current position:Home>As for the technology released on AI day, Tesla has been leading auto companies for at least 20 years

As for the technology released on AI day, Tesla has been leading auto companies for at least 20 years

2021-08-26 02:19:56 Luca car

hold " tesla " As the protagonist of the topic , I'm a little tired of this kind of thing . Similar new technology to almost every other enterprise , Whether advanced or not , Have to compare with Tesla's current performance , To get the answer , What's going on .

Essentially , Before the formal implementation of laws, regulations and supervision , Today's technology competition is more of a period of chaos , It has not yet entered a truly white hot era .

From the timeline , Tesla has already had its own autopilot days before 、 Battery day , Looking forward to their next technological development . among , There are many new developments , But it also triggered discussions in the industry , The general view is " Is this ?"

and 2021 Year of AI DAY, We have a new answer sheet in front of us , There is also a need for discussion , About AI Artificial intelligence .

These technical objectives , Other enterprises 20 Years may not catch up with

First, we need to pave the way , Tesla's 2021 AI DAY, The purpose is clear , The small meaning is to show your muscles , Tell you what the next development path of this enterprise is . And great significance , It's recruitment , So read down to the core of recruitment , You will gradually find , Why on earth , This press conference will be like this .

Just get a wave AI DAY Dry goods of , Don't put too much hard core data , Easy to understand :

1. There are four big things , Talked about computer vision 、 Chatting TESLA chip 、 Tesla robots will be in 2022 The prototype was launched in , And the most important Dojo supercomputer . Keep in mind ,Dojo above all , I'll go into detail below .

2. chip , Name is D1, Tesla self-developed and produced (designed and built entirely in-house), 7 Nanomaterials . At Tesla AI DAY On , Its data has such a line description "GPU Level of computing power ,CPU Of connectivity ,I/O Bandwidth is " Today's most advanced network switching chip " Twice as many ". With processing power 362 teraflops Of D1 chip , Tesla can use this to increase bandwidth 、 Reduce the delay 、 promote AI performance , Then it is calculated and transmitted at the same time ( Focus on ), Instruction set architecture customization ISA, Then with the training tile , form Dojo This supercomputer has pure learning ability .

A single one D1 Have 500 100 million transistors 、362TFLOPS Peak computing power of (BF16/CFP8), Power waste 400W within . except D1 chip , And the performance of training tiles , It integrates multiple chips for higher bandwidth , The ultimate computing power is , Each tile 9 petaflops And every second 36 TB Computing power . Final , The cabinet machine consists of 3000 star D1 Tesla composed of chips ExaPOD The calculation force is directly pulled to 1.1EFLOPS.

In the field of supercomputers , At the top of the world , The difference is , Its core is for learning and training , Not other compatible , Almost means more efficient .

3. Computer vision , This is one of the ways to build a digital model , You can understand it as modeling and matching similar to computer games . Every Tesla car , Will scan and observe like human eyes through their own camera head , Then it's like the cerebral cortex , Build your own nervous system . Through different algorithms and systems and Tesla's internal control system 1000 People's manually tagged data team , Put a panoramic world in front of the machine , Then make the next decision .

4. Dojo supercomputer , The king's bombing . The above chip and training tile , I believe the figures they bring are not particularly sensitive , It can be directly understood as refreshing mass production records , Even super double . before ,GPU There are no such high requirements , So the industry is mainly CPU+GPU Collaborative related processing scheme , Tesla uses a visual neural network FSD programme , Therefore, stronger GPU performance .

Far away ,Dojo, This cloud supercomputer , According to Tesla , It will be the fastest computer of its kind , At the same time, it will be directly connected with Tesla's models 、 Robots, etc , You can understand it as , Constantly in all practical modeling , Through high computing power , Try all the possibilities 、 Solution , Keep training 、 Keep on learning .

5. Tesla robot , Needless to say , Is to use... At the same time FSD The core 、 Use the camera 、 utilize Dojo Training 、 Edge calculation, etc , Finally let the robot learn , Have certain human abilities .

If the above section has removed a lot of obscure information , You read it and read it , Then we can talk about the next topic , What is the meaning 、 What can be done 、 What's the problem .

Yes, of course , remember AI DAY Starting point , The core is recruitment . Tesla has revealed a series of its core , Look for those who are willing to join it , And have the ability to promote technological progress , Then solve a series of existing problems , To create new breakthroughs .

If you believe everything above , Then the global auto companies must at least catch up with Tesla 20 year

We need to explain one thing in advance , You and I have seen a lot before PPT、 A great deal of PPT Car making , What they tell is generally , What should I am going to do? 、 What is the future in our eyes , Behind this, consumers need to recognize 、 Capital is willing to further overweight 、 Find your own entry point and promote it .

tesla AI DAY Upper PPT Relatively more advanced , No concept , Talking about all kinds of details , It's about " I'm going to do this ". in other words , It's far from jobs iphone 4 The kind of war of God PPT, Only a little short . After all , It hasn't really been made yet , It can't be sold , It startled everyone " So what he said is true ".

well , Let's talk about why we can lead 20 This topic in .

tesla AI DAY Everything posted on , In essence, it is further consolidating its own advantages : chip 、 Calculate the force 、 framework 、 Software . In today's , Almost the vast majority of automotive enterprises have not yet done so in terms of electronic architecture " On the way ", But Tesla has overlapped to the third place within itself X generation . after , yes GPU、 Chip capabilities , The above figures are simply translated , If you want to achieve the same ability , It may not be the ability of automobile enterprises , It's national capacity . And in this , Tesla's Dojo、D1 Later chip 、 Training tiles, etc , There will also be further iterations , So it almost means ," tesla AI DAY Brought about by the , Alien technology almost equivalent to dimensionality reduction attack ", a AI company , Will make all auto companies in the world , There's no way to chase . Yes, of course , Chasing or not is another topic , After all , Standards in the automotive field are not set by Tesla , Major countries in the world will pay attention to the intelligent network connection 、 Intelligent driving gives its own corresponding standards . therefore , That means ,Dojo、D1 Behind a series of exploding data , It may not exactly match the corresponding regulations , for instance L4 Level 1 autopilot requires mandatory matching of medium and short range lidar , That pure visual solution , It doesn't make much sense .

then , Let's talk about AI DAY behind , This series of points that need reflection , How many in the world .

1. How Tesla landed , The above series , The core problem is this aspect . First of all , chip 、 Training tile , How to make , Who will make . Design level , On the one hand , At the same time, we also interviewed experts in the field of chips , They are for Tesla D1 The interpretation of self-developed and self-produced chips is quite interesting , Because self research is indeed possible , And in the process of self production , Design your own chips / Design backend / Self production , Can be called self-produced . As for real production , Intel 、AMD、 Thomson technology ? well , The current ones don't match . meanwhile , The technology in TSMC's hands ( Or Samsung's solution ), Whether it can meet Tesla's index needs , Everything is uncertain .

2. Existing technology path ,AI DAY I didn't talk about too many concepts and directions , In essence, it is no different from the current situation of the industry . So the question came , Even in the national standard 、 Before the implementation of global standards , The vision brought by Tesla makes some clouds 、 Extremely high simulation ability , these , It will really have a great impact on lidar ? It's not .

Of course , Musk also offers other possibilities , Like the Tesla robot . since , In the future, automatic driving technology may not be a pure visual world or something ,AI DAY Tesla on , In terms of the results , Not much focus on cars , It is AI And software . In other words , If Tesla is allowed to build a country or world , The prospect of this approach is infinite .

3. Alien technology really landed ? Too much calculation , It's really necessary to have such a high computational power ? If from a certain point of view , Next , Will consist of each vehicle FSD4.0 Functional models -Dojo Supercalculation , Together, the edges are calculated + Large edge computing network , Accurately speaking , Such a high edge computing power , Not suited to the business model . As for what model can digest this computing power , This subversion of traditional means of production , answer , I really understand everything .

4. Last question , About Tesla's business model , Business model .

AI DAY A series of releases , We can see that Tesla is really an optimist /X War thinking / More engineer oriented companies , What does the industry do ? Is not important , What others think I'm doing ? Is not important . Anyway , Capital markets are recognized , At the same time, new technologies continue to drive the recognition of the capital market , With enough money ( And further coding cycle ), The stock has been rushing , therefore , Tesla can jump out of many conventional rhythms , Directly and boldly to all kinds of science and technology enterprises all over the world , To pass on , How did you land .

At the end :

Last , Show me your shallow conclusions .

Everything is not really made , And meet the corresponding standards mentioned in the press conference , I'm not sure , Leading all car companies 20 This will really happen in .

After all , tesla 、 The musk Flag Frequent instability . such as X Millions of Tesla driverless taxis will run on the road , Like our FSD Often it's automatic driving 、 Not for a while . Before these benchmark questions , Vehicle delay 、 bounced check 、 Update delay, etc , None of this is a big deal .

therefore , The whole is like this , about 2021 AI DAY Come on , Full of eyes " Alien technology " It's too big , Can Tesla surpass Intel at one time +AMD+ Mercedes +BMW+ Honda + Toyota , It depends on how you perform in the end .

copyright notice
author[Luca car],Please bring the original link to reprint, thank you.

Random recommended