-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
MCU3 Release Date?
- Thread starter AMehta
- Start date
Lol... can’t wait to see Elon time on these.i am waiting for MCU4 too
I'll chime in as a contrarian. I believe MCU2 is built on an Intel reference platform that is already shipping it's next version. I don't think Tesla saves any money in cost of goods by staying on the old platform and the next tech will certainly have tighter security. So I think at the same time Tesla undergoes the trouble to switch they will implement 5G as well. Could be a while and I think it will be positioned as more of an MCU 2.1, but I believe it will happen in the next few years. I'm just not waiting around for it. Any possible increase in CPU performance will likely be traded partially away for lower thermal and power footprint, so it isn't likely to feel much faster.
milleron
Member
They never offered an an MCU “upgrade” when they went from 3G to LTE. An owner could get the upgrade by bearing the cost of the LTE receiver and labor. If there’s a new MCU2.1, I’d guess that they would simply start inserting it in the upgrade kit as soon as they run out of MCU2 units, but that’s IF they’re still offering the upgrade kit which includes the new MCU, the new IC display, some supplemental wiring, and labor. Finding that package deal when so-called MCU2.1 with 5G comes out (if Tesla ever moves to 5G) is certainly not assured.
I’d not wait if I had MCU 1 and planned on keeping my car for several years. It’s probably wiser to take the package deal now and hope there’s a relatively inexpensive 5G upgrade like there was for LTE (about $300).
Last edited:
I like your thinking but waiting for a future proof Tesla means you would never buy. Didn't Elon say they were already working on next custom chip during tech day last year? If I were you I would wait for another few months when many expect battery improvements and refreshed MS announcements.
You may be referring to Elon's comments regarding the self driving hardware HW4, which like HW3 will be a custom chip unlike HW2x which was Nvidia and HW1 which was mobileye. The MCU or center screen computer which now drives the instrument cluster is mostly reference hardware from other manufacturers, formerly based on Tegra ARM but now on Intel Atom x64
I concur there is no future proof Tesla.
Last edited:
milleron
Member
In addition, there's no reason to believe that Tesla would switch to the next Intel platform unless the current one is no longer manufactured. Lord knows they had every opportunity to switch to an adequate sized eMMC that would solve the premature-failure problems with MCU 1. It would have been much simpler than switching to a new Intel platform, but they never did that. Probably for good reasons, they like parts to remain exactly the same.
BrownOuttaSpec
Active Member
If i had to guess it would be at least 2022 just because ALL the 3's and Y's have MCU2. Maybe the roadster would get it but I really think we are at least a couple years off at minimum. but who knows, I am just a random guy in the internet and Elon is... spontaneous.
ahkahn
Active Member
Well, track record-wise:
MCU1: 2012->2018 (6 years)
MCU2: 2018->2024?
Seems that MCU2 can handle everything necessary for today's app requirements, and that is without the software streamlining that they had to do with MCU1 last year. I suspect there will be some updates along the way (5G, etc).
MCU1: 2012->2018 (6 years)
MCU2: 2018->2024?
Seems that MCU2 can handle everything necessary for today's app requirements, and that is without the software streamlining that they had to do with MCU1 last year. I suspect there will be some updates along the way (5G, etc).
MCU1: 2012->2015 (3 years)
MCU1.1: 2015->2018 (3 years) i.e. LTE
MCU2: 2018->2022 (4 years)
MCU2.1 2022+ i.e.5G
Speculation of course.
Last edited:
ahkahn
Active Member
Distinct possibility. I'd say 2021 (MCU2), and then 2.1 -> 2021-2024, which still keeps it all within the timelines...
But yes, all speculation.
ahkahn
Active Member
Tesla will announce MCU3 about a week after you upgrade to MCU2 if your luck is anything like mine.
Thanks all! I’m still debating. I have the same dilemma with iPhones.. If I don’t buy the newest model within a month or two of release, I wait until the next release.
Since MCU2 has been out for 2 years, my concern is that they’ll announce the upgrade the day after I pay for the install - trm2, seems our luck is the same.
Since MCU2 has been out for 2 years, my concern is that they’ll announce the upgrade the day after I pay for the install - trm2, seems our luck is the same.
I would like to add some facts to the MCU3 speculation, since it's quite hard to find specific information on this if you look for it. It got a bit long, I hope this is ok.
What is the MCU2 Hardware actually based on?
from everything I was able to gather the MCU2 is based on the Intel Gordon Peak automotive platform. You can read on several sites that it uses an Intel E3800 cpu (which is not very specific because this is a series and not a cpu) since the Gordon Peak datasheet mentions this CPU. You can read on other sites that it uses an intel E8000 cpu, but I think this is complete horseshit or a typo. I found [1] and [2] as evidence of what is actually used: an Atom A3950, which seems to be a variant of the Atom E3950 (the A3950 cannot be found in Intels "ark" product database). While it might be possible that Tesla has used different CPUs in the MCU2 over the years, I don't think so since the A3950 was introduced by Intel in Q4/2016 (see intel ark link) and is the right one for the use conditions, including "Automotive" and "Industrial Extended Temp". I found a datasheet addendum which lists specifications for the E3950 and the A3950 and the A3950 seems to have a slightly lower TDP of 9.5W (A3950) instead of the 12W TDP (E3950).
How old is this MCU2 cpu?
the A3950/E3950 is a 4-core 1.6GHz base (2.0GHz boost) Intels Apollo Lake series CPU based on the Goldmont architecture which was unveiled April 2016 and supposedly released Q3/2016. Its siblings are commonly used in NAS-Servers, and the cpu architecture was reasonably powerful for an Atom based cpu, but it's not really a high performance CPU (like all cpus derived from atom cores). The refreshed "Goldmont Plus" architecture which debuted a little over a year later in Q4/2017 was a lot more powerful at the same power usage (see here for an appropriate comparison) but never made it into cars. With an Intel introduction date of Q3/2016 the A3950 CPU in the MCU2 it is close to 4 years old and it was never really powerful to begin with. That it still performs pretty well is an indication of how crappy the CPU in the MCU1 nVidia design was
Speculation: MCU3 - switch to a competitor (ARM, AMD):
Tesla could switch from Intel to a competitor for the next MCU. While AMD has great CPUs these days, I think this is possible but unlikely from both sides (cars/industrial not a focus for AMDs). If Tesla switches back to an ARM based cpus, then all bets are off regarding the timeline. They could buy a CPU like a snapdragon and base their board on this, or build their own ARM based chip like Apple does (or tesla does for the AP system), though I would consider Tesla going for their own chip design unlikely as it simply does not offer a lot of advantages and it costs a lot. It made sense for AP, which has very specific compute requirements, but not for the MCU which is basically a PC and has no specific requirements.
Speculation: MCU3 - stay with the Intel automotive platform:
if however Tesla stays with Intel (which I would consider likely, since the MCU2 was a good design which is not in need of revolutionary changes), then we can say something about the timeline. The A3950 is an automotive/industrial-grade cpu, and for the successor architecture "Goldmont Plus" Intel never released an automotive grade cpu. In fact you can read here (in German, sorry), that the embedded CPUs from the A3950 series will be succeeded once the "Elkhart Lake" series cpus based on the Intel Tremont architecture come out in late 2020.
What will Intels Elkhart Lake (series) / Tremont (architecture) offer compared to the old Goldmont architecture?
- CPU performance is hard to guess since the very first Tremond based CPUs have only arrived a few days ago in the significantly different "Lakefield" configuration (which is an entirely different design in a 4+1 big/little configuration) and so far tests were limited to a thermally limited 5W TDP laptop. Tremont however was promised to significantly increase single threaded performance, and that comes on top of the significant improvements that already came with Goldmont Plus. I would guess a real-world performance factor of 1.66x to 2.0x compared to the cpu in MCU2 sould be realistic, but this is of course very much a guess until comparable cpus are actually on the market!
- GPU performance should be significantly improved, the new GPU is supposedly Intel Ice Lake derived, which has pretty good performance for an integrated GPU solution. How good the GPU performance will be cannot be said, since Intel has different versions of the iGPU and we don't know what variant the automotive chip will get. Difference should probably only matter for games, but even if the automotive cpu gets a low-tier GPU, the difference should be big, the old "HD Graphics 505" in the E3950/A3950 was really really bad across the board, so there is a lot of room for improvement.
When will this new Intel Automotive platform be available?
The first Tremont based chips have only now been released after significant delays, probably owed to Intels troubles with the 10nm fabrication issues which are still not resolved. More variants (like those for NAS servers and automotive) are likely to follow by the end of this year (if the 10nm production works) or sometime next year (if there are additional 10nm screwups). Integrating this in a new system and doing some validation will take additional time, so even if intel lists an automotive part with a Q4/2019 release date it will take at least half a year, probably more like 1 year or more until this shows up in a car. Or the platform could be ready and validated and only have been delayed because Intel could not get the 10nm yields up to actually produce the cpus. But even in the most optimistic scenario, I would be surprised if a switch to a new Intel MCU platform would happen before early summer 2020 and likely later.
TLDR:
if MCU3 stays with the intel automotive platform, a release is probably at least 9 months off and very likely later. If I were a betting person I'd say fall 2021.
What is the MCU2 Hardware actually based on?
from everything I was able to gather the MCU2 is based on the Intel Gordon Peak automotive platform. You can read on several sites that it uses an Intel E3800 cpu (which is not very specific because this is a series and not a cpu) since the Gordon Peak datasheet mentions this CPU. You can read on other sites that it uses an intel E8000 cpu, but I think this is complete horseshit or a typo. I found [1] and [2] as evidence of what is actually used: an Atom A3950, which seems to be a variant of the Atom E3950 (the A3950 cannot be found in Intels "ark" product database). While it might be possible that Tesla has used different CPUs in the MCU2 over the years, I don't think so since the A3950 was introduced by Intel in Q4/2016 (see intel ark link) and is the right one for the use conditions, including "Automotive" and "Industrial Extended Temp". I found a datasheet addendum which lists specifications for the E3950 and the A3950 and the A3950 seems to have a slightly lower TDP of 9.5W (A3950) instead of the 12W TDP (E3950).
How old is this MCU2 cpu?
the A3950/E3950 is a 4-core 1.6GHz base (2.0GHz boost) Intels Apollo Lake series CPU based on the Goldmont architecture which was unveiled April 2016 and supposedly released Q3/2016. Its siblings are commonly used in NAS-Servers, and the cpu architecture was reasonably powerful for an Atom based cpu, but it's not really a high performance CPU (like all cpus derived from atom cores). The refreshed "Goldmont Plus" architecture which debuted a little over a year later in Q4/2017 was a lot more powerful at the same power usage (see here for an appropriate comparison) but never made it into cars. With an Intel introduction date of Q3/2016 the A3950 CPU in the MCU2 it is close to 4 years old and it was never really powerful to begin with. That it still performs pretty well is an indication of how crappy the CPU in the MCU1 nVidia design was
Speculation: MCU3 - switch to a competitor (ARM, AMD):
Tesla could switch from Intel to a competitor for the next MCU. While AMD has great CPUs these days, I think this is possible but unlikely from both sides (cars/industrial not a focus for AMDs). If Tesla switches back to an ARM based cpus, then all bets are off regarding the timeline. They could buy a CPU like a snapdragon and base their board on this, or build their own ARM based chip like Apple does (or tesla does for the AP system), though I would consider Tesla going for their own chip design unlikely as it simply does not offer a lot of advantages and it costs a lot. It made sense for AP, which has very specific compute requirements, but not for the MCU which is basically a PC and has no specific requirements.
Speculation: MCU3 - stay with the Intel automotive platform:
if however Tesla stays with Intel (which I would consider likely, since the MCU2 was a good design which is not in need of revolutionary changes), then we can say something about the timeline. The A3950 is an automotive/industrial-grade cpu, and for the successor architecture "Goldmont Plus" Intel never released an automotive grade cpu. In fact you can read here (in German, sorry), that the embedded CPUs from the A3950 series will be succeeded once the "Elkhart Lake" series cpus based on the Intel Tremont architecture come out in late 2020.
What will Intels Elkhart Lake (series) / Tremont (architecture) offer compared to the old Goldmont architecture?
- CPU performance is hard to guess since the very first Tremond based CPUs have only arrived a few days ago in the significantly different "Lakefield" configuration (which is an entirely different design in a 4+1 big/little configuration) and so far tests were limited to a thermally limited 5W TDP laptop. Tremont however was promised to significantly increase single threaded performance, and that comes on top of the significant improvements that already came with Goldmont Plus. I would guess a real-world performance factor of 1.66x to 2.0x compared to the cpu in MCU2 sould be realistic, but this is of course very much a guess until comparable cpus are actually on the market!
- GPU performance should be significantly improved, the new GPU is supposedly Intel Ice Lake derived, which has pretty good performance for an integrated GPU solution. How good the GPU performance will be cannot be said, since Intel has different versions of the iGPU and we don't know what variant the automotive chip will get. Difference should probably only matter for games, but even if the automotive cpu gets a low-tier GPU, the difference should be big, the old "HD Graphics 505" in the E3950/A3950 was really really bad across the board, so there is a lot of room for improvement.
When will this new Intel Automotive platform be available?
The first Tremont based chips have only now been released after significant delays, probably owed to Intels troubles with the 10nm fabrication issues which are still not resolved. More variants (like those for NAS servers and automotive) are likely to follow by the end of this year (if the 10nm production works) or sometime next year (if there are additional 10nm screwups). Integrating this in a new system and doing some validation will take additional time, so even if intel lists an automotive part with a Q4/2019 release date it will take at least half a year, probably more like 1 year or more until this shows up in a car. Or the platform could be ready and validated and only have been delayed because Intel could not get the 10nm yields up to actually produce the cpus. But even in the most optimistic scenario, I would be surprised if a switch to a new Intel MCU platform would happen before early summer 2020 and likely later.
TLDR:
if MCU3 stays with the intel automotive platform, a release is probably at least 9 months off and very likely later. If I were a betting person I'd say fall 2021.
Similar threads
