The split was caused by Tesla pushing targets that Mobileye was not comfortable with. I don't think we know enough about the situation without a report from someone inside the company but it's a little extreme to call them Tesla's table scraps.
Autopilot is floundering in the absence of Mobileye. It will take Tesla a while before their much more expensive hardware suite with 8 cameras and a Px drive chip (just the chip is worth a few thousand dollars) can match the performance of Mobileye's EyeQ3 and single camera.
To be fair, Tesla is only using a single camera for their current functionality with the other cameras remaining idle. I agree though that they're not up to AP1 standards yet (even though they stated they would be by end of December).
Tesla would have been stuck at level 2/3 with Mobileye. With PX2 you have enough compute power to do much more, and a mature and rapidly developing GPGPU ecosystem (CUDA for ML, SLAM, SfM etc). I do think Nvidia is going to have to bring the price of PX2 down significantly though if they want mass market adoption.
It seems Tesla and Mobileye were at one time planning on a bright future together, because Tesla didn't have any kind of a back-up plan.
But Mobileye isn't sitting still, so if the relationship hadn't been terminated, Tesla would've been stuck with whatever Mobileye's latest and greatest offering might be.
What's interesting is that if you compare the HW2 sensor suite to Mobileye's next-gen sensor plans, they are very similar, almost the same. They were both working on something together before the relationship went sour.
Musk hinted recently that 1 PX2 is not enough, and that 2 may be necessary for L4. There's certainly no consensus as to how much compute is needed, the more the better, but I'm skeptical that the hardware costs will be low enough near term to justify putting it in privately owned autonomous cars. I think Tesla has made costly strategic errors in the Autonomy department, and that Musk in particular has repeatedly underestimated the challenge of developing a fully autonomous vehicle. I don't think he's making that mistake anymore.
However I'd wager to say that betting on a powerful Nvidia GPU/chipset (piggybacking on general ML/CUDA ecosystem progress) to be able process rapidly advancing autonomous driving algorithms is a more future proof bet for Tesla than relying on already outdated hardware/algorithms from Mobileye.
I thought $16B for yet another autonomous car developer R&D outfit seemed excessive even by Intel's standards, but it looks as if Mobileye has been providing assistant/enhancing technologies (eg collision warning systems) for more than a decade.
This is probably not too bad a spot to be in, at least until fully autonomous vehicles have become mainstream. They seem to be profitable, at least.
I didn't think they were really into self-driving back then. There is a whole range of ADAS capabilities that fall short of taking control of a vehicle.
$16B is a great price for Mobileye's team, but I am not sure it is the best move for Intel. But I guess Intel will try to sell package deals to car markers and thus use this to try to keep NVIDIA and ARM from locking up the in-car market.
Thus it is a strategic move on Intel's part to fend off competitors rather than reflective of the intrinsic value of Mobileye.
BTW what was Mobileye's revenue last year and profit margin? About $350M in 2016, with probably $500M in 2017 it seems. So the multiple is around 30x revenue or so?
Intrinsic value usually refers to the present (discounted) value of future earnings.
>There is price and price ($16B) is determined by the market.
How is it determined if all valuation is supposed to be pointless?
>Talking about PE and multiples is pointless for tech companies where synergies and rapid adoption/changes are common
For a public company that has existed for almost two decades it absolutely makes sense to consider revenue multiples among other factors.
It may well be that the value of the company in Intel's hands is worth more than it is worth for other, non-strategic, investors. I agree that there is a lot of potential variability in tech, but that doesn't mean any fantasy price is right just because some people buy some shares at that price (which is what market prices mean)
Worst case in a few years they can write down 16 billion or some amount and they can lower their taxes. Effectively the government will pay for part of the purchase.
More like tax credits. For example you can buy an EV and knock some money off your taxes. For example supposed a company makes 5 billion they have a 30% tax rate. They would owe 1.5 billion. Supposed they bought a company for 6 billion. They realized next year hey the company is really only worth $1 billion. They write off the 5 billion which offsets the 5 billion profits so instead of the paying 1.5 billion in taxes they pay 0. So government basically ended up paying 25% of the purchase price.
Lot of consolidation in Autonomous Driving(AD) market. Samsung bought Harman, Google buying waze etc. As Continental CEO puts it "Traditional OEM/suppliers want to go from inside to outside, traditional IT companies are moving from outside to inside" ..a car basically. To track AD market this could be of interest to you https://www.2025ad.com/
Worth mentioning their factories in Israel are side by side and they already work together on projects in Jerusalem. A lot of Mobileye workers are ex-intel and sometimes vice versa.
The companies are a lot more linked than it seems.
There is urgency in the market around autonomous vehicles - for better or worse, it is believed that it will be the next big technology market - but I think that the Nvidia push around it is the main thing spurring Intel.
My wife has Mobileye in her car - it works quite well - warning you when you're too close to the car in front or when you're wandering out of your lane.
The collision avoidance systems (I presume OP is speaking about) is aftermarket. That's a small part of Mobileye's business, but it is based on the same technology.
BMWs and other high end cars have had Mobileye as OEM for a long time, but the company itself is a bit of a legacy company and not always run well (and is based in Jerusalem which is even more legacy) - but like many Israeli tech stories has great tech.
I've noticed a bunch of these in Uber/Lyft cars in NYC, but didn't know what they were until I saw TechCrunch linked to a Mobileye press release about installing 4500 of these systems:
I got one with my car (a Hyundai) recently. Works rather well, although has some false positives once in a while, where it screams that I'm about to hit someone while there's no one near me. Just think about the scale and quality of the data they collect like that!
I'm not sure, it came pre-installed with the car. It warns me when I'm drifting between lanes without signaling first, it tells me if I'm speeding (no warning sound though), it indicates if I'm within a safe distance of the car ahead of me, and it triggers a very loud alarm if it thinks I'm about to hit someone or not keeping my distance.
It actually probably saved me from a minor accident on my first drive in my brand new car, getting it out of the dealership: It was raining heavily and visibility was crap, and I was a bit distracted finding all the controls and looking for the back window wipers (turns out there aren't any!). And I didn't notice the cars ahead of me stopped. The alarm went off and I hit the brakes (it was very low speed), and everything was fine.
Intel has made acquisitions of almost $40 billion in the past couple of years. It will be interesting to see if they can make this money back as competition becomes more fierce in the PC and server markets because of AMD and Qualcomm.
They have 95%+ marketshare in server chips. They have an enormous manufacturing advantage. Even though they charge so much for their chips if they feel threatened by Ryzen they can just lower the price of their chips to Ryzen prices because they have a manufacturing advantage. It will just hurt their margins. 40-50% margins forever seem pretty absurd from a pure economics perspective.
They didn't miss mobile. Mobile makes poor margins at least for intel so they are not in it. PC sales are already down. Where they make a ton of their money is data centers.
Partially because they felt they could gain some advantage by owning the client side of what is powering a lot of their data center chips. Investing in modem tech and trying to get into mobile phones, they could have a completely integrated story from the phone, modem, cell tower (see Axxia acquisition), to the data center. I think they also were hoping the Intel Security story would be compelling from an end-2-end perspective.
Based on Microsoft's recent decision and some other rumors ARM on server might reduce Intel to only 90+% over the next few years but that's nearly the same cashflow.
Currently it looks like the manufacturing advantage will run out in the next ten to twenty years. There isn't very much innovation in the chip manufacturing business. Intel managed to stay ahead of the curve with their massive R&D budget, but we're well into the diminishing returns territory. Eventually all other players will catch up, unless some breakthrough technology enters the market first.
> if they feel threatened by Ryzen they can just lower the price of their chips to Ryzen prices because they have a manufacturing advantage.
If it would've been so easy, they would've done it with Atom in the mobile market. But it wasn't. Intel was losing $1 billion a quarter in mobile just to be competitive on price.
Intel absolutely lacks the cost structure to compete with Qualcomm's ARM chips, and that has already been proven. It's not a theory.
As for Ryzen, they can lower prices, but not by as much as you may think. Lowering prices is directly proportional with how many people Intel will have to fire to accommodate for a slightly lower cost structure that's more competitive with AMD.
For every 10% average reduction in price, Intel may have to fire at least a few thousand employees (across divisions, but more likely by gutting some non-profitable research projects, in the beginning).
However, those layoffs will also look bad on Intel's public image, so don't expect Intel to lower prices strategically across product lines to compete against AMD without serious consideration. But I really don't expect them to lower their average price point by more than 10% for the next few years, until Intel comes out with a more "streamlined" (fewer features) microarchitecture, that's more cost-effective to build.
I can only assume that Intel will axe some of the C-Suite and attempt to rekindle the flame with Tesla. Mobileye was functioning a bit too much like a traditional (and glacial) vendor in the automotive industry. Tesla wanted to develop and implemented quickly, while Mobileye became a bit too focus on the what-ifs.
The truth is, Tesla vehicles are safer than their counterparts in almost every situation imaginable. Mobileye screwed the pooch when they began issuing negative press releases that attempted to push all the responsibility and potential liability onto Tesla.
Intel is smarter than that. Intel has become fairly accustomed to implementing and following through with aggressive development and launch cycles.
What's Intels play here?
I'm glad the mobileye team get some nice purim bonus but we'll see in couple of years if Intel won't kill the product (and the team)...
Diversification. They see the writing on the wall for their CPU business and it's associated margins. They need other high-margin highly in demand tech products to go after.
Furthermore, I'm guessing x86 has almost no representation in the self-driving car space at the moment. They're probably hoping to change that.
>> Diversification. They see the writing on the wall for their CPU business and it's associated margins. They need other high-margin highly in demand tech products to go after.
I completely agree.
>> Furthermore, I'm guessing x86 has almost no representation in the self-driving car space at the moment. They're probably hoping to change that.
But this, not so much. x86 is way too power hungry (especially for EVs) and way too expensive for automotive use.
Why do we still have this misnomer. The front end which decodes the x86 instructions is a small part of the overall chip. Intel made low power chips that were on phones using x86. They just had no competitive advantage in that space and left.
> But this, not so much. x86 is way too power hungry (especially for EVs) and way too expensive for automotive use.
it isn't a power or efficiency problem. it's a margin problem. vehicle electronics have margins that consumer devices don't. just look at the price of 95% useless in-car satnav.
> But this, not so much. x86 is way too power hungry (especially for EVs) and way too expensive for automotive use.
When you're running a electric motor that pulls power on the magnitude of 1 * 10^5 W (i.e. 100 kW+) I doubt you care about a few extra watts total power load if you have an inefficient processor. Similarly for conventional fuel, an extra few watts on the load is hardly going to worry the alternator.
OK. I've worked on several EVs with 100kW motors. Lets do some math. A really good EV takes around 200 to 250 watt-hours per mile. On the EPA urban drive cycle you average just under 20 miles per hour. If you have 100 miles range, that will take around 5 hours in urban driving. A 40 watt load for 5 hours is 200 watt-hours. Your Intel processor running near 100 watts is going to cost you 2.5 percent of the range. Or if you look at costs, it's going to take that percentage of the battery cost to run it.
Sure an EV can suck 100kW or more of power, but only for a short time. And it can be put back into the battery via regenerative baking. Range is incredibly sensitive to electrical loads.
The power used by a processor is mostly a matter of performance rather than ISA. You might be able to make a non-x86 processor with similar performance that only takes 95 rather than 100 Watts but that's the extent of it.
The reason that ARM processors are able to take so much less power than Intel is because their single threaded performance is much lower. As single threaded performance increases linearly the power required goes up as the square of that for architectural improvements and as the cube for frequency improvements modulo FO4 delays.
Intel's Atom processors haven't gotten huge amounts of love from the parent company but for DSPish code of the sort that cars are mostly running they should perform comparably to the equivalent ARM options.
You're stuck in the mindset that the consumer CPUs that you can purchase are the only ones Intel can make. They can and do make custom chips, extremely custom, for large customers and government agencies. It would be very, very easy for them to mass produce a chip that drops all the unnecessary features for an "appliance" like self-driving cars.
Let's look at it a different way. The MacBook battery is 55 watt-hours, the Tesla battery is 100 thousand watt hours. They both run their respective workloads for a similar amount of time. So, less than a tenth of a percent?
Why would you assume they'd be using a CPU running at 100W???
I can get an off-the-shelf Xeon-D 8-core processor that runs at 35W and any single core would STOMP every ARM chip on the market. If I were to build that chip specifically for a car platform my power envelope would drop accordingly.
This is an embedded application... not a gaming computer. If I move to Atom - which is what would likely be used for this sort of thing, you're talking less than 5W.
>> This is an embedded application... not a gaming computer. If I move to Atom - which is what would likely be used for this sort of thing, you're talking less than 5W.
Correct. But Atom processors performance are no better than the competition and don't currently have any of the common automotive peripherals (CAN for example).
Intel is the device manufacturer... it doesn't matter if they're "better". They get to choose what they put into their own hardware, this isn't an open audition.
Regardless, you're missing the point. Intel has the means to get into the space and create appropriate CPUs to compete. This acquisition makes it clear they're planning on doing just that.
X86 has some inherent disadvantages to arm and other newer risc architectures. The main one being that modern x86 is actually a risc architecture with a x86 to risc conversion layer. Pure risc architectures like arm do not pay this tax. There are other reasons as well. For example, the lack of many registers in the isa exacerbates the register renaming cost. Jim Keller estimates that fully developed arm chips would be ~15-30% more efficient than their x86 counterparts.
well advances in self driving tech will have payouts in all robotics. any robot needing to operate in an environment which isn't 100% predictable will improve.
simplest way, think of robots used to automate cleaning everything from the mall, your house, to the sides of the roads. or one day pulling up to a fueling facility and it automatically connects to your car for gas/electric/etc all is improved by better sensors and such
Yeah, they're old school. They've got all sorts of deep industry partnerships and various driver assist and safety products. What they don't have is a Level 4 autonomous driving OS. Nobody does, it doesn't exist yet.
The whole 3 ring circus of chipmakers, OEM's, and various component suppliers that stand to benefit from Autonomous vehicles is held aloft by a single tentpole that doesn't actually exist yet. It's just a twinkle in the eye of some AI genius who may very well still be an undergrad at the moment. Intel just spent 15 gigabucks assuaging their fear of missing out.
The field of AI and deep learning in particular is changing and progressing so fast that the experts of yesteryear are almost certainly not the people who will build the autonomous OS of the future. There will be a lot of misspent capital between now and the first safe, reliable, market ready, publicly accessible fully autonomous vehicle.
It don't believe we're headed into any kind of AI trough. Very exciting things are happening in the field of deep learning, and the people who are best at understanding and applying all that new theory and research are kids who never leave the lab.
Drive.ai is an autonomous driving startup that's made up of mostly unknown Stanford AI lab labmates who dropped out of their PHD programs to build an autonomous OS, and what they're doing is so much more futuristic than what anybody else is doing. They've made more progress in less time with fewer resources than any of the big players. They've got 4 test vehicles, a team of 40 people, they're running on $12 million in venture capital, and they've demoed capabilities that the multi-billion dollar players don't have. They're just pure deep learning guys, they could have gone into fintech or NLP but they went for AVs starting with a blank slate and an abundance of unproven talent.
They're still behind Waymo and Cruise, but they're progressing at a higher rate of speed. Everybody in this game is subject to disruption from some nobody coming out of nowhere. Something like transfer learning drops out of the sky, it has the potential to totally revolutionize the way one might think about an Autonomous driving stack, and it all comes down to the genius who figures out how to apply it effectively. That genius could be anyone, no amount of money you spend can buy certainty in this race. Which makes it really fun to watch.
I'll just be upfront, I'm a total pessimist on the near-term possibility of autonomous driving.
>I don't believe we're headed into any kind of AI trough.
Why? AI winter is a well-known phenomenon, AI hype has permeated tech, now business and is starting to push into the layman's consciousness. Does it not seem that a bust is near? The laymans perception of autonomous driving is a magic box that will drive them around while they read their phones and that it will be here in 5/10 years. I think that's just unsustainable. A car which can handle northern snow, midwestern winds, country potholes, city traffic, unpredictiable/aggressive drivers and which can do that with a lower accident rate than humans is our generations cold fusion. Fundamental breakthroughs are needed.
What strikes me is that we think because we can describe "driving" in one word, we think it is a specific task, but it seems much more general and AI developments on general learning are notoriously fraught. Your thoughts?
I don't know what people mean when they say 'the near-term possibility of autonomous driving'. Some people mean a car that can go everywhere and do everything a human can. I'm thinking about robotaxis in the Bay area that drive along established circuits and do pick-ups and drop offs in designated zones like bus stops. That's what I'm expecting to see in the 5 year time frame, and being able to apply autonomous driving technology to that limited extent opens a very lucrative new market.
Any way you look at the autonomous driving problem it's a hard problem, but every few days it seems there's some paper coming out that demos a new optimization to one aspect or another of the autonomous vehicle development pipeline.
The world of modular deep learning systems, where you have various parent and child subsystems that can be quickly trained and swapped out is very much an open and field of research and my bet is on agile teams that can build and test models faster, and can take things from research to implementation in less time than the competition.
Drive.ai's best trade secret is the system they have for vastly reducing the amount of training data they need to get a result, and vastly reducing the amount of human labour needed to annotate or clean up said training data.
I don't know what the hype is. People are expecting terminator robots or something, that's not going to happen anytime soon. But progress in Machine learning has been steady and driving for a decade and shows know signs of letting up in the forseeable future, because we just keep opening up new frontiers to explore. There's so much to do.
I expect a car that can get me to and from work safely and reliably so i don't have to waste an hour a day driving - and i hear i'm lucky that it's only an hour.
What about a a car that can only do level-4 on the highway, but that can very reliably sense when it's near an urban area, and tells the driver to take control ? maybe we're close to that ?
kinda funny since we was very anti mobileye and said they didn't know what they were doing and claimed he was gonna be the ceo of a billion dollar company etc and then the gov shuts him down immediately and mobileye gets acquired for 16B.
dude needs to get back to hacking iphones it seems.
Their plan is probably to build machine learning/autonomous driving chips combining Intel's own chips/expertise, FPGA (as with their Altera acquisition) and specialist ML chips (like Movidius that they bought recently).
$16B for Mobilieye as a route to market (they already have the relationships and supply chains set up with car manufacturing) sounds like a lot but Intel needs a fast track if they want to compete with Nvidia (and ARM and Qualcomm) in that space.
Also of note is that MobileEye is the company that built the sensor and software suite for Telsa's Autopilot generation 1 system. Tesla threw it away due to disagreements and Tesla now uses Nvidia Drive PX 2 (a customized version no doubt) for the number crunching and their own camera suite.
If you've got a good enough sense of your algorithms and have a large enough scale to be rolling your own ASIC components you probably don't need an FPGA in you chip.
I have to wonder what sort of market size projections support that kind of price for what is essentially an accessory technology. There's a lot of fanciful fantasies about self driving cars floating around, but I'm not familiar with any of them that would support the idea of a growth market compared to today's dumbcar market. Even today, you can buy entire car companies for that kind of money.
104 comments
[ 4.1 ms ] story [ 221 ms ] thread"It's been a bear to get done, but prob 10 days or so, depending on full speed autosteer test results" https://twitter.com/elonmusk/status/841041891747545089
But Mobileye isn't sitting still, so if the relationship hadn't been terminated, Tesla would've been stuck with whatever Mobileye's latest and greatest offering might be.
What's interesting is that if you compare the HW2 sensor suite to Mobileye's next-gen sensor plans, they are very similar, almost the same. They were both working on something together before the relationship went sour.
Musk hinted recently that 1 PX2 is not enough, and that 2 may be necessary for L4. There's certainly no consensus as to how much compute is needed, the more the better, but I'm skeptical that the hardware costs will be low enough near term to justify putting it in privately owned autonomous cars. I think Tesla has made costly strategic errors in the Autonomy department, and that Musk in particular has repeatedly underestimated the challenge of developing a fully autonomous vehicle. I don't think he's making that mistake anymore.
You couldn't train a good autonomous driving agent even if you used a massive GPU cluster.
I'm not saying better hardware doesn't help, it definitely helps incrementally but it's not going to be enough to solve this problem.
However I'd wager to say that betting on a powerful Nvidia GPU/chipset (piggybacking on general ML/CUDA ecosystem progress) to be able process rapidly advancing autonomous driving algorithms is a more future proof bet for Tesla than relying on already outdated hardware/algorithms from Mobileye.
This is probably not too bad a spot to be in, at least until fully autonomous vehicles have become mainstream. They seem to be profitable, at least.
Thus it is a strategic move on Intel's part to fend off competitors rather than reflective of the intrinsic value of Mobileye.
BTW what was Mobileye's revenue last year and profit margin? About $350M in 2016, with probably $500M in 2017 it seems. So the multiple is around 30x revenue or so?
Talking about PE and multiples is pointless for tech companies where synergies and rapid adoption/changes are common.
Intrinsic value usually refers to the present (discounted) value of future earnings.
>There is price and price ($16B) is determined by the market.
How is it determined if all valuation is supposed to be pointless?
>Talking about PE and multiples is pointless for tech companies where synergies and rapid adoption/changes are common
For a public company that has existed for almost two decades it absolutely makes sense to consider revenue multiples among other factors.
It may well be that the value of the company in Intel's hands is worth more than it is worth for other, non-strategic, investors. I agree that there is a lot of potential variability in tech, but that doesn't mean any fantasy price is right just because some people buy some shares at that price (which is what market prices mean)
"Intrinsic value" is a basic financial term that is widely used and has a well defined meaning - just type the term into Google: https://en.wikipedia.org/wiki/Intrinsic_value_(finance)
The companies are a lot more linked than it seems.
BMWs and other high end cars have had Mobileye as OEM for a long time, but the company itself is a bit of a legacy company and not always run well (and is based in Jerusalem which is even more legacy) - but like many Israeli tech stories has great tech.
http://www.mobileye.com/en-uk/products/mobileye-5-series/
http://www.mobileye.com/en-uk/products/mobileye-6-series/
http://www.prnewswire.com/news-releases/mobileye-completes-i...
It actually probably saved me from a minor accident on my first drive in my brand new car, getting it out of the dealership: It was raining heavily and visibility was crap, and I was a bit distracted finding all the controls and looking for the back window wipers (turns out there aren't any!). And I didn't notice the cars ahead of me stopped. The alarm went off and I hit the brakes (it was very low speed), and everything was fine.
If it would've been so easy, they would've done it with Atom in the mobile market. But it wasn't. Intel was losing $1 billion a quarter in mobile just to be competitive on price.
Intel absolutely lacks the cost structure to compete with Qualcomm's ARM chips, and that has already been proven. It's not a theory.
As for Ryzen, they can lower prices, but not by as much as you may think. Lowering prices is directly proportional with how many people Intel will have to fire to accommodate for a slightly lower cost structure that's more competitive with AMD.
For every 10% average reduction in price, Intel may have to fire at least a few thousand employees (across divisions, but more likely by gutting some non-profitable research projects, in the beginning).
However, those layoffs will also look bad on Intel's public image, so don't expect Intel to lower prices strategically across product lines to compete against AMD without serious consideration. But I really don't expect them to lower their average price point by more than 10% for the next few years, until Intel comes out with a more "streamlined" (fewer features) microarchitecture, that's more cost-effective to build.
The truth is, Tesla vehicles are safer than their counterparts in almost every situation imaginable. Mobileye screwed the pooch when they began issuing negative press releases that attempted to push all the responsibility and potential liability onto Tesla.
Intel is smarter than that. Intel has become fairly accustomed to implementing and following through with aggressive development and launch cycles.
Furthermore, I'm guessing x86 has almost no representation in the self-driving car space at the moment. They're probably hoping to change that.
I completely agree.
>> Furthermore, I'm guessing x86 has almost no representation in the self-driving car space at the moment. They're probably hoping to change that.
But this, not so much. x86 is way too power hungry (especially for EVs) and way too expensive for automotive use.
it isn't a power or efficiency problem. it's a margin problem. vehicle electronics have margins that consumer devices don't. just look at the price of 95% useless in-car satnav.
When you're running a electric motor that pulls power on the magnitude of 1 * 10^5 W (i.e. 100 kW+) I doubt you care about a few extra watts total power load if you have an inefficient processor. Similarly for conventional fuel, an extra few watts on the load is hardly going to worry the alternator.
Sure an EV can suck 100kW or more of power, but only for a short time. And it can be put back into the battery via regenerative baking. Range is incredibly sensitive to electrical loads.
The reason that ARM processors are able to take so much less power than Intel is because their single threaded performance is much lower. As single threaded performance increases linearly the power required goes up as the square of that for architectural improvements and as the cube for frequency improvements modulo FO4 delays.
Intel's Atom processors haven't gotten huge amounts of love from the parent company but for DSPish code of the sort that cars are mostly running they should perform comparably to the equivalent ARM options.
I can get an off-the-shelf Xeon-D 8-core processor that runs at 35W and any single core would STOMP every ARM chip on the market. If I were to build that chip specifically for a car platform my power envelope would drop accordingly.
https://ark.intel.com/products/family/87041/Intel-Xeon-Proce...
This is an embedded application... not a gaming computer. If I move to Atom - which is what would likely be used for this sort of thing, you're talking less than 5W.
https://en.wikipedia.org/wiki/List_of_Intel_Atom_microproces...
Correct. But Atom processors performance are no better than the competition and don't currently have any of the common automotive peripherals (CAN for example).
Regardless, you're missing the point. Intel has the means to get into the space and create appropriate CPUs to compete. This acquisition makes it clear they're planning on doing just that.
simplest way, think of robots used to automate cleaning everything from the mall, your house, to the sides of the roads. or one day pulling up to a fueling facility and it automatically connects to your car for gas/electric/etc all is improved by better sensors and such
The whole 3 ring circus of chipmakers, OEM's, and various component suppliers that stand to benefit from Autonomous vehicles is held aloft by a single tentpole that doesn't actually exist yet. It's just a twinkle in the eye of some AI genius who may very well still be an undergrad at the moment. Intel just spent 15 gigabucks assuaging their fear of missing out.
The field of AI and deep learning in particular is changing and progressing so fast that the experts of yesteryear are almost certainly not the people who will build the autonomous OS of the future. There will be a lot of misspent capital between now and the first safe, reliable, market ready, publicly accessible fully autonomous vehicle.
>the experts of yesteryear are almost certainly not the people who will build the autonomous OS of the future
is because it will be a very long time before we have anything like a Level 4 vehicle?
Drive.ai is an autonomous driving startup that's made up of mostly unknown Stanford AI lab labmates who dropped out of their PHD programs to build an autonomous OS, and what they're doing is so much more futuristic than what anybody else is doing. They've made more progress in less time with fewer resources than any of the big players. They've got 4 test vehicles, a team of 40 people, they're running on $12 million in venture capital, and they've demoed capabilities that the multi-billion dollar players don't have. They're just pure deep learning guys, they could have gone into fintech or NLP but they went for AVs starting with a blank slate and an abundance of unproven talent.
They're still behind Waymo and Cruise, but they're progressing at a higher rate of speed. Everybody in this game is subject to disruption from some nobody coming out of nowhere. Something like transfer learning drops out of the sky, it has the potential to totally revolutionize the way one might think about an Autonomous driving stack, and it all comes down to the genius who figures out how to apply it effectively. That genius could be anyone, no amount of money you spend can buy certainty in this race. Which makes it really fun to watch.
>I don't believe we're headed into any kind of AI trough.
Why? AI winter is a well-known phenomenon, AI hype has permeated tech, now business and is starting to push into the layman's consciousness. Does it not seem that a bust is near? The laymans perception of autonomous driving is a magic box that will drive them around while they read their phones and that it will be here in 5/10 years. I think that's just unsustainable. A car which can handle northern snow, midwestern winds, country potholes, city traffic, unpredictiable/aggressive drivers and which can do that with a lower accident rate than humans is our generations cold fusion. Fundamental breakthroughs are needed.
What strikes me is that we think because we can describe "driving" in one word, we think it is a specific task, but it seems much more general and AI developments on general learning are notoriously fraught. Your thoughts?
Any way you look at the autonomous driving problem it's a hard problem, but every few days it seems there's some paper coming out that demos a new optimization to one aspect or another of the autonomous vehicle development pipeline.
The world of modular deep learning systems, where you have various parent and child subsystems that can be quickly trained and swapped out is very much an open and field of research and my bet is on agile teams that can build and test models faster, and can take things from research to implementation in less time than the competition.
Drive.ai's best trade secret is the system they have for vastly reducing the amount of training data they need to get a result, and vastly reducing the amount of human labour needed to annotate or clean up said training data.
I don't know what the hype is. People are expecting terminator robots or something, that's not going to happen anytime soon. But progress in Machine learning has been steady and driving for a decade and shows know signs of letting up in the forseeable future, because we just keep opening up new frontiers to explore. There's so much to do.
But that is different to another AI winter. AI is making real companies real money this time.
dude needs to get back to hacking iphones it seems.
$16B for Mobilieye as a route to market (they already have the relationships and supply chains set up with car manufacturing) sounds like a lot but Intel needs a fast track if they want to compete with Nvidia (and ARM and Qualcomm) in that space.
Their intelligence services are also known to "sabotage" western citizens involved in BDS. https://electronicintifada.net/content/israels-new-strategy-...