Welcome to Tesla Motors Club
Discuss Tesla's Model S, Model 3, Model X, Model Y, Cybertruck, Roadster and More.
Register

2017 Investor Roundtable:General Discussion

This site may earn commission on affiliate links.
Status
Not open for further replies.
Why do people focus so much on this? Its a GPU, they're not expensive. It does not take supercomputer-class computing power to do the sort of computations required for self driving - any semi-modern gaming GPU could do it.
Please have a clue what about are talking about before you post.

They are not using a gpu. They are using a server with multiple CPU's and even more gpu's. Elon said that it is a super computer. It takes a lot of computing power to process the radar images. They send out a signature with every pulse that has the time, angle and direction of the pulse that they use to build an image, similar to what they could obtain with Lidar, which by comparison requires a lot less additional processing.

Advanced Sensor Coverage
Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.


They also need to process the camera images which is easier than the radar but non trivial, recognizing the difference between a cardboard box and a small child for example. Plus the ultrasonic sensors and they need to do all of that in real time!
 
Please have a clue what about are talking about before you post.

They are not using a gpu. They are using a server with multiple CPU's and even more gpu's. Elon said that it is a super computer. It takes a lot of computing power to process the radar images. They send out a signature with every pulse that has the time, angle and direction of the pulse that they use to build an image, similar to what they could obtain with Lidar, which by comparison requires a lot less additional processing.

Advanced Sensor Coverage
Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.


They also need to process the camera images which is easier than the radar but non trivial, recognizing the difference between a cardboard box and a small child for example. Plus the ultrasonic sensors and they need to do all of that in real time!

His point stands though. GPU chips aren't very expensive and the many-chip module Tesla is buying is also not tremendously expensive, probably. This community has spent a fair amount of time trying to figure out the HW cost as though that provides insight as to the vehicle's COGS. It doesn't provide that much insight since there are so many unknowns the price of the rest of the car.
 
Moran said he has been working at Tesla for 4 years. Assuming this is true, he has been at Tesla since around the time the UAW first attempted to bring Tesla into the "UAW Family".

UAW Looks to Organize Tesla | 1853 Chairman.com

Moran says he went through a lot of trouble to pass around the article, to recruit workers to join his "cause". However, Moran also implies in his article that Tesla is negligent when it comes to worker safety. The moment he stated this his entire article in my view crossed a line from being a reasonable complaint to a dubious hit piece.

UAW did not deny they were at some point or are currently supporting Moran with his agenda. UAW simply denied Moran "has been paid by UAW".
Moran stated UAW has been actively helping him.

Many things about Morgan's article don't make sense. Also, his decision to remove the comment section from his article certainly doesn't help his case.

Given the tactics UAW has used in the past, nothing would surprise me.

UAW officials take an auto worker for a ride - World Socialist Web Site
In a weird sort of way, the market would interpret interest from the UAW into tesla as a sign that tesla is now profitable... Just sayin' the market moves in strange ways...
 
Please have a clue what about are talking about before you post.

They are not using a gpu. They are using a server with multiple CPU's and even more gpu's. Elon said that it is a super computer. It takes a lot of computing power to process the radar images. They send out a signature with every pulse that has the time, angle and direction of the pulse that they use to build an image, similar to what they could obtain with Lidar, which by comparison requires a lot less additional processing.

Advanced Sensor Coverage
Eight surround cameras provide 360 degrees of visibility around the car at up to 250 meters of range. Twelve updated ultrasonic sensors complement this vision, allowing for detection of both hard and soft objects at nearly twice the distance of the prior system. A forward-facing radar with enhanced processing provides additional data about the world on a redundant wavelength that is able to see through heavy rain, fog, dust and even the car ahead.


They also need to process the camera images which is easier than the radar but non trivial, recognizing the difference between a cardboard box and a small child for example. Plus the ultrasonic sensors and they need to do all of that in real time!

Please don't assume I'm an idiot before retorting to my simplified explanations. I know you've seen me make hundreds of posts here complete with sources and a rational dissection of what I'm saying. I know you've seen me speak tangentially of my career.

I work in software development for a company that manufactures a real-time data acquisition and analysis system that runs on an RTAI Linux platform. Some of our products process and extract data from images from high-speed cameras (in excess of 4000 fps) in real-time. Comparatively speaking, car-speed real-time is SLOW.

GPU's typically have many many more processing cores per physical chip die. They are ideal for this type of application.

A GPU cannot be used without a CPU to control it, so yes, there is additional supporting hardware. The point is, that it does not take a 'trunk full of computers' to do it. A modern household PC with a decent graphics card has the chops to do this stuff, and those can be cleverly packaged to fit a space substantially smaller than a typical PC tower, and doesn't cost more than a 3 digit number of dollars.

My anecdote I relayed about the desktop GPUs being used to process data replacing the server? They were processing ground penetrating radar images for the mining industry to locate fractures in rock layers separating their mine from the water table. I understand more about this than an average Joe.
 
Last edited:
I apologize for offending you.

First picture of Tesla’s new NVIDIA onboard supercomputer for Autopilot installed in a car
NVIDIA describes the Drive PX 2 as “the world’s first AI supercomputer for self-driving cars”. Its computing power is comparable to about 150 MacBook Pros and the company estimates that one can support a level 4 self-driving system while two would be necessary for a fully self-driving level 5 vehicle, but Tesla is aiming for its software to be efficient enough to run level 5 on one.
racer26 said:
Why do people focus so much on this? Its a GPU, they're not expensive. It does not take supercomputer-class computing power to do the sort of computations required for self driving - any semi-modern gaming GPU could do it.

A GPU cannot be used without a CPU to control it, so yes, there is additional supporting hardware. The point is, that it does not take a 'trunk full of computers' to do it.
That's not what you said. Way more than "a GPU".

racer26 said:
A modern household PC with a decent graphics card has the chops to do this stuff, and those can be cleverly packaged to fit a space substantially smaller than a typical PC tower, and doesn't cost more than a 3 digit number of dollars.
Wrong t according to Elon. Modern household PC's don't come with "computing power comparable to about 150 MacBook Pros ".

I'm sure that supercomputer-class computing servers are somewhat affordable, but according to Elon that's exactly what's required.
 
Wrong t according to Elon. Modern household PC's don't come with "computing power comparable to about 150 MacBook Pros ".

I'm sure that supercomputer-class computing servers are somewhat affordable, but according to Elon that's exactly what's required

One thing that everyone MUST be cognizant of, is that "computing power" can be measured many different ways. Workloads and the benchmarks that measure a system's effectiveness at handling those workloads can be very diverse.

Most simply put, a modern GPU performs specific types operations extremely quickly through a large number of units operating in parallel. On those types of workloads, a GPU or multiple GPUs could handily defeat a MacBook Pro running the same algorithms on an Intel Skylake Core i5 or i7. However, that same GPU doesn't have the flexibility to do all the things than an Intel chip can handle.

I also want to note that MacBook Pro does not refer to a single model of computer. In the current generation, the entry level MacBook Pro has an integrated Intel graphics system, while the higher end MacBook Pros have discrete AMD GPUs. So if someone says, X computing system is 150x as powerful as a MacBook Pro on Y workload, one of the first things I'm going to ask is: which MacBook Pro from what year?
 
I apologize for offending you.
'Offend' isn't really the right word. I have a pretty thick skin, but I certainly wasn't going to allow my intelligence and knowledge of the topic to be maligned.
I'm aware of what they said. I'm also aware that getting people to understand complex abstract topics like processing power requires hyperbole sometimes. GPUs are capable of very impressive feats of computing power for very low cost when they're crunching numbers and data instead of drawing pretty pictures. Laptops have notoriously bad GPUs because they aren't typically used for gaming. I assume '150 MacBook Pros' is a hyperbolic poor comparison that's only sort of correct.

That's not what you said. Way more than "a GPU".
The OP I was replying to asked about the cost of the Titan processor Tesla is using (which underpins DrivePX2, or at least Tesla's system that is built with nVidia tech). It *is* a GPU. Its attached to some supporting hardware, yes, but they didn't ask about the supporting hardware, which isn't that much more complex than any of the hardware driving competitors products anyway, so I didn't worry about it. You can cram it all up under the dash somewhere with no trouble at all, and it doesn't cost much more than a modest PC. All in cost for all of the computing hardware is likely in the realm of $500-800.

Wrong t according to Elon. Modern household PC's don't come with "computing power comparable to about 150 MacBook Pros ".

I'm sure that supercomputer-class computing servers are somewhat affordable, but according to Elon that's exactly what's required.
Modern household PCs with a decent graphics card do have sufficient power. The Linux system I mentioned in my post runs on ~6-8 year old quad-core desktop-class processors with a pricetag < $200 at the time, and 4GB of RAM. It processes live streaming video at over 4000 fps, and does real-time PID-control of the computed output by controlling hardware attached to the thing in the image. If we needed more power, we could use a graphics card. A desktop PC version of the Titan GPU that Tesla is using is running for about $1000 in quantity 1 right now. It has 3072 CUDA cores and 12 GB of RAM. I suspect its probably this that is underpinning whatever is under the hood of AP2, but repackaged onto a single PCB with its host CPU and other supporting hardware.

Supercomputer class servers have pricetags in the 5 digits range. I guarantee you that Tesla doesn't require anything like that.
 
Why do people focus so much on this? Its a GPU, they're not expensive. It does not take supercomputer-class computing power to do the sort of computations required for self driving - any semi-modern gaming GPU could do it.

Its actually quite amazing what can be done with cheap GPUs. A former contracted coworker worked on an outside project for which the customer originally bought a 24 core IBM server w/ 72GB of RAM for around $50,000 to crunch numbers with. They later achieved better performance with 2 nVidia ~$600 desktop class graphics cards in a SLI configuration doing the crunching in an ordinary desktop PC and literally gave him the $50,000 server, because they were no longer using it. That server is now the corporate server at my workplace.

The pricing quoted by nVidia for their DrivePX2 platform was for development kits, not actual production quantity equipment.

I imagine the nVidia chip driving AP2 is no more than $200 in the quantities Tesla is buying it.
Hey I'm just curious, no need to get worked up on this. Some NVidia high end graphics chips can go close to $1K, which is not insignificant, similar to what GM spends on advertising for Cruze. If you have the info that it's indeed $200 then that's great. I've got lots of eggs all in TSLA basket because I trust Elon.
 
  • Informative
Reactions: neroden
Thank you for everyone who got to the bottom of the $17-21 hour question. I hope Tesla does not unionize, as I think it would kill productivity, but I also hope the compensation is somehow better than those numbers imply (Maybe stock options?). Hopefully as they continue to automate production and reduce man hours per car, they can take some of the savings and put it toward paying whatever workers remain in the process well.
I live in Seattle, which I believe is lower cost of living than Fremont, and I can't get an illegal Mexican day laborer at the Home Depot parking lot to do a day of yard work for less than $14 hr. I find it very surprising that they can find enough people in that area, to work for $17-21 and hour.
 
  • Informative
Reactions: neroden
I'll take some of what your having ;)
Plus you get Mad nerd points for correcting, not grammar, but the accuracy of a historically significant launch pad. To be honest, with some of the nerd chops here, can't understand why there's not an icon with a pair of glasses or a slide rule for nerd points. It would be the gold standard of props.
 
Please don't assume I'm an idiot before retorting to my simplified explanations. I know you've seen me make hundreds of posts here complete with sources and a rational dissection of what I'm saying. I know you've seen me speak tangentially of my career.

I work in software development for a company that manufactures a real-time data acquisition and analysis system that runs on an RTAI Linux platform. Some of our products process and extract data from images from high-speed cameras (in excess of 4000 fps) in real-time. Comparatively speaking, car-speed real-time is SLOW.

GPU's typically have many many more processing cores per physical chip die. They are ideal for this type of application.

A GPU cannot be used without a CPU to control it, so yes, there is additional supporting hardware. The point is, that it does not take a 'trunk full of computers' to do it. A modern household PC with a decent graphics card has the chops to do this stuff, and those can be cleverly packaged to fit a space substantially smaller than a typical PC tower, and doesn't cost more than a 3 digit number of dollars.

My anecdote I relayed about the desktop GPUs being used to process data replacing the server? They were processing ground penetrating radar images for the mining industry to locate fractures in rock layers separating their mine from the water table. I understand more about this than an average Joe.
Now this is the kind of post that SCREAMs for nerd points....double bonus nerd points. I gave it a "like", but that seems....inadequate, and slightly disappointing.
 
And he's been doing this same thing (trying to get Tesla unionized) ever since he got there. Not as if he didn't know the job would require extra effort. What's interesting is that he's never succeeded yet. Unionization evidently doesn't offer that many benefits for their dues. Personally, I think he ought to go work for Chrysler.

Extra effort is one thing, injuries are another. I'm ambivalent about unions. I would like to believe that Tesla employees aren't exploited or abused. If the concerns mentioned in the article are real, I hope Tesla will address them so unions won't seem like the only way to get treated humanely.

I imagine most of the people on this forum have had to work overtime in one sense or another, but for most of us that meant making our brain tired. A vacation or good night's sleep will fix that.

But how would we feel if our working conditions / working overtime meant our hands or back could become permanently damaged?
 
Everytime I see the 'Moran', my mental spellchecker converts it to 'Moron'.

No one is forcing you to work at Tesla. Go somewhere else if the market determines you are worth more than you are being paid.

The ones who complain about wages the most, are the ones doing the least to increase their value.

My first job paid $5.15 an hour with zero benefits and I am forever grateful to this day that employer saw me fit enough to be given a chance.
 
Status
Not open for further replies.