV market in China.
Want some awesome Dr. Know-it-all merch? Check out our awesome new Merch store: drknowitall.itemorder.com/sale
For a limited time, use the code "Knows2021" to get 20% off your entire order!
You can help support this channel with one click! We have an Amazon Affiliate link in several countries. If you click the link for your country, anything you buy from Amazon in the next several hours gives us a small commission, and costs you nothing. Thank you very much for helping support the channel by clicking our link and shopping on Amazon! (If you live in a country not listed here, please let us know and we will add it!)
USA: amzn.to/39n5mPH
Germany: amzn.to/2XbdxJi
United Kingdom: amzn.to/3hGlzTR
France: amzn.to/2KRAwXh
Spain: amzn.to/3hJYYFV
If you would like to purchase a more powerful wireless iPhone/Android charger for your Tesla Model 3 or Model Y, check this link: bit.ly/3b5fHSW
If you use the coupon code DRKNOW on your order, it will automatically take off the price of the 64GB MicroSD card, which is very nice!
If you are looking to purchase a new Tesla and would like 1,000 free Supercharger miles, just click this link and use our referral link: www.tesla.com/referral/john11286. Thank you!
Music by Zenlee. Check out his amazing music on instagram -@zenlee_music
or CZcams - / @zenlee_music
Video edited by Daniel Gualdron. Check out more of his work here: / @danielgualdron8504
If you enjoy more outdoorsy, adventure type videos--like climbing the Matterhorn or trekking to Mount Everest Base camp--and a few product reviews too, please check out our sister channel, Whole Nuts and Donuts, right here: / @wholenutsanddonuts5741
Please ask questions in the comments below, or at: DrKnowItAllKnows@gmail.com. Thank you!
Twitter: @jkundert
Instagram: @johnthekg - Věda a technologie
Nicely presented for such a complex subject. Tesla owners can point friends to this video when people say that, "I am not handing control of my car over to a computer because what happens when it crashes !!". Great video, but I am slightly biased...
Thanks, Paul. And than you so much for all the insider information and insight. Couldn't have done this episode without you!
Even before this I'd point them to the question "what happens if you have a brain aneurysm?" People seem to forget that our brains are basically just squishy computers.
@@DrKnowitallKnows
X-peng just announced a chip 7 times faster than tesla
@@golfish8589 it's actually 4 processors from Nvidea . and they are comparing to tesla's first chip.
Couldn't ICE cars catch up by using an NVidia chipset already purposed for EVs, stuck behind the glove box? That's not all they'd have to do, but that would get them a long ways there. (It's my understanding that the jungle of other embedded processors poses a significant challenge to legacy auto makers.)
I feel that this technology developed by Tesla is highly under appreciated
For good reason.
So much data. So quick. So entertaining. Thank you!
Very good and timely explanation, thanks!
Especially with the extremely bloated introduction by NIO... 👌🏼
very useful presentation, thanks !
Great video... I work for Infineon in Kentucky...
Thanks for that information.
Great and new content information!
I've ordered the tri motor fsd cyber truck..I may also buy the S Plaid.. I CAN'T WAIT!! i got lots of solar to charge them BOTH & a backup battery system as well..
Thanks. John
Thank you...!!
Wow man, super technical! but supper informative. Thats so interesting all this technology behind the chips. People and automakers may have no clue about how much behind tesla they are. Holly crap that blows my mind.
This has been, more or less, public knowledge for years. Thanks for the video. My point is that how the heck do people like Gordon Genius Johnson get away with lies and slander on national media outlets when they are very easily proven as such. 🤦♂️🤦♂️
GORDON!! He's my favorite. I get a smile every time something (not sure I can say the word here) comes out of his mouth.
Thanks to people like Gordon, people like us could buy TSLA for less!
You and me both friend, why isn't Trevor Milton locked up. I mean C'mon just like dickless Johnson every word he spoke was an utter Lie.
gordon has made people a ton of money by doing the exact opposite of what he says.@@DrKnowitallKnows
Thank you!
Great video thank you! Only comment calling it the "$10,000 Autopilot option" could be confusing to people since "Autopilot" comes standard and it's the extra FSD features that cost $10,000
How does the millimeter wave radar ties in with the system you are describing?
It's fed into the information, but from what Andrej Karpathy has said (see my videos on him) it's mostly a redundant backup for the vision system--as are the ultrasonic sensors.
great and smart video
Brilliant, this is unique insight into Tesla HW3, which I think is not available anywhere else.
What about tesla hardware 4.0 and probably versions 5 and 6 which are being worked on already.
V 4 was already being worked on when Elon revealed Hw 3
for those not familiar with "chip design" _ yes I were a computer developer - software, hardware, firmware...
Anyway, aside from the machine code, there's the "instruction set" that lead to ASIC, RISC, and a few other application related designs. ASIC= Application Specific Integrated Circuit/computer"... eliminate the unneeded functions and the system is much quicker/efficient.
Nice technical info. Would like to know the comparison against other EV computers (comparison with ICE is worthless nowdays in my opinion). For a future program I would like to know more about the Safety System like the architecture, functionality, what independent sensors (if any) are wired to this safety system, what actions does it the safety system takes and the interaction with the control CPUs. BTW, Iam a Control System engineer. Thanks!
Highly intelligent, I see what you did here. You only made the comparison that matters Tesla vs ICE. ⚡️🥃🥃
My new favorite channel
Nice explanation. I was hoping for a comparison with nVidias solution(s) though, which is what NIO says they'll be using (though according to nVidia's press release back in '19 those chips will not be available in volume until early '22). If I remember correctly one of the key parameters that prompted Tesla to roll their own was power consumption (and probably price too).
Test will release something better by 2022.
Couldn't have explained it better myself.
Great content. You are rushing the delivery. Thanks for this video.
Nice presentation. I disagree regarding Wrights law. They leverage existing manufacturing platforms (fabs) so their manufacturing costs for the chips are pretty much fixed. However, the chips are probably relatively inexpensive and the volume amortizes the cost of the chip's development over time.
My head hurts listening to this but probably because it is good stuff
I don't see the merchandise link
Hey Dr Know it All, thanks for another great video! Can you provide me with an answer to the following question: as i read the FSD is fully redundant, because Tesla alway’s runs the current software version and the next version on every car. To my knowledge, both software versions are compared, so every output from the current version is compared to the next version and in case of different outputs, it’s send to Tesla to see what happend so they can adjust (read: improve) the new software. As i understood, this makes that all drivers are testdrivers for Tesla to improve the software. As i see in your explanation, both computers run the same software and compare the outcomes. Now i’m confused, because why would they compare the output from the samen input, same software and same hardware, only for safety and redundancy? It seems to me that comparing software versions is way more plausible / effective.
ICE cars could have a similar chip placed in the passenger compartment and also be fine right?
Could you make a comparison of the Tesla chips vs the new NIO ET7?
NIO use Nvidia with about 7 times the pure computing power (TFLOPS), but their 11 8MP cams will need about 9-10 times the computing power, and then there is also LiDAR. And with the higher visibility of the cameras (more objects to be tracked), the computing power seems to me to be too little. Or the specified visibility is simply not used and is mainly used for marketing.
Great video as usually are you telling us that Tesla cars are more computers on wheels LOL.
I wish the sensors were redundant. The triple cam unit behind my mirror failed and I lost autosteer, cruise control, auto headlights and auto wipers for two weeks until Tesla could replace it.
Nio just released the info that they are using 3 Nvidia chips and tout 7 times more computing power than Tesla. I think Tesla optimized for power efficiency on their silicon. Can you do a segment on other EV ADAS computers? Thank you! Great content!
Not only is the FSD computer tucked inside of the passenger compartment, but my understanding is that the entire motherboard is WATER-COOLED as well. The housing of the FSD computer ties directly into the vehicle’s cooling system to extract waste heat. With Model Y and now Model 3 having heat pumps and octo-valves, Tesla can route chilled water (just above freezing) to the FSD computer, keeping it even cooler than before.
It's very interesting how the difference in engine heat affects the performance their computer is able to achieve. I hope other EV makers also take advantage of that.
John, one thing I didn't hear you talk about is the fact that the Tesla CPU board is connected to a heat sink that has fluid from the car's heat management system moving through it. I assume this is connected to the "octavalve" for controlling the temperature.
That's awesome.
@@neeljavia2965 Well hey, I did not know that. Thanks for that really awesome info!
@@DrKnowitallKnows Actually Jim said that.
Not me.
He provided the awesome info.
Link for explanation and photos of the octavalve. First used in the Model Y, I think Tesla is migrating the octavalve to other models. Its a genius system. www.tesmanian.com/blogs/tesmanian-blog/deep-looks-into-tesla-model-y-s-octovalve
With heat scavenging, as all CPU power is more or less eventually turned into heat, that could offset other heating sources, so may not be such a loss, except in the heat of summer.
When do you think Tesla will be putting their new revision self driving chip in their cars will it be by June?
As I’ve said on other videos, I believe that Tesla is actually producing their HW4 chip already, just that they haven’t announced it yet. Tesla asked for a importation tariff waiver by the Trump Administration for chips that are manufactured in China specifically for their FSD computers. However, during Autonomy Day (April 2019), they stated that the chips in HW3 were made in Texas by Samsung. Why the change from Texas to China? New architecture? New supplier?
I read some articles yesterday that there was a worldwide shortage of semi conductors. They mentioned a lot of other car manufacturers but they didn’t mention Tesla, I wasn’t sure if it is affecting Tesla as well?
I think rather less. The other manufacturers had corrected their expected sales figures downwards and thus probably also canceled orders for the chips, which are now missing again due to the slow ramp-up. Tesla stayed with the 500k vehicles, despite Covid.
What if the motor fails on steering or breaking? Is there redundancy there?
Yes, the position information is feed back into the FSD computer. Also Tesla measures acceleration which also provides feedback that steering and braking changes are taking effect
also steering has some redundancy
14nm? I expected finer tracings on the chips but ok still makes sense as it's still very efficient and can't wait to see what the new 7nm chips can do, the efficiency must be insane and I imagine it would be very robust of course due to it's smaller scale.
Actually, "robust" goes down as semiconductors scale "smaller". Quantum fluctuations are becoming significant for the bleeding edge of industry.
Other "competitors" can use NVidia's Drive solutions. NVidia are pretty good at this stuff too! Probably a couple of problems though, you still need to develop your FSD software solution, collect data, pay a higher price for the hardware (NVidia's markup). End of the day seems Tesla again have an unobtainable advantage over other EV/autopilot manufacturers for the foreseeable future.
Tesla and SpaceX are both producing a lot of unobtainium isotopes. Others cannot compete.
Next iteration will be out this year.
Whay are the CZcams CCs in Dutch? 😳🙈
hello, Wery good video! I probably didn't understood all of it :) but I have a question: When you say "Tesla’s self driving chips are so much better than the competition’s" Are you also considering the recent NIO announcements? They seem to claim they are 7 times better than TESLA in at least some aspects.
I think a video regarding the NIO solution would be a good idea.
7 times faster on raw computing power (TFLOPS), that doesn't mean better.
@@Zedus-rl9hp exactly :) but it might... I do not know what does it mean? this is why i asked
@@procentskladany it depends on what runs via software and what has been integrated directly into the hardware -> ASICs are far more efficient, but only for a special function.
Since Nvidia is an "open" system, there will definitely be one or the other weak point here.
The "7X" performance advantage is comparing present Tesla product vs Nvidia volume release in 2022. But Tesla is a moving, accelerating target.
SOC are impressive design
Since the call can control its own temperature if the it is in an extreme temperature climate, does that mean it will burn battery slowly over time just to maintain?
even when just in idle or not in use?
Good question - there are options.
Firstly if there are very high temperatures in the car parked and you want to use, then you may want to pre cool the cabin (via the Tesla app) which fixes the problem. Same for cold temps.
Very quickly after starting the car the AC will come on, and when in spec, the FSD computer will be available.
This is for now, and won't drain the battery.
In the future, with FSD taxi's this is never a problem as the vehicle is always in use.
@@zachtaylor226
Yes. If CPU/battery temperature is critically high/low, the batteries are going to power a heat pump to maintain a safe temperature.
Even while within temperature range, the computer runs often enough to implement the standby requirements, consuming power.
This is why you should maintain your residential electricity and keep your idle Tesla plugged-in, especially when on a long vacation or below freezing.
@@Paul.Chaffey very good point, i was under the same impression for the robotaxi as you!
@@imconsequetau5275 beautiful response! exactly the information i was looking for thank you
I think Nano Dem tech will help with new chip and board design
6:00 spider on the wall
Ladybug?
Actually, the Tesla vehicle chips could be better thermally protected. Each door, rear hatch, boot, and cabin roof will need a thermally robust embedded chip to operate items at this location, all the chips running stuff under the hood, or just inside the 'fire wall', could go in a thermally regulated box under the hood that would be far easier to service. In fact, a better car layout would be the charging port just forward of the front passenger door, all the high voltage electronics including inverters in a thermally regulated box next to the wheel-well with cables out to the drive motors, heat pump motor and battery. Then all the low voltage electronics could go in another thermally regulated box next to it. The rest of the bay could be taken up with the 12V battery, heat pump, cabin air con system, air filtering system, brake system, steering system, driving lights, storage space for cables, and the radiator with its airflow path isolated from the systems bay. The electronics boxes are are cooled by pumping fluid through a cooling plate to which the circuit boards are mounted.
Someone asked me this question so I thought I ask you.... What happens with Tesla's FSD when its hardware get old (say 5-7 years old) and may become unreliable or just fail when FSD is engaged? Do they have a safe mode so that when FSD fails, it will "safely" stop the car?
When they get to 5 nm they'll probably be sub 50W. Imagine that, an LED bulb's worth of power driving your car.
A 50w LED is nothing to scoff at though
No, because they will need bigger chips to improve their self driving software.
@@user232349 That's partially right.
@@eugeneputin1858 Sub fitty includes anything _undafitty._ 🤣
So are these computers up for the task of full self driving?
Outlier conditions will always cause problems, but the important measure is whether the public in and around any vehicle is safer overall with FSD turned on. So, when the statistical balance is 100:1 in favor, is there a real issue? And each encountered outlier is a learning experience for the entire fleet, not an individual.
@@imconsequetau5275 I was just wondering if this gen computer has enough processing power to grab that last "9" that makes the regulators happy? If it is, what is the reasoning behind the next gen? If the next gen is required to grab that last nine , when will it be out, and what will retrofit cost be for all those who have purchased FSD in car that doesn't have enough computer to actually pull it off?
@@pathfollower
The issue is both with training the optimal neural network parameters and with carrying them out, thus both Dojo processing and FSD hardware levels.
The latest cars have enough power to implement some arbitrary "depth" of neural network, just not necessarily what Tesla or Elon is totally satisfied with. There is satisfactory and then there is Elon aspirational.
Let's say, for example, a FSD-3 computer can always keep up (real-time) with a _12-level_ neural net and pass the regulatory hurdle. It is still possible that the Dojo supercomputer has built a 10X better-behaving solution that requires processing a _15-level_ neural network in real-time, but the present car computer cannot process that fast.
Tesla would want to offer the faster hardware that can handle the optimal solution. It might be attainable for mass production within the normal CPU improvement curve in two years. Then Tesla would offer the FSD-4, 15-level, 99.999% version, standard on new models but the FSD-3 version operating to only 12 levels assures merely 99.99%, so that becomes a hardware upgrade option.
Great video! Would be great if you can do a video comparing Tesla's HW3 to the recently announced Nio hardware & sensor suite (claimed to be 7X faster than Tesla's FSD, comprised of 4X Nvidia Orin chips) and also the higher megapixel camera. And also the overall significance/relevance of that (if any) in terms of overall autonomy capability.
This is hard to judge exactly. The figure may well be x7 computing performance over the Tesla HW 3, but you have to take a number of things into consideration. Firstly Tesla has 12 ARM cores (3 x quad ARM cores) which initially was used to make the SW perform the same as the older Tesla HW based on the Nvidia chip set. Initially the features for FSD were behind the HW2 but caught up with SW updates.
What has actually happened is Tesla has been moving more and more features into the neural network part of the chip freeing up the ARM cores for more standard processing items. With what I have seen with the beta FSD video's is that Tesla is not performance limited with their chipset, and software is key. I would think Tesla knows how much processing power is necessary and have designed accordingly. Thus saving money and power comsumption.
Wanted to ask the same.
@@Paul.Chaffey Thnx for your insight.
But don't you think that Tesla would upgrade to even powerful chips in hw4.
@@neeljavia2965 It's on the list now. Thanks!
Yes I would think that they will add more powerful chips in HW4. The thing that I would expect for HW4 is a technology node reduction (7nm) which will reduce area, improve perforance and power. Also as doing a respin of their chip at this small technology I would expect them to add / refine the chip architecture even further as Tesla will have a better idea what they need going forward with their SW. Power comsumption is important with BEV's - Tesla (~6 billion transistors) < 100 Watts for the system. The NEO Nvidia solution requires (~17 billion transistors) and ~ 65 -> 70 watts per chip = 260 Watts !
This topic is music to my eyes 👀
This subject is paintings for my ears
@@polishfish lol
🤣
Re space, although peripheral, you say you like 'space history.' How can you have flexible spacesuit in a vacuum? In fact, more than flexible, the 'suits' hang loose on the 'moon-walkers.'
Insulation and most other layers are _outside_ the pressure barrier, so they "hang loose".
The new Orin from Nvidia is about 2x the compute with only half of the power. For example, the dual-chip version gets 400 TOPS at only 130 W power consumption or the smaller version gets 100 TOPS at 40 W, compared to HW3 from tesla that gets 144 TOPS at 100 W power consumption.
This is for 1 chip. NIO are using 4 NVidia devices per complete FSD computer - I would estimate ~ 260 Watts for this board. Need to compare Apples to Apples...
TOPS is not all. If the chip is not optimized, the highest computing power is worthless.
Nvidia is an "open system", so not always the best choice, keyword ASIC.
@@Zedus-rl9hp Nvidia has tensor cores that are specialists for the operations that are used in ML applications. If you look at all the importer metrics the new Orin chips are better than HW3, faster memory, more tensor operations, direct support for RNN type networks, more moderns CPU cores, and build on more modern architecture. In the Chip industry you the development is fast and hardware that was the best for an application can be outdated in tow years.
Overall a good summary, one correction, you stated that they have produced a million cars with these chips, this is incorrect. The hw3 only came out in April 2019 and in limited quantities. Many cars in 2019 were produced with non hw3 chips. 2020 manufactured 500k cars, and 2019 was around 300k with 25-50% of that pre hw3. So realistically 700k cars with hw3. Just splitting hairs I suppose. Overall your assessment that Tesla is far in advance of legacy auto is correct. Though a mention that as legacy auto transitions to electrification, they to will have the opportunity to redesign and learn from the choices that Tesla has implemented.
Just splitting hairs, but are not the hardware 2.5 cars being upgraded to hardware 3.0. That would take the total to around 1 million! 😉😎
@@cazcam2000 I was also counting what's in the pipeline for all the cars they're making now. They have to be sitting in a warehouse somewhere waiting to be installed :)
@@cazcam2000 they are only being updated for the buyers that purchased FSD. If they didn't then they are stuck at 2.5. If they choose to upgrade later the hardware will be installed. I love the fact that the older cars have an upgrade path.
Where do TESLA find their engineers. Thx again but this is significant in terms of redundant systems considering the flexibility you have in a system like the FSD board. Eg. you can run different code on each CPU that should give the same result and only compare the result. This reduces the chances of software bug killing a system considerably So not only can you now check if the CPU is working properly you can now verify that the software is working properly. WOW!
Really makes the fleet robust. Compare this to released software updates that have bankrupted other companies or permanently damaged their reputation.
Actually, it's quite the opposite. Expecting ALL your raw input processing to be done by ONE UNSPECIALIZED CHIP is NOT redundancy but rather putting all your eggs in one basket. As computers have also proven, one fast processor will never match specialized processors feeding one main processor, and DEFINITELY not a multi-core PARALLEL system.
So it's neither redundant NOR faster...typical of the thinking that also goes into CrApple products.
So much better ...😏😏😏😏
U tells its better 😂😂
Tasha Keeney couldn’t put it better!
Not only is the tesla FSDC better then any ICE computer, but it is also better than any EV computer.
Yeah but what about NVIDIA's Orin?
It has about 3x the performance at only 30% more power usage so let's not talk about that. /s www.golem.de/news/drive-agx-orin-nvidias-automotive-aquaman-liefert-200-teraops-1912-145610.html
@@MrAntidodo Tesla knows what kind of tech they need for their self-driving.
Do you hide "plant food" in your plants ? ;)
Answer to question in title starts 4:30
You are wrong about temps variance last I checked the hottest it gets anywhere in the world is like 130degrees f not 212 and the coldest would be negative 30 degrees f for anywhere the only mode of transportation isn't snowmobiles. The main reason for a hot engine is the fact it's a combustion engine the worst an electric vehicle would have to deal with is heat from the battery but even then if your batteryg is getting super hot you want to shut the car down anyway
Interesting. I think that the biggest obstacle to FSD will be the safety regulations. Looks like Tesla is taking the right step to resolve that.
I just ordered a performance version of the model Y with the 10k FSD 9 days ago.
I feel that this decision is the wright one.
Hahaha. Nice. Wright's law at work again! Enjoy the car. I'm _this_ close to getting the $2k speed boost option on my MY long range. It's just so much fun to dust everyone silently when a traffic light turns green.
@@DrKnowitallKnows get it, then make some videos and monetize your model Y
@@DrKnowitallKnows make a video about the upgrade. Perhaps you can write off the cost as a business expense.
@@DrKnowitallKnows yes do it! I learnt a lot from your videos on the model Y and it did help me make an informed decision.
So why wouldn't the ICE makers just put the computers in the cabin?
Of course they can. Most computers in an ICE car (and there are lots of them) are not in the engine bay. It would be no problem at all to put a 50w chip in a ICE car and it would work just fine.
Because it was customary for every supplier to deliver their own control unit, which is then often installed directly on this component.
The whole thing was then only plugged together via cables and the various control units could then partially communicate with each other or the information came together in the instrument cluster.
But Tesla _actively_ heats or cools both the circuit board and battery pack.
Tesla chip resembles Apple M1 SoC, having ARM cores with gpu, npu and memories.
I agree - seems like the big players are all moving in this direction - Apple, Nvidia, & Tesla
Automotive computers have been located behind the glove compartment and the in passenger cab for decades. Computers located in the engine compartment are located in the intake airbox for cooling. GM, Toyota and Honda make great computers. Tesla pushes the envelope right the edge. The big automakers are more conservative because of their size. GM is going after Tesla and they will give Tesla a run for their money.
That's funny GM is all talk.
plz compare the tesla chip to the nio setup; nio says their boardcomputer is 7 times faster than teslas but they have 11cam's with 8Mexapixel=88 and tesla has only 8 cams with 1.2MP=9,6
88/9,6=9,166 times the data but only 7times the speed => therefore the tesla computer should still be faster cause they have way less data to process
And Tesla has over a million cars on the road delivering data and training the neural network. Soon they will have the Dojo neural net training computer. Tesla is only using a fraction of the computing power for their FSD beta, it is only a tenth of the overall capacity, as far as I remember. Why to use significantly more expensive and energy consuming hardware when it is not necessary. It only drives the purchase price up.
Nio probably needs the extra compute power for LiDAR computations. I suspect that compute power more than anything else was the reason Elon said that LiDAR is doomed. Four Nvidia Orin chips = 288W vs Tesla's 100W consumption.
@@tsh4k Obviously this has some interest. I will definitely put it on the list of topics to do. Thanks!
@@tsh4k Very insignificant energy consumption difference. It's all about safety with NIO so they want to be perfecting the tech even if it's not as energy efficient yet. TESLA is taking the route of experimentation and improvement down the line.
The major issue I have with their "7x" claim is that those nvidia chips are very generic chips. That means they have all kinds of processing units so every manufacturer can use them with whatever software they come up with. Tesla's chips on the other hand are custom designed for the software that runs on it. There's no overhead from modules they don't need for their software.
Also, Tesla's chips were never designed for maximum performance. Otherwise they'd be 10x as big, and would use 10x the power, and there would be 10 of them on a board, and the whole passenger compartment would be filled with those boards. They were designed to be just as powerful (with some margin) as needed fro the software while using the minimum amount of energy possible.
nio's solution might very well be comparable to putting a 600 Watt desktop GPU into a mobile phone. Great graphics, but the battery would implode in a second while burning away your hand.
Stupid question, why can’t ICE car manufacturers also place the chips inside the cabin?
Yes they can do so, but if the cabin temperatures get very hot or cold, the ICE engine cannot just turn on and start the air conditioning - just imagine being in a closed space with ICE cars starting themselves, the air would become toxic.
@@Paul.Chaffey makes sense. Thank you!
The problem is rather that the control units are often integrated into the supplied parts, distributed everywhere in the vehicle (window lifters, transmission, engine, brake system, comfort systems, etc.). So that would be a huge number of "small, primitive computers" in plastic housings that would have to be relocated indoors.
And to run the whole thing on a central computer, the competence of the OEMs is still missing. This could mean losses for the suppliers as well.
So here several factors come together, the one with the temperature is probably the least.
@@Zedus-rl9hp that makes alot of sense. Thank you!
We will all be driving (or not driving) autonomous cars in just a few years. Once it is there ut will be l8ke the iPhone moment when in a few years everything changed very quickly (and Nokia, Blackberry like companies equivalent in automotive will die)
I read where several ICE manufactures are out of computer chips, so they are forced to shut down manufacturing their cars. Does Tesla have their own supply?
This could be a serious problem for Tesla. I'm not sure how much "lead" they have with Samsung, but I know everyone is feeling the squeeze right now.
@@DrKnowitallKnows
Would Tesla really ever negotiate a single-source vendor license?
Tesla
699 👍🏽
0 👎🏽
🙌
what about patents , how safe is tesla tech.
Thank you! This FSD computer (card) was the reason Herbert Dietz was talking about 'catching up' with Tesla would cost Volkswagen €52 billion of US$ 60 billion. The self-sufficient smug board members of GM and Ford have not the faintest clue about what this video is all about and think they can catch up with ease. They look at market penetration figures and are not worried because they expect the big EV breakthrough to happen after they are gone ("not my problem"). Look at their remuneration packages and you know what is going on. They need to buy car companies and chip companies and AI companies and even then they will be hard pressed to keep up for integrating all that is hard, especially as boards consist of bean counters and sales men. Don't believe me, just look at their board member roster. Tesla has a rocket scientist at the helm, GM and Ford have either sales-marketing people or bean counters at the helm. That says it all. One is only interested in creating a better future, the others are only interested in a better Christmas bonus.
One redundancy may not be enough. About 63% chance that it will work sounds low for the task.
Yeah sure it's great, but Nvidia isn't a joke neither. NVIDIA DRIVE AGX Orin chip can perform 200 trillion operations per second, which is almost seven times as many as NVIDIA's previous Xavier chip (30 trillion operations) and more than Tesla's FSD Computer (144 trillion operations per second). The Orin chip will be capable of scaling from level 2 to level 5 autonomous driving. And all other car companies have access to that Nvidia Drive AGX Orin chip. And I don't think the ICE engine influences that chip much I think they put the self-driving chipsets somewhere inside the passenger area type of area of the car. I think self-driving works better for EVs mainly because electricity is much cheaper than running petrol in an ICE car, especially when doing an Uber like self-driving network if driving around with these cars 24 hours a day. But if Teslas don't soon start supporting automated battery swapping on their whole range, then 24-hour Uber like self-driving service won't be doable. Comon Elon, Swap those batteries now!
i do own 2 stocks only...Tesla and E hang
TeSlA hAs No TeChNoLoGiCaL aDvAnTaGe
The computer is also cooled by the cars coolant system via the octovalve.
I'm not sure that the position in the cabin of the CPUs is actually an unique advantage to Tesla. Every manufacturer can place their, more or less sophisticated, control unit in the cabin. Maybe I'm missing the point... 🤔
Probably due to legacy concerns they don't, however. They certainly could in the future--though that will require a pretty big hardware shuffle, as most embedded chips are installed where they are needed.
Tesla actively heats or cools the CPU board and batteries. In-cabin location is just incidental to being close to the dashboard _heating/coolant lines._
"Everything that's wrong here is on me" is a strange intro disclaimer for someone who claims to know it all ... A bit like the muscle bound dude with the skull tattoo who walks up to a hot girl and then looks at his own shoes and says "can i buy you a drink?"
X-peng just announced a chip 7 times faster than tesla
I think Tesla knows what they need for self-driving. There's a reason they ditched Nvidia which is what X-peng using.
putting the processing unit inside the cab isn't new. Comma.AI does it, so does waymo and so did all of the participants of Darpa grand challenge. In the case of grand challenge, they couldn't fit it in the engine compartment if they wanted to. Most of these design choices were also done by darpa teams, if I remember correctly. Why are people surprised by this? It's not new at all. Just go read the papers published by stanford and CMU. Is it that hard to google or look at arxiv.org? The next gen isn't what Tesla did. It's making it a SOC to reduce power, increase efficiency and add FPGA to make certain tasks run in real-time sub ms speed. Microsoft pioneered brainwave to compile NN models into FPGA that run more efficiently. If you look at what xilinx is doing to make FPGA work better with next gen NN, that's the future. As transformer technique matures, better custom hardware is needed. Tesla is probably already designing the next gen hardware, which means in 3-4 years that hardware will be available and real level5 we will be feasible.
What is hiding in your pot plant?
That would be baby Yoda hand sanitizer. My son put it there. Good catch! ;)
wow. don't bet against Elon and team
Haha Americans just can't call it Celcius.. ;-)
2:35 I thought you were talking about political voting machines.
Yeah ok, but traditional auto could do the same as tesla and just put it inside. They dont need to put it under the hood. You put it as if they would not be alle to do that.
The Tesla CPU and battery have a conditioned environment at all times, is running at some % all the time; mostly thanks to the large storage battery.
@@imconsequetau5275 this is true but this could also be true for a traditional automaker. Nothing they could not change. This is no real unique advantage.
@@Johny1
Reminds me of "Everybody could have, Somebody should have, but Nobody did it. That's why Nobody got ahead of Everybody."
Tesla has a TWELVE year lead because they just keep _doing_ everything they can, unlike the ICE manufacturers that don't.
@@imconsequetau5275 yeah but putting it like that, is kinda Bad. This will be the norm, and it will not restrict them using consumer Hardware, they will not have to use anything less than consumer Hardware. This is not something to say when talking about what other automakers are limited to implement. Cause its literally possible and very likely to be implemented it this way.
So, ICE producers only need to relocate their car chips to the cabin, and Tesla's advantage is gone? That is like it sounds to me.
Tesla cars keep the cabin cooled in the summer even when not using the car. You can't let and engine run at anytime since it could be parked in an attached garage and suffocate someone.
@@bluetrepidation So maybe just turn the self driving off when parked?
@@Epacman17 it is mostly off. If you take a cell phone, turn it off, then leave it sit on your cars dashboard for a year in the sun it will eventually fail. Thermal cycling will kill it. This is why automotive chips are held to a higher standard. Tesla is using the HVAC to cheat a little bit and keep those extreme temperatures under control. Plus automotive chips are expected to last much longer too.
If you’re okay with feedback, might I suggest you get a better teleprompter or keep it farther away? Most of your videos are brilliant and really well researched. But, it’s clear you’re focusing on the words and you can see your eyes scrolling left to right. Finally, your speech patterns are mostly monotonous and could improve (which surely they will over time) so that it doesn’t seem like you’re reading a script. Use of words like “thus” etc. also makes it seem very scripted.
Hate to say it, but 'legacy' (i.e. REAL) car makers' vehicles rely on human brains which do 2 things NO chip (including Tesla's/CrApple's) can do:
Interpret and decide independently.
Also, parallel computing isn't about redundancy. It's about COMPUTING/PROCESSING SPEED. So if one processor goes 'oof,' you have a non-working computer.
If the processors aren't in true parallel, and each has only one core, then these chips are no match for a collection of specialized chips sending data through a controller CPU. Kind of like how CrApple got its ass waxed by WIntel. And this is why Teslas crash into semi trailers when 'inferior' ICE vehicles stop.
Sorry...Tesla is returning to the late 90s as far as SAYING they're ahead when their tech is actually BEHIND.
And how many years on active duty working on / around the most advanced technologies..?
And your advanced Engineering Master's / PHD (+ Military Experience...???)
I'm going go out on a limb here and guess .... ZERO...
Salty.
Do people realize that autonomous cars will not go over the speed limit and will brake going down hill to stay under 25 mph. We are going to be on a slow road to china. All following the one in front and I don't know how they will merge into a long line of cars. To stop a whole line of cars you just have to walk out in the road and take a nap.
Do you realize that you are talking bunk? Autonomous cars will do whatever we* tell them to do. Currently FSD will drive over the speed limit, so you are already wrong. Why on earth would thy stay under 25 mph going down hill (which they don't)?
@@CorwynGC I just got done watching two videos of Tesla newest beta update and they would go 5 over. We will see but if these cars can be programmed to go over the speed limit then that is guilty. by default. Maybe do a 5 second over limit to pass a car in front but full autonomous cars programmed to go even 10 over limit will never pass scrutiny. The 25 comment was if that is the speed limit which many residential and business limits are. My opinion. Every motor now is governed to a maximum which can be changed but usually unlawful.
@@CorwynGC I suppose Elon could put in a "We're having a baby" mode.
So what's wrong with complying with speed limits?
@@Zedus-rl9hp Nothing. I think people assume these cars will be like a cruise control but since these cars use maps the speeds across the country are known and will have to be obeyed. One mile over is against the law even if you don't get pulled over. Actually I think they could implement now that all cars can not speed whether it is autonomous or not. All it takes is a map and GPS. All throttles are by wire.
Sounds like you just read the Tesla marketing bull crap without actually looking at the actual competition. NVidia, Qualcomm, Intel and a number of other companies have systems they are selling that are at least as advanced as Tesla's system. By the way, which Tesla system are you talking about? Are you talking about the first, second or third system of "full self driving" boards? Tesla is going to release their fourth hardware version this year so they have been talking about full self driving for years without being able to implement it properly, they are main road and highway driving systems, not actual full self driving. I don't think you actually know what embedded chips are used for, Tesla would be using plenty of them. You don't actually think Tesla would use a 2.2Ghz multi-core chip to control the AC or ABS in the car I hope. Embedded chips control a lot of the input and output of anything that moves or changes in any way. The powerful computer board takes the information from the embedded chips and does its stuff. I recommend learning about what the other competitors are using before declaring that Tesla is the best.