News Tesla Model Y Switches From Intel to AMD Ryzen Processors

[Admin]

Tesla's Chinese versions of the Model Y performance variants are getting AMD Ryzen APU hardware as replacement for Intel SoC's used in the car's infotainment system.

Tesla Model Y Switches From Intel to AMD Ryzen Processors : Read more

 

[HideOut]

The "performance" version will likely change in the future once they go thru old supply. No need to run two different systems. More coding work.

 

[jkflipflop98]

Cool. Designed around a chip thats in extremely short supply and high demand. This will work out well for them.

 

[mitch074]

Cool. Designed around a chip thats in extremely short supply and high demand. This will work out well for them.

I wouldn't be surprised to learn that this chip is the same as the one planned for the Steam Deck - Intel haven't come out with anything that can compete with that, making Tesla's choice for them.

 

[Liquidrider]

AMD isn't as constrained as you may think. They have had orders in place with TSMC years in advance. Beyond that I strongly believe AMD will split its orders with Samsung & TSMC. Similar to what happened when AMD had orders in from both GlobalFoundries & TSMC.. Add in that Xilinx and Samsung already have a joint venture.

Musk didn't tweet "AMD has been great to work with!" for nothing.

Read the reuter's article "AMD books capacity years ahead to ease crunches"

 

[Kj1985]

Beyond that I strongly believe AMD will split its orders with Samsung & TSMC.

Doubt they will do that. That would end their preferred partnership with tsmc.

 

[jkflipflop98]

Doubt they will do that. That would end their preferred partnership with tsmc.

So TSMC will give you a discount on your chips if you ONLY buy from them? Hmmmm I remember someone else that makes chips getting the pants fined off them for exactly the same thing a few years back.

 

[rluker5]

Going from Intel atom to Ryzen is an improvement. I imagine infotainment use is pretty light so they may want a ULT Intel instead for responsiveness if they are going low power, but considering they were using atom, they probably don't want to spend that much.

That and gaming on dedicated Tesla hardware is silly when you can get a laptop that will sit right in front of you, or even mount a game console in the trunk if you need better performance. An off center dashboard gaming setup could have a 6900xt and it would still be trash due to the size and location of the display.

If they have the gpu/cpu for ai assist purposes then AMD is the wrong choice and this indicates they are giving up on that.

 

[digitalgriffin]

If they have the gpu/cpu for ai assist purposes then AMD is the wrong choice and this indicates they are giving up on that.

Why do you say this? While Intel has a lot of serious accelerators for AI, that only exist on higher end chips where the money is.

The move is because Intel is 14nm and 10nm constrained. Margin for what's left of the Atom product line is negligible compared to something like Alderlake/Icelake & Server Chips. So intel charges a premium which no longer makes them viable against AMD's much better performance : price offerings at the ultra low end systems.

The iGPU portion of the AMD chips include a lot of vector processing instructions which is inherent to AI. (Even on GCN) AMD's semi custom SOC chips provides a lot of power to customers at a reasonable price. $80 for an AMD Semi Custom SOC or $80 for a much older, and less powerful Intel Atom SOC?

 

[rluker5]

The move is because Intel is 14nm and 10nm constrained.

Nonsense.

AMD's much better performance : price offerings at the ultra low end systems.

That's probably the reason. Atom was probably a little too laggy and AMD was just good enough and still cheap.

The iGPU portion of the AMD chips include a lot of vector processing instructions which is inherent to AI.

Just because it has some doesn't mean it is anywhere near enough. I bet most modern higher end phone soc's have more.
Maybe Tesla is just going full cloud AI with it only working via remote connection. Good for a subscription model.

 

[Kj1985]

So TSMC will give you a discount on your chips if you ONLY buy from them? Hmmmm I remember someone else that makes chips getting the pants fined off them for exactly the same thing a few years back.

While getting close TSMC does not have a monopoly. And it is not so much a discount as access to their engineers so the customer can make the best product. TSMC does not want to leak secrets to their competitors through their customers.

 

[digitalgriffin]

Nonsense.

That's probably the reason. Atom was probably a little too laggy and AMD was just good enough and still cheap.

Just because it has some doesn't mean it is anywhere near enough. I bet most modern higher end phone soc's have more.
Maybe Tesla is just going full cloud AI with it only working via remote connection. Good for a subscription model.

https://www.oregonlive.com/silicon-...slower-pc-sales-and-constrained-supplies.html

You can't run AI over a cloud for automated driving algorithms.

 

[watzupken]

Doubt they will do that. That would end their preferred partnership with tsmc.

I will be honest here, the only company that will get preferential treatment from TSMC is Apple. No other company pay TSMC as much as Apple each year for the most cutting edge node. The only exception being this year where Apple is using N4 instead of N3 because the cycle is messed up.
Under normal circumstances, I think it makes sense to stick with TSMC. But given the constraints of late, I feel it may be better to pay more than to sell less. If a chip company cannot supply enough chips to their clients, it is a matter of time that the clients may cut down on orders by not offering AMD products.

 

[watzupken]

Going from Intel atom to Ryzen is an improvement. I imagine infotainment use is pretty light so they may want a ULT Intel instead for responsiveness if they are going low power, but considering they were using atom, they probably don't want to spend that much.

That and gaming on dedicated Tesla hardware is silly when you can get a laptop that will sit right in front of you, or even mount a game console in the trunk if you need better performance. An off center dashboard gaming setup could have a 6900xt and it would still be trash due to the size and location of the display.

If they have the gpu/cpu for ai assist purposes then AMD is the wrong choice and this indicates they are giving up on that.

Performance aside, I suspect cost is also a factor. If you look at low cost PCs now, I can find mini PCs utilizing the aged Ryzen 2500U for a slight premium over one that uses an Intel Celeron/ Pentium Silver chip. The former while being the first gen Ryzen chip is still very fast when compared the the Celeron/Pentium parts, and with an iGPU that wipes the floor when compared to the cut down Intel UHD solution.

 

[spongiemaster]

Doubt they will do that. That would end their preferred partnership with tsmc.

There's no such thing as preferred status. What and how much future capacity a company is able to reserve depends almost exclusively on how much that company is willing to pay for that capacity. That's how Intel jumped AMD in the "importance queue" for 3nm capacity, more money.

 

[spongiemaster]

Performance aside, I suspect cost is also a factor. If you look at low cost PCs now, I can find mini PCs utilizing the aged Ryzen 2500U for a slight premium over one that uses an Intel Celeron/ Pentium Silver chip. The former while being the first gen Ryzen chip is still very fast when compared the the Celeron/Pentium parts, and with an iGPU that wipes the floor when compared to the cut down Intel UHD solution.

Most definitely. AMD has traditionally been more willing to cut margins than either Intel or Nvidia, which is why AMD has landed so many semi-custom clients.

 

[Giroro]

More importantly, why does that car have a yoke instead of a steering wheel?
It's like, the Windows 11 of cars.

It's like, cool you can play the Witcher 3... .... But maybe you should stop it from crashing, first?

 

[Kj1985]

There's no such thing as preferred status. What and how much future capacity a company is able to reserve depends almost exclusively on how much that company is willing to pay for that capacity. That's how Intel jumped AMD in the "importance queue" for 3nm capacity, more money.

Or AMD was not ready for the 3nm product yet. TSMC does have a preferentially partner program, and one of the major stipulations is that they single source from TSMC. That is the major reason Qualcomm lost capacity at TSMC.

 

[mitch074]

Nonsense.

That's probably the reason. Atom was probably a little too laggy and AMD was just good enough and still cheap.

Just because it has some doesn't mean it is anywhere near enough. I bet most modern higher end phone soc's have more.
Maybe Tesla is just going full cloud AI with it only working via remote connection. Good for a subscription model.

Atom has a power to performance ratio that's just too low for any modern application. A quad core Zen2 chip will blow it out of the water within the same power envelope.
AMD's RDNA2 is way ahead of whatever mobile chips provide (except Nvidia's Tegra, and that's debatable) - there's a reason Samsung wanted AMD to design their next iGPU. Heck, even Intel are still chasing after AMD's perf-per-watt in embedded! Remember Hades Canyon?

 

[rluker5]

https://www.oregonlive.com/silicon-...slower-pc-sales-and-constrained-supplies.html

You can't run AI over a cloud for automated driving algorithms.

So what do they have left that is adequate? Like around 100 tops? We aren't talking GTA5 npc driving ai here. It just looks like they are giving up on in car ai.

Performance aside, I suspect cost is also a factor. If you look at low cost PCs now, I can find mini PCs utilizing the aged Ryzen 2500U for a slight premium over one that uses an Intel Celeron/ Pentium Silver chip. The former while being the first gen Ryzen chip is still very fast when compared the the Celeron/Pentium parts, and with an iGPU that wipes the floor when compared to the cut down Intel UHD solution.

Cheap core mini pcs are at a price premium. Probably also is the case for a custom build like in a Tesla.

Concerning igpu, depends what you are talking about. For desktop image display and video playback Ryzen isn't significantly ahead of a 2w total system power soc atom igpu from 2014. But the Ryzen cpu would access files and start programs quicker and smoother, but maybe not first gen Ryzen. No need to limit it to 1st gen though. Ryzen's igpu advantage is only in gaming, which doesn't matter for those with dgpus or not gaming on an igpu. For infotainment display the new intel igpus probably beats anything Ampere, much less Radeon or AMD igpu. But AMD does have a relatively good gaming igpu for what that matters.

Atom has a power to performance ratio that's just too low for any modern application. A quad core Zen2 chip will blow it out of the water within the same power envelope.
AMD's RDNA2 is way ahead of whatever mobile chips provide (except Nvidia's Tegra, and that's debatable) - there's a reason Samsung wanted AMD to design their next iGPU. Heck, even Intel are still chasing after AMD's perf-per-watt in embedded! Remember Hades Canyon?

Atom has a high enough performance to power ratio to compete with ARM in phones, it just gets worse software support than windows mobile. Neither the Zenfone2 nor Leagoo T5c were lacking vs arm in perf/watt. The Leagoo had 8 cores and no heatsink. Just a bare black die and my daughter would still be using hers if she hadn't trashed the charge port. AMD isn't even on the same planet in that regard. Their I/O chiplet alone needs a fan.
Where Atom is lacking for light use is cache speed and memory latency. It is terrible and you can see the lag in lower powered parts. Ryzen over 15w doesn't have this problem. But neither does Core over 5w. Even Broadwell from 2015 (check reviews). Ryzen can win in "productivity" like cinebench, but this is an infotainment display.
How heavy of use do you think infotainment is?

Likely the Ryzens are good enough at the best price. Maybe Musk sees some Ryzen hype wave and is all about exploiting hype. Maybe both