4 ms·

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

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: Drive PX Xavier - 30 TOPS - 30 Watts - https://en.wikipedia.org/wiki/Nvidia_Drive#Drive_PX_Xavier 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 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.

cptskippy 8y ago
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.
- kingosticks 8y ago
  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.
  - cptskippy 8y ago
    Not to mention getting the right people cutting your cables.
    - torpfactory 8y ago
      They did hire Jim Keller to design their chips so... using your analogy they definitely hired the right guy to cut the cables :). https://en.wikipedia.org/wiki/Jim_Keller_(engineer) https://en.wikipedia.org/wiki/Jim_Keller_(engineer)
timdorr 8y ago
They said during the presentation they had a 100W limit. The previous Nvidia hardware is ~50W and this new hardware is ~72W (36W per chip). So, 500W is going to be well beyond that and wouldn't fit their requirements.
- 7e 8y ago
  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.
- FireBeyond 8y ago
  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.
  - perilunar 8y ago
    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.
    - 7e 8y ago
      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.
- sundvor 8y ago
  I'm watching the video. Highly interesting. 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.
torpfactory 8y ago
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. https://en.wikipedia.org/wiki/ASCI_Red https://en.wikipedia.org/wiki/ASCI_Red
- ADefenestrator 7y ago
  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.

cptskippy 8y ago

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.

kingosticks 8y ago
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.
- cptskippy 8y ago
  Not to mention getting the right people cutting your cables.
  - torpfactory 8y ago
    They did hire Jim Keller to design their chips so... using your analogy they definitely hired the right guy to cut the cables :). https://en.wikipedia.org/wiki/Jim_Keller_(engineer) https://en.wikipedia.org/wiki/Jim_Keller_(engineer)

timdorr 8y ago

They said during the presentation they had a 100W limit. The previous Nvidia hardware is ~50W and this new hardware is ~72W (36W per chip). So, 500W is going to be well beyond that and wouldn't fit their requirements.

7e 8y ago
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.
FireBeyond 8y ago
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.
- perilunar 8y ago
  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.
  - 7e 8y ago
    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.
sundvor 8y ago
I'm watching the video. Highly interesting. 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.

torpfactory 8y ago

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. https://en.wikipedia.org/wiki/ASCI_Red https://en.wikipedia.org/wiki/ASCI_Red

ADefenestrator 7y ago
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.