Slashdot Mirror
Tesla Is Building Its Own AI Chips For Self-Driving Cars (techcrunch.com)
Yesterday, during his quarterly earnings call, Tesla CEO Elon Musk revealed a new piece of hardware that the company is working on to perform all the calculations required to advance the self-driving capabilities of its vehicles. The specialized chip, known as "Hardware 3," will be "swapped into the Model S, X, and 3," reports TechCrunch. From the report: Tesla has thus far relied on Nvidia's Drive platform. So why switch now? By building things in-house, Tesla say it's able to focus on its own needs for the sake of efficiency. "We had the benefit [...] of knowing what our neural networks look like, and what they'll look like in the future," said Pete Bannon, director of the Hardware 3 project. Bannon also noted that the hardware upgrade should start rolling out next year. "The key," adds Elon "is to be able to run the neural network at a fundamental, bare metal level. You have to do these calculations in the circuit itself, not in some sort of emulation mode, which is how a GPU or CPU would operate. You want to do a massive amount of [calculations] with the memory right there." The final outcome, according to Elon, is pretty dramatic: He says that whereas Tesla's computer vision software running on Nvidia's hardware was handling about 200 frames per second, its specialized chip is able to crunch out 2,000 frames per second "with full redundancy and failover." Plus, as AI analyst James Wang points out, it gives Tesla more control over its own future.
From the previous thread about Tesla, I expected this headline to read "Tesla is now building their own arcade cabinets".
#DeleteFacebook
You know if you've ever said anything nasty about Elon.
Now, his vehicles know.
Be afraid. Be very, very afraid.
Mimetics Inc. Twitter
So what, Musk is trying to be Bezos and do everything?
For better and worse, keeping things proprietary means it's by definition both closed source, and tested only to one's own environment. Although it produces fast yields, it doesn't have many eyes. Many eyes and many hours are needed to vet the integrity and edge cases (like cliff edges) before safety can be assured.
It's a risky, expensive, and proprietary endeavor. If everyone (systems builders) were using similar development, the testing age could be completed in a concurrent time, rather than a serial/iterative time. I'm betting against this turning out well.
---- Teach Peace. It's Cheaper Than War.
I'll be in the basement hiding from the cars
-- Thou hast strayed far from the path of the Avatar.
What number of Time You Lost This Droid Chance... On the off chance that you wana get a sensible salary through workstation and on the off chance that you have a solid web association then you ought to have the capacity to know how you make your wage by laptop.this is extremely easy to know just vist my site. HERE www.kingstar5.com kindly don't duplicate In Url Thanks Alot
And they created a rat with only 2 neurons.
"We had the benefit [...] of knowing what our neural networks look like, and what they'll look like in the future,"
Really? If they take their neural network development seriously I don't think they know what their networks will look like in ten years. It's a research area in the middle of a transformation. Using architectures molded into hardware is probably just costly and will act as an antagonist to innovation. I don't think having 2000 vs 200 frames per second right now outweighs that downside.
A car travelling at 90 mph is moving about 4 cm/millisecond. So going from 200 fps to 2000 fps is going from 20 cm to 2 cm per cycle. What's the use of recognizing a car every two centimeters? For a jogger at 9 mph it's down from 2 cm to 2 mm. It's neat and all but I don't see how that necessary to react in the time frames a car needs to react. Even if it takes 3-4 frames for the car to get a motion vector 0.2 seconds is still way quicker than a human and 0.02 seconds doesn't bring that much.
Live today, because you never know what tomorrow brings
As an existing owner, how do I upgrade to the new hardware system? Is it an over the air update?
I, for one, welcome our new Tesla AI overlords. Imagine a Beowulf cluster of Teslas.
This is how the Tranformers began. Onc humans are obsolete they will forget their creators.
Wow Atari managed to do this on 8/16bit MCU's ..,
And tesla requires a custom AI chip just to play games...wow
What happens if (or, when) Tesla realizes they need to make a significant change to their code?
Automated driving and AI are both hot research areas. I wouldn't take a bet that there won't be big changes in the near future.
This smells like an unholy combination of two things: a development team getting burnt by premature optimization, with just a hint of "painting oneself into a corner".
Between this and the omission of lidar, I'm not enthusiastic about Tesla's self-driving capability. My pessimism applies across the board---to its timeliness, reliability, and safety. I really liked it when Tesla delivered premium electric vehicles instead of the same hideous abominations that everyone else made. If I bought a Tesla, however, it would be in spite of their self-driving tech.
Volvo is putting a lot of effort into developing the tech safely before marketing it, and, most importantly, before putting it into consumer goods. At this point, I'd expect to feel safe enabling self-drive on a Volvo way before I'd do it on a Tesla. An OTA update doesn't reattach my head.
---
According to the latest ruleset, this post should be modded as Vorpal Flamebait +5.
Chip design is a very expensive endeavor. Musk does not sell enough cars to spread the cost. It will take consortiums to reach the volumes to make this a good idea.
As they have been advertising for a long time that what you buy now contains all the hardware required for fully autonomous driving with a free software upgrade in the future.
Nothing but dust.
Beat ya there are a lot of eyes on space x assets, ip, talent and contracts though.
Neural nets are remarkably dangerous for high level of safety applications; they inherently create edge cases that are remarkably difficult to find in M&S, and should never be certified for life safety critical applications, like autonomous vehicles.
So in your words, this is rocket science, yeah?
Too bad Elon knows nothing about rocket science. Not like he has created the most successful rocket company in history. Not like he was heavily involved in the design of his first rocket. Hmmm, wait a sec.... Nah, some random internet user who has never achieved anything def knows better.
After Intels intense lack of concentration, and soon to follow Nokia, it is now Nvidias turn.
“Knowing” what “your” equipment is doing is fundamentally important to everyone.
In other words, they're making an ASIC. Great, but I would have expected that most of these computer vision-based self-driving platforms would indeed be using ASICs. So the news here is probably that they are moving away from NVidia's turnkey hardware solution.
Imagine the networks topology needs to be changed. Oh no's!
Ups.
Musk was involved in the attempts to purchase his first rocket. He wasn't involved in the design, because he's not qualified.
Very practical question regarding Tesla handling (never drove a Level 2+ car before, only my parents' car with FCAS and LDWS, and various similar system on all the various fleets of car-sharing).
Normally, a Telsa should handle both steering and accelerating/braking.
I the driver where to grab the wheel to adjust steering - BUT keeps the feet hovering above the pedals without pushing them (Yet) - would this disengage only the autopilot steering ? While keeping engage the usual distance keeping / emergency autonomous braking features (FCAS) ?
Or can the driver only fall back to fully manual ? (Absolutely needs to also take control of the pedals)
Also, how good are the FCAS in city ? (Im so much happy with my parent's car that I currently mostly drive on automatic distance keeping in city too. Most of the time, I'm inputing speed directly through the FCAS interface, and only use pedal in a few emergency situations that the FCAS misses.)
On one hand, given all the boasts Tesla makes, I would suspect that their system should compare well to system like Volvo's.
On the other hand, Tesla is an US company (a country where to us it seems people spend a couple hours on the highway every single day), and might only work reliably on highway. Also all the in city performance of FCAS I've seen relies on some sort of LIDAR (Volvos have a forward laser field in addition of cameras for in city mid-speed short distance obstacles, all VW have the same laser for in city safety even on the lower Up! models, etc.) which Tesla completely lacks.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
There's two components on an autonomous system.
The sensors and the computer.
Tesla has publicly stated that they've on purpose designed the computer to be modular, and the current cars since recently (forgot the exact date, it's google-able, I think it's since the introduction of the triple front cam) are designed in such way that you could swap the computer with a newer one in the future.
So in theory yes, if they keep their word, you should be able to install the newer computer with the better NN when it's out in a couple of years. (Or get all the torches and pitchforks and be angry at them not keeping their word).
This makes a lot of sense : computers (thanks to Moore's law) are progressing and getting obsoleted much faster than cars. It would be stupid to be stuck for 8 years on the same computer. There are even start-ups organizing their business around this concept (e.g.: Comma.ai)
The problem is going to be the sensors. Tesla currently seems to be lagging a bit behind others : yup they might have a bit more side cameras, but they still don't have any LIDAR-like features, where most of the competition does (from high range stuff like Volvo, all the way down to the cheaper VW Up!), and they only very recently started putting multiple front-facing cameras (like lots of German and Japanese brands do) (Currently, Tesla's triple cams are setup to each be optimal at a different range/field of view, but in a pinch they might help some amount of stereoscopy to better discriminate 3D. A better computer would definitely help for that).
These things are depending on the model of the car, AND CANNOT be retro fitted (according to the same announcement from Tesla - e.g.: they can't retro fit the triple cams to current owners).
So in short, in a couple of years, you could pay to get a better computer in your car, but you'll be lacking all the 10 new additional sensors that Tesla would be launching at that time with an announcement "This time we promise, with all these the car would be finally ready for Level 5 at some point in the future (like every other time we added sensors)" .
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Do you work in marketing?
So many words, so little content
This smells like an unholy combination of two things: a development team getting burnt by premature optimization, with just a hint of "painting oneself into a corner".
Well, on the other hand, It's just number-crunching hardware, used to run their NN.
Maybe they'll come up with a good computer, maybe in the future they'll realise that silicon by Nvidia or AMD ends up being the best to run their nets.
It's more important to them (they'll be divesting money into that R&D) than to users (it's just number-crunching silicon to run a NN on it. You're supposed to be able to swap the computer for a better one in the future, according to Tesla).
Between this and the omission of lidar, I'm not enthusiastic about Tesla's self-driving capability.
I totally agree with this. At a time when every body else relies on one (from high range Vovlo which add a laser field to their camera, down to the cheapest VW Up! that also has the laser as a standard for City Safety), I'm also uncomfortable with Tesla still skipping it.
On the other hand they seem to be slowly learning from their mistake. They started making big promises with a single front facing camera, but now have switched to a triple cam (currently mostly setup so that each cam has a different range/field of view compromise, but in a pinch could also be leveraged for some stereoscopy like most of the German and Japanese brands do).
They seem that each each they'll announce a "This time we got all the sensors, for good !" new platform with even more sensors.
They'll probably add a LIDAR down the line in a few years (probably at the time when the recent Caterpillar buyout eventually leads to their hoped smart-phone-with-a-LIDAR), and then probably redesign the LIDAR inhouse for the next iteration after that.
On the third hand : Tesla has specifically stated that their sensors are NOT retro-fitable (They can't retrofit the current triple cam), only the computers are.
We should probably wait at least a couple of years, until Tesla offer a amount of sensors we're comfortable with, before being able to consider it.
As you said other manufacturer who are longer at this game than Tesla would probably have beaten them to it by then. (But will probably only offer smaller batteries with shorter ranger than whatever range-monster Tesla would have on the marker, or still be stuck with hybrid tech).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
The reasoning is because Apple did it?
Apple has to fit its chip into a 4 oz container slightly larger than a credit card. You've got an entire f'in car, put 100 chips in it genius.
//TODO: Insert catchy phrase
If your trip has more than two stops you may get to see places you've never seen before - the "hidden" stops; hopefully there is a supercharger along that path.
Taking calculations from general purpose processing to custom hardware is going to obviously increase performance dramatically. The other benefit is energy efficiency. Once you make efficient use of the silicon, power loss will dramatically drop for the same CPU output.
Imagine rendering your favorite games with GPUs instead of CPUs. You would need ALOT of CPUs and each one would be used way less efficiently than a GPU.
This will probably increase driving ranges and give better self driving performance. It's a win win.
And yes, in 10 years, they'll probably go back to general use computing because it will be cheaper to do so.
Caveat: These custom chips could probably have specialized use cases so they'll probably sell the hardware to various consumers at a nice profit margin.
How's sensor fusion in extraction of needed information, as well as dealing with noise in general working out?
Tesla currently uses the nVidia PX2. They could upgrade to the PX Pegasus or autochauffeur and get more than a 10x improvement.
That isn't going to somehow magically make the laws of mathematics that prevent autonomous tech from working evaporate, and it's pretty obvious who they're aping. Could the announcement be timed to piggy back on Apple's recent market cap news? It's possible. It's funny, too: Apple got no end of shit (including predictions of their end) in the 90s for using proprietary chips (RISC, though they didn't manufacture them themselves). Musk is pathetic. It's a shame, too. If he weren't so delusional and narcissistic, he could actually use his resources to make some cool, actually functional stuff.
I thought it was premature ejaculation,?
Why does he think he can start up a new business like that and eclipse Nvidia that has decades of experience? He's been the dog that caught the car. The dog can't keep doing that.
Could be the first (or second... or so on) nail in this coffin.