Tuesday, April 23rd 2019
NVIDIA Responds to Tesla's In-house Full Self-driving Hardware Development
Tesla held an investor panel in the USA yesterday (April 22) with the entire event, focusing on autonomous vehicles, also streamed on YouTube (replay here). There were many things promised in the course of the event, many of which are outside the scope of this website, but the announcement of Tesla's first full self-driving hardware module made the news in more ways than one as reported right here on TechPowerUp. We had noted how Tesla had traditionally relied on NVIDIA (and then Intel) microcontroller units, as well as NVIDIA self-driving modules in the past, but the new in-house built module had stepped away from the green camp in favor of more control over the feature set.
NVIDIA was quick to respond to this, saying Tesla was incorrect in their comparisons, in that the NVIDIA Drive Xavier at 21 TOPS was not the right comparison, and rather it should have been against NVIDIA's own full self-driving hardware the Drive AGX Pegasus capable of 320 TOPS. Oh, and NVIDIA also claimed Tesla erroneously reported Drive Xavier's performance was 21 TOPS instead of 30 TOPS. It is interesting how one company was quick to recognize itself as the unmarked competition, especially at a time when Intel, via their Mobileye division, have also given them a hard time recently. Perhaps this is a sign of things to come in that self-driving cars, and AI computing in general, is getting too big a market to be left to third-party manufacturing, with larger companies opting for in-house hardware itself. This move does hurt NVIDIA's focus in this field, as market speculation is ongoing that they may end up losing other customers following Tesla's departure.
Source:
NVIDIA Press Statement
NVIDIA was quick to respond to this, saying Tesla was incorrect in their comparisons, in that the NVIDIA Drive Xavier at 21 TOPS was not the right comparison, and rather it should have been against NVIDIA's own full self-driving hardware the Drive AGX Pegasus capable of 320 TOPS. Oh, and NVIDIA also claimed Tesla erroneously reported Drive Xavier's performance was 21 TOPS instead of 30 TOPS. It is interesting how one company was quick to recognize itself as the unmarked competition, especially at a time when Intel, via their Mobileye division, have also given them a hard time recently. Perhaps this is a sign of things to come in that self-driving cars, and AI computing in general, is getting too big a market to be left to third-party manufacturing, with larger companies opting for in-house hardware itself. This move does hurt NVIDIA's focus in this field, as market speculation is ongoing that they may end up losing other customers following Tesla's departure.
38 Comments on NVIDIA Responds to Tesla's In-house Full Self-driving Hardware Development
Nvidia wants to compare to something the Tesla cars never had. If Nvidia wants that comparison they should reveal how much they were selling their previous solution to Tesla and how much the Xavier solution would have cost. There is a reason why it was cost effective to produce it in-house. Instead they want the comparison to be on the Drive AGX Pegasus that consumes 500w to a 72w Tesla solution.
Must be some real good reasons for Tesla to spend that kind of cash, time, and resources to do this themselves on top of everything else they have to do. Doesn't look good at all for nvidia.
Just like its nice warm and cosy to be a customer of nVidia.
What a lovely company they are.
Tesla is on an aggressive campaign to get market share. Their cars are pretty cheap for what they offer. But it has to mean some savings.
Both Volvo and Tesla sell premium hatchbacks for around $40k (Volvo V40 and Tesla 3). Volvo makes enough margin to afford Nvidia AGX and Tesla doesn't. Simple stuff.
Also, Tesla is not making a profit at the moment. It might simply be that shareholders demanded Musk to cut production costs.The problem is that AGX Pegasus is a chip of the future. It's designed to cover level 5, which is many years away. However, thanks to this chip car makers can develop and test cars.
Until level 5 becomes a thing, power consumption of these systems will be greatly reduced.
Also, lets not demonize these 500W. It's a car. It draws 15kW during steady 50km/h (30mph) driving.
Air-conditioning in a cheap hatchback draws 2000-3000W. AC in a premium sedan can hit over 10kW.
Moreover, 500W is peak power - used when there are a lot of signals to analyze (junctions, large traffic etc).
Going straight on a half-empty highway, it'll draw maybe 10%. We know this for sure because <50W chips (like Nvidia's basic Xavier) can already do that.Because AMD is famous for their FPGAs? :-)
This was never about Pascal gpus or mining. Consumer GPU market is highly predictable, yes even despite mining adventures and even despite Turing's lukewarm reception; there simply is not much of an alternative and the demand still exists. The reason Nvidia's stock soared was because they were looking at several new products and markets, and these are now pretty much locked for them, for the foreseeable future. Their smaller attempts like SHIELD are also not very big sales cannons.
Don't underestimate Elon Musks' power and influence in the automotive business right now. Almost every other car conglomerate has major problems getting their products to market, or even off the assembly line. Audi was and still is struggling for batteries (!). VW has only just gotten serious about developing an actual platform for E-vehicles. In a sense, that is the largest car company in the world forced to admit Elon was right about electric cars. For Audi, its a confirmation that Elon was right about keeping most of his production in-house instead of sourcing everything from everywhere.
Where is this going? Well, when Musk says we're going to do best developing true autonomous vehicles in-house, you can rest assured that is what it'll be, at least for the next ten years. He already pulled out a rabbit by announcing the entire fleet was already self-learning as it is. The man is sitting on the largest self-driving dataset in the world. Meanwhile, everyone else, including Nvidia, was still on the old-boys network style of thinking, making their solution fit in an expensive ecosystem of suppliers.
for Tesla's sake, I hope they know what they are doing.
Hope that others will do the same as Tesla. /happy.
Price point doesn't indicate the cars actual value or if it's a premium brand/product. That of course is what you consider premium. Same goes for technology. I'd rather focus on a specific product not the brand cause this vary greatly.
Ngreedia have a long track history of undercutting features. HDR performance losses on 1xxx series cards for example, when AMD had almost none. Nvidia was forcing lower quality through drivers that HDR rendering exposed. Tessellation in games they backed was being forced on to levels that provided no tangible quality (concrete barrier, water under ground that couldn't be seen), but was there to lower performance when run on competition. Physx that featured non real time precooked effects, that of course was listed as real time and was only there to slow down competitive hardware, then removing the ability to run Phsyx on a Nvidia card if a competitive card was also present.
Working in machine automation professionally gives me knowledge into what issues are faced when a vendor doesn't give almost complete diagnostic access to the system.
The 100W budget makes no sense. The car still uses 15kW just for slow cruising.
Stronger AI chip would give them great possibilities and advantages. This frugal one will give them maybe 5km of range on top of 350km Tesla 3 already has. Awesome.
Tesla S comes with a 200W sound system by default, but they let you upgrade to 560W. Think about that.
Yet, a fundamental part for an autonomous car will be limited to 100W. Good job. And great message for all the people concerned about autonomous cars' safety.
I really doubt power is a problem. Tesla has plenty of it. I think they simply can't afford a more powerful solution (neither Nvidia's nor own).I have and I know the prices. So?They decided to save money on a fundamental component for an AI car. They could save money on countless other parts.
For example they could use cheaper interior materials. At the moment it's much better than what most Americans are used to, but far from Mercedes. There is room to cut costs.
It's a cuthroat market.Which goddamn press release stated that was the reason, why are you people so freaking easy at making shit up?!!??
Musk, no less, told forbes Tesla's own chip was 10 times faster.
That's as direct a statement about PERFORMANCE of the chip as it gets, where did the crap about "more features" come from again??!!?
News here is that even a low volume car manufacturer (dozens of thousands, perhaps 1-2 hundred thousand annually at this point) cann tell Huang to go have kamasutra with himself and roll out own chip and use it instead.
Quite surprising.