Like Steve Jobs, he’s become a victim of his own reality distortion field. But he doesn’t have the charisma of Steve Jobs, so he just comes off as an inauthentic creep.
But with the rumours around his drug use and partying, documented issues with Ambien, erratic tweeting behaviour, calling people pedophiles etc. He isn't exactly the straightest CEO around.
Holmes never delivered a product that worked. Musk has delivered 4 Tesla models and counting, as well as VTOL rockets among many other accomplishments. Sure, he over-promises and under-delivers (or at least misses his timelines), but I wouldn't compare that to Holmes level fraud by any stretch.
> as t increases, the delta between Musk and Holmes is decreasing.
Not at all.
Musk says wacky stuff on Twitter and has real trouble with public speaking. He is objectively not charismatic at all. Holmes was super polished and had everything locked down.
Musk is able to attract and retain top talent, and push them to accomplish a lot. Holmes could attract politically powerful people with her charisma, but she ended up pushing competent technical staff away.
I don't see much substance in the article. E.g. Nvidia argues that they have chips to meet Tesla's specs, I don't know what constraints Tesla had, but they did mention one - power consumption:
I just scrubbed through the presentation quickly today and happened to stop on a reporter's question about why they built their own chip and at what cost. Musk said basically there were more powerful chips and less powerful chips available but this one was with in their power requirements and had exactly the power they needed at a cost less than what they were paying including R&D costs.
I can understand that rationale, it's like making custom fit patch cables for your network rack that are exactly the length they need to be.
Except if you screw up and cut the cable too short it's going to really hurt your finances, and maybe more importantly, your time to market. There's quite a lot of risk involved but I guess a big enough budget can help mitigate some of that.
Practically speaking there is no power limit. The traction motors on Tesla's cars themselves will pull two orders of magnitude more power than these chips will.
Granted, there are other factors (obviously), but it seems odd to be quibbling about even 400W on a car with a 100KWh battery pack. 0.4% difference in consumption accounts for less than 2mi of range.
400W x 3 hrs = 1.2 kWh = 1.2% of a 100kWh battery.
But that was only one of the issues. They also mentioned waste heat and being able to retrofit the computer to older cars as reasons to keep the power down.
If you're driving a Tesla for three hours, you have likely drained a huge chunk of the battery already, either via the traction motors alone, or in combination with climate control and waste from braking. Tesla air conditioning alone can draw 600-2000 watts. Inference power draw is really nothing.
That level of power use would take *noticeable" dents into the range of the car, especially in inner city driving.
Furthermore that's a lot of thermals to get rid of. 500w, crikey, speaking as someone who water cools his PCs, forget about retrofitting older cars! (That was one of the criteria!)
The existing units sit behind the glove box in front of the firewall, and are rather compact. The thermal budget was a hard one.
Practically speaking there is no power limit. The traction motors on Tesla's cars themselves will pull two orders of magnitude more power than these chips will. Thermals are nothing compared to the heat generated by the battery, especially when supercharging.
All power usage adds up. More important, the unit sits behind the glove box in front of the firewall. You'd need to redesign the car to fit a cooler worthy of 500w in there, thus making retrofitting a much harder task. My 280mm Corsair AIO water cooler struggles with a 140W TDP Intel CPU at full load.
All typical car manufacturers' message when rolling out something new: Buy the new car. Here, existing owners get access to a paradigm shifting feature, by just swapping out what will be an inexpensive module.
Tesla has nailed this upgrade, in my eyes. I don't own any of their cars (although I wish I did), nor do I have any shares.
Let's just all pause for a moment of silence to ponder just how amazing even the PX Xavier is compared to somewhat recent history:
ASCI Red was the first 1 TOP computer. It consumed 850kW of power and cost $46m. It was the fastest supercomputer in the world 19 years ago this June. Now we're arguing over whether all our cars will be driving around with a computer either 144 or 300x as fast using either 8500x or 1700x less power and costing in the neighborhood of 100,000x less.
A bit apples and oranges, though. The ASCI Red is a general-purpose computer that ran whatever you wanted via the full x86 instruction set. The Xavier and Tesla chips are basically very specific FMADD modules that can't do much else.
The blog post literally calls out Musk for inaccuracies:
> But while we agree with him on the big picture—that this is a challenge that can only be tackled with supercomputer-class systems—there are a few inaccuracies in Tesla’s Autonomy Day presentation that we need to correct.
Ask yourself why nVidia would feel the need to put out anything given that they and Tesla are no longer collaborators. Do you believe they’d put out a “Cool claims by Tesla” press release if they didn’t feel the need to point out corrections?
>Ask yourself why nVidia would feel the need to put out anything given that they and Tesla are no longer collaborators.
nVidia must be short-selling Tesla stock. If Hacker News has taught me anything, it's that we can rest assured about our TSLA investments because the only reason anyone ever says something negative about - or disagrees with - Musk or Tesla is that they're short sellers and we've just gotta ride it out.
I'm not defending Tesla's statements on the comparison between Tesla's hardware and Nvidia's hardware, but it feels that the author of the article took an advantage of Nvidia's post to dismiss Tesla's progress on this area.
“The new 14nm based self-developed chip is fabbed by Samsung and is capable of 144 tera operations per second (TOPS) (two chips each 72 TOPS) compared with Nvidia’s Drive Xavier’s theoretical performance of 21 TOPS. It features two fully independent packages, each with their own with DRAM memory, flash storage chips, and power supplies. If one fails, the failsafe is the second one.” [1]
So the correct way to look at this is that Tesla has a redundant 72 TOPS chip compared to nVidia Xavier 21 or 30 TOPS.
It’s nonsensical to compare a stack of chips versus another. NVidia claiming the fair comparison is a 320 TOPS system with 2 Xavier’s and 2 TensorCores pulling 500 watts total. A single Xavier provides 30 TOPS of Int8 at 30 watts.
The redundant Tesla setup by comparison pulls 100 watts total. [2]
A fact check that just parrots Nvidia’s blog with no actual research or fair presentation of the numbers may be par for the course but not terribly useful.
> The redundant Tesla setup by comparison pulls 100 watts total. [2]
Hmm... I recalled him saying 72w in the presentation compared to 54w in the previous hardware. Then he said this was within the allotted power envelope and wouldn't impact range.
It’s a chip for doing exactly the convolutions they want to do a whole lot of ... Nvidia gpus present a very different and more flexible programming model — why is it not the expected case that Tesla’s hardware would have the price and wattage advantage for the problem they intend to throw at it ...?
Samsung fabs it. Do you mean chip design and tape out NRE?
Tesla spent several years designing their current chip, finishing about 1.5 years ago. They hired a industry guru to help them do it, I’m sure the financial investment was massive. Elon stated somewhat wryly “their entire expense sheet is FSD” which is to say their R&D spending in FSD dwarfs their other R&D spending.
They are also already working on a Gen2 of their just released chip for the last year plus.
For Tesla the hardware design is essential to achieving their goal. If they get FSD in the real world before everyone else, they can print money with the Tesla Network. And they simply can’t get there first using general purpose off the shelf hardware.
It’s a core part of the competitive advantage of the Tesla fleet. The fact that you can slot this new hardware into any AP2.5+ car shipped in the last couple years is an amazing perk of owning a Tesla. No other carmaker comes close to offering this kind of long-term support with software and hardware updates to a late-model-year car.
Absolutely it is possible to do better by specializing, that doesn’t mean we generally expect an auto-maker to be able to successfully make nVidia’s best effort in the same marketplace obsolete in their first go at designing their own chip.
This isn’t quite like GPU vs ASIC for Bitcoin mining, for instance, because nVidia Xavier is already a special purpose chip for doing the Int8 calculations. Tesla isn’t competing against a graphics card being repurposed for FSD, they are competing against nVidia’s FSD offering, and using an industry standard metric (Int8 TOPS) to measure their performance.
I think you're massively under-estimating the difference in specialization. Tesla know the resolution of the data going into the Neural Nets, they know the depth of the data, they know the types of non-convolution stages (pooling, ReLU Sigmoid etc). They've probably got a fixed number of layers too. So they're basically completely getting rid of the control logic. That's an enormous chunk of the problem, and lets' remember these are peak performance numbers - the Tesla programming model may be a disaster. Hell, they could probably have bought the IP for the actual computation core.
I'd love to see a real comparison of the two technologies under the Tesla workload, both for what it would say about the chips and what it would say about the workload.
Having said that, an investor call isn't going to be that and Elon isn't the person to give it, so I'm not sure what Bloomberg (or Nvidia) wanted here. "CEO likes own technology" isn't exactly news.
Right, much of the actual hard work in chip design would be outsourced. They pull in enormous amounts of templated chip design ("IP"), and clearly hired top talent to keep the whole thing going. They'd need simulators and labs for testing, but maybe outsource PCB layout. And I believe Elon even stated out loud that the fab itself is in Texas. Nobody makes chips on their own except perhaps Intel or Samsung. Even with all that industry support it's still a remarkable piece of work in a tight time line, but given the funding and the people, entirely reasonable.
The difference is that Trump knows when he is lying but doesn't care, whereas Musk actually thinks he is telling the truth. Musk's pathology is more serious, which is why he probably will never get elected to public office.
Apples to oranges comparison. Of course nVidia are going to claim their GPU processes stuff faster, and ignore the part where you have to get the results of the calculations off the GPU.
52 comments
[ 2.9 ms ] story [ 109 ms ] threadBut with the rumours around his drug use and partying, documented issues with Ambien, erratic tweeting behaviour, calling people pedophiles etc. He isn't exactly the straightest CEO around.
Not at all.
Musk says wacky stuff on Twitter and has real trouble with public speaking. He is objectively not charismatic at all. Holmes was super polished and had everything locked down.
Musk is able to attract and retain top talent, and push them to accomplish a lot. Holmes could attract politically powerful people with her charisma, but she ended up pushing competent technical staff away.
Drive PX Xavier - 30 TOPS - 30 Watts - https://en.wikipedia.org/wiki/Nvidia_Drive#Drive_PX_Xavier DRIVE AGX Pegasus - 320 TOPS - 500 Watts - https://en.wikipedia.org/wiki/Nvidia_Drive#Drive_PX_Pegasus
Both chips if scaled to 144 TOPS consume double the power of Tesla's solution.
I can understand that rationale, it's like making custom fit patch cables for your network rack that are exactly the length they need to be.
https://en.wikipedia.org/wiki/Jim_Keller_(engineer)
So, 500W is going to be well beyond that and wouldn't fit their requirements.
But that was only one of the issues. They also mentioned waste heat and being able to retrofit the computer to older cars as reasons to keep the power down.
That level of power use would take *noticeable" dents into the range of the car, especially in inner city driving.
Furthermore that's a lot of thermals to get rid of. 500w, crikey, speaking as someone who water cools his PCs, forget about retrofitting older cars! (That was one of the criteria!)
The existing units sit behind the glove box in front of the firewall, and are rather compact. The thermal budget was a hard one.
Nvidia are full of crap on this one.
All typical car manufacturers' message when rolling out something new: Buy the new car. Here, existing owners get access to a paradigm shifting feature, by just swapping out what will be an inexpensive module.
Tesla has nailed this upgrade, in my eyes. I don't own any of their cars (although I wish I did), nor do I have any shares.
ASCI Red was the first 1 TOP computer. It consumed 850kW of power and cost $46m. It was the fastest supercomputer in the world 19 years ago this June. Now we're arguing over whether all our cars will be driving around with a computer either 144 or 300x as fast using either 8500x or 1700x less power and costing in the neighborhood of 100,000x less.
https://en.wikipedia.org/wiki/ASCI_Red
Read the source: https://blogs.nvidia.com/blog/2019/04/23/tesla-self-driving/
Then read the Bloomberg article again, and tell me how that isn't complete spin towards a narrative that doesn't exist in the source.
Bloomberg is completely shameless in their anti-anything-Tesla narrative.
> But while we agree with him on the big picture—that this is a challenge that can only be tackled with supercomputer-class systems—there are a few inaccuracies in Tesla’s Autonomy Day presentation that we need to correct.
Ask yourself why nVidia would feel the need to put out anything given that they and Tesla are no longer collaborators. Do you believe they’d put out a “Cool claims by Tesla” press release if they didn’t feel the need to point out corrections?
nVidia must be short-selling Tesla stock. If Hacker News has taught me anything, it's that we can rest assured about our TSLA investments because the only reason anyone ever says something negative about - or disagrees with - Musk or Tesla is that they're short sellers and we've just gotta ride it out.
I'm not defending Tesla's statements on the comparison between Tesla's hardware and Nvidia's hardware, but it feels that the author of the article took an advantage of Nvidia's post to dismiss Tesla's progress on this area.
So the correct way to look at this is that Tesla has a redundant 72 TOPS chip compared to nVidia Xavier 21 or 30 TOPS.
It’s nonsensical to compare a stack of chips versus another. NVidia claiming the fair comparison is a 320 TOPS system with 2 Xavier’s and 2 TensorCores pulling 500 watts total. A single Xavier provides 30 TOPS of Int8 at 30 watts.
The redundant Tesla setup by comparison pulls 100 watts total. [2]
A fact check that just parrots Nvidia’s blog with no actual research or fair presentation of the numbers may be par for the course but not terribly useful.
[1] - https://www.guru3d.com/news-story/tesla-develops-own-self-dr...
[2] - https://electrek.co/2019/04/23/nvidia-disputes-tesla-fsd-com...
Hmm... I recalled him saying 72w in the presentation compared to 54w in the previous hardware. Then he said this was within the allotted power envelope and wouldn't impact range.
I went back and it was around the 1hr 30min mark when they had the chip designer on stage answering questions.
Tesla spent several years designing their current chip, finishing about 1.5 years ago. They hired a industry guru to help them do it, I’m sure the financial investment was massive. Elon stated somewhat wryly “their entire expense sheet is FSD” which is to say their R&D spending in FSD dwarfs their other R&D spending.
They are also already working on a Gen2 of their just released chip for the last year plus.
For Tesla the hardware design is essential to achieving their goal. If they get FSD in the real world before everyone else, they can print money with the Tesla Network. And they simply can’t get there first using general purpose off the shelf hardware.
It’s a core part of the competitive advantage of the Tesla fleet. The fact that you can slot this new hardware into any AP2.5+ car shipped in the last couple years is an amazing perk of owning a Tesla. No other carmaker comes close to offering this kind of long-term support with software and hardware updates to a late-model-year car.
This isn’t quite like GPU vs ASIC for Bitcoin mining, for instance, because nVidia Xavier is already a special purpose chip for doing the Int8 calculations. Tesla isn’t competing against a graphics card being repurposed for FSD, they are competing against nVidia’s FSD offering, and using an industry standard metric (Int8 TOPS) to measure their performance.
“But, you are driving that fraud :-)”
Having said that, an investor call isn't going to be that and Elon isn't the person to give it, so I'm not sure what Bloomberg (or Nvidia) wanted here. "CEO likes own technology" isn't exactly news.
> [Video] How can a company that has never designed a chip, design the best chip in the world?
They hire one of the best chip design teams in the world. Musk said this no fewer than two times during the presentation.