• Light/Dark
  • Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
  • We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.

Evidence growing that new Switch models will use die shrunk Tegra X1 variant, same as new Shield TV models.

Inugami

Inugami

Member
Oct 25, 2017
14,995
A quick summary: We know of a new Shield TV coming out which is using a new model of Tegra X1 with slightly better clocks and thermals. We also know that Switch firmwares since 5.0 have referred to a new hardware arrangement as "Mariko". There is enough of a solid case here to say that the upcoming rumored models of switch will use this same chipset.

Thraktor said:
A few things I've got from the Linux 4 Tegra source code:
  • The m in mdarcy stands for Mariko. This is generally referred to as T210B01 in the L4T source code, but occasionally referred to as T214, and at least once as Mariko. It's the same chip referred to in the Switch firmware and expected in a future Switch device (there it's referred to as both Mariko and T214).
  • The mdarcy platform seems to be a slightly modified darcy (which is the code name for the 2017 Shield TV), but with a Mariko chip (hence the name).
  • There's a newer platform code named sif which appears to be a dedicated Shield TV design with the Mariko/T214/T210B01 chip.
  • The Mariko chip runs at higher clocks & lower voltages than the T210, which would be expected from a newer manufacturing process, but there doesn't appear to be any indication in the publicly available code on CPU or GPU architecture or configuration.
Basically, mdarcy seems to be a test system for the new chipset, and sif will be the actual one released to the public. As to why the developer page refers to the old X1, old clocks, etc., it could simply be that this is an early development platform, so they don't want to reveal the specs until they've publicly announced the new Shield TV.
Click to expand...
Click to shrink...



All very convincing. What we don't know yet is if both upcoming Switch models (the cut down portable, and the 'pro') will use the same chip. If I had to personally guess, I'd guess they both do with software setting these chips for different purposes. Since the chip should be able to both hit higher clocks at similar power draw and heat, as well as the same clocks at much less power draw and heat there is little reason not to use it. Especially since a smaller node means this variant should be cheaper than the original Switch chip at least over a long time span (may initially be a little more expensive if yields aren't great).

(Thanks to everyone in the two previous threads for gathering this info, thought it was best to make a new thread versus it getting lost in all of the other discussions in them)
 
Abicus

Abicus

Member
Oct 25, 2017
489
Australia
What we looking at performance increase wise? I'll need this explained with reference to DBZ power levels and number of Gamecubes taped together.
 
Pluto

Pluto

Member
Oct 25, 2017
6,455
So I shouldn't get a regular Switch right now? I want one for Mario Maker 2.
 
OP
OP
Inugami

Inugami

Member
Oct 25, 2017
14,995
Abicus said:
What we looking at performance increase wise? I'll need this explained with reference to DBZ power levels and number of Gamecubes taped together.
Click to expand...
Click to shrink...
We don't have a ton of specifics yet, just a lot of circumstantial stuff. That said, if this is a die-shrink only (plus security fixes) don't expect anything huge. Maybe a guru powerup on Krillin versus base Krillin?

Pluto said:
So I shouldn't get a regular Switch right now? I want one for Mario Maker 2.
Click to expand...
Click to shrink...
Nothing has been officially announced yet, but there is more than just a little smoke. It's up to you on whether you want to wait or not. If you're just getting it for Mario Maker, I doubt any pro model is going to have any effect on that game. (this won't make games play in 4k or anything extreme like that, and Mario Maker is unlikely to have performance or resolution issues that a slightly faster chip may fix)
 
RockmanBN

RockmanBN

Visited by Knack - One Winged Slayer
Member
Oct 25, 2017
27,985
Cornfields
So don't go expecting native 720p undocked and 1080p docked for games like DOOM, Crash, or Xenoblade 2?
 
Darknight

Darknight

"I'd buy that for a dollar!"
Member
Oct 25, 2017
22,842
Uh, we don't know a new Shield TV is coming out with the new chipset. Just because one was spotted doesn't confirm it's coming out.
 
OP
OP
Inugami

Inugami

Member
Oct 25, 2017
14,995
Darknight said:
Uh, we don't know a new Shield TV is coming out with the new chipset. Just because one was spotted doesn't confirm it's coming out.
Click to expand...
Click to shrink...
This is true, but we do know the Mariko chip is being used in the Shield TV listed and is referenced in the code for Linux for Tegra. So even if this Shield doesn't materialize, it doesn't in any way effect this rumor as it pertains to Mariko and upcoming switch models (which have referrened this chip for months now)
 
Elfotografoalocado

Elfotografoalocado

Banned
Oct 27, 2017
5,618
Spain
There's no evidence it's a shrunk Tegra X1, the specs in the Google developers place are a placeholder. All we really know is that (surprise) it uses ARMv.8, which isn't a whole lot.

The only news here is that, on top of Nvidia making a Jetson module with the new chip, it also seems they are making a Shield TV. But we will have to see.
 
kirbyfan407

kirbyfan407

Member
Oct 25, 2017
2,114
A die-shrunk chip where the "Lite" model runs at the original clocks to extend battery life while the "Pro" model runs at Switch's highest clocks (with a new docked GPU boost mode?) at all times makes sense to me.
 
z0m3le

z0m3le

Member
Oct 25, 2017
5,418
Well we know the max clock speed for the new X1 210b chip was 1.267GHz, if this is indeed the case, and the configuration is the same, the max performance a Switch "pro" would have is 648GFLOPs. I would also expect any vendor to not use max clock, so you'd probably end up with 1.152GHz or 590GFLOPs. I don't think we should take a conclusion from a 14 month old tweet so literally though, the clocks might be higher for Tegra 214 than 210b for instance. They are different profiles, and a 12nm Tegra X1/X2 chip should be able to reach 1.6GHz minimum, considering the 16nm version of such a GPU can hit 1.5GHz. It's worth noting that the 16nm and 12nm process is the same, it's actually a free upgrade offered by TSMC, and is only 12nm in performance metrics, based on TSMC's expectations, it is not physically shrunken.

I've been sitting on this for a little while, because the configuration of Mariko is unknown to us, we don't know what profile differences there are, we don't even know if the 210b profile is simply a faulty chip, so when binning the Mariko chip, they turn off part of the GPU that didn't pass testing, and ship those as Tegra T210b chips instead of T214 chips. Mariko might have more cuda cores, and 210b might just have them disabled in the Shield TV. Just one option of why we can't really be sure about what this means.
 

Deleted member 8593

User requested account closure
Banned
Oct 26, 2017
27,176
I just finally wanna know how big the hardware bump for the "enhanced" model will be. Not much, I assume, but I still need confirmation.
 
Darknight

Darknight

"I'd buy that for a dollar!"
Member
Oct 25, 2017
22,842
KojiKnight said:
This is true, but we do know the Mariko chip is being used in the Shield TV listed and is referenced in the code for Linux for Tegra. So even if this Shield doesn't materialize, it doesn't in any way effect this rumor as it pertains to Mariko and upcoming switch models (which have referrened this chip for months now)
Click to expand...
Click to shrink...

For all we know, they could be planning a standard Switch revision which will use this chip in order to plug security issues. There's no doubt a revised chip exists, but I'm not sure you can say one way or the other what this really means. I do think it was misleading to say a new Shield TV was coming when we don't know that at all. I know for a fact there exists a Shield Portable 2 and is completely functional but will never see the light of day so there's a history of NVIDIA creating devices in development that never materialize as a consumer product.
 
antonz

antonz

Banned
Oct 25, 2017
5,309
Nvidia is making lots of different X1 variants at this point. The Jetson nano is a X1 Variant with a Quad-core Arm A57 with only 128 Maxwell Cores. They cut half its GPU power and removed the other quad core cpu.
 
z0m3le

z0m3le

Member
Oct 25, 2017
5,418
antonz said:
Nvidia is making lots of different X1 variants at this point. The Jetson nano is a X1 Variant with a Quad-core Arm A57 with only 128 Maxwell Cores. They cut half its GPU power and removed the other quad core cpu.
Click to expand...
Click to shrink...
This is good to keep in mind, Mariko and 210b were separate chips, and there is a separate Jetson board coming out, which I feel like wouldn't be needed if it was just the 210b.
 
Elfotografoalocado

Elfotografoalocado

Banned
Oct 27, 2017
5,618
Spain
z0m3le said:
Well we know the max clock speed for the new X1 210b chip was 1.267GHz, if this is indeed the case, and the configuration is the same, the max performance a Switch "pro" would have is 648GFLOPs. I would also expect any vendor to not use max clock, so you'd probably end up with 1.152GHz or 590GFLOPs. I don't think we should take a conclusion from a 14 month old tweet so literally though, the clocks might be higher for Tegra 214 than 210b for instance. They are different profiles, and a 12nm Tegra X1/X2 chip should be able to reach 1.6GHz minimum, considering the 16nm version of such a GPU can hit 1.5GHz. It's worth noting that the 16nm and 12nm process is the same, it's actually a free upgrade offered by TSMC, and is only 12nm in performance metrics, based on TSMC's expectations, it is not physically shrunken.

I've been sitting on this for a little while, because the configuration of Mariko is unknown to us, we don't know what profile differences there are, we don't even know if the 210b profile is simply a faulty chip, so when binning the Mariko chip, they turn off part of the GPU that didn't pass testing, and ship those as Tegra T210b chips instead of T214 chips. Mariko might have more cuda cores, and 210b might just have them disabled in the Shield TV. Just one option of why we can't really be sure about what this means.
Click to expand...
Click to shrink...
There's already Pascal in mobile Tegra, its called Tegra Parker, and it doesn't reach 1.6GHZ, ever. It's capped at 1.3 at maximum, and in the 15W power mode it only reaches 1.12.
Darknight said:
For all we know, they could be planning a standard Switch revision which will use this chip in order to plug security issues. There's no doubt a revised chip exists, but I'm not sure you can say one way or the other what this really means. I do think it was misleading to say a new Shield TV was coming when we don't know that at all. I know for a fact there exists a Shield Portable 2 and is completely functional but will never see the light of day so there's a history of NVIDIA creating devices in development that never materialize as a consumer product.
Click to expand...
Click to shrink...
The security issue with the Switch is already plugged and has been for a year. Only the models sold before mid-2018 or thereabouts (Depending on the manufacture date) are vulnerable to the hardware exploit. Maybe there are other issues, but the hacking community is very focused on the hackable Switches.
 
R0987

R0987

Avenger
Jan 20, 2018
2,837
Of course when i finally buy a shield tv its revealed they have a new one in the pipe line.
 
Last edited:
killer7

killer7

Banned
Nov 22, 2018
609
KojiKnight said:
A quick summary: We know of a new Shield TV coming out which is using a new model of Tegra X1 with slightly better clocks and thermals. We also know that Switch firmwares since 5.0 have referred to a new hardware arrangement as "Mariko". There is enough of a solid case here to say that the upcoming rumored models of switch will use this same chipset.





All very convincing. What we don't know yet is if both upcoming Switch models (the cut down portable, and the 'pro') will use the same chip. If I had to personally guess, I'd guess they both do with software setting these chips for different purposes. Since the chip should be able to both hit higher clocks at similar power draw and heat, as well as the same clocks at much less power draw and heat there is little reason not to use it. Especially since a smaller node means this variant should be cheaper than the original Switch chip at least over a long time span (may initially be a little more expensive if yields aren't great).

(Thanks to everyone in the two previous threads for gathering this info, thought it was best to make a new thread versus it getting lost in all of the other discussions in them)
Click to expand...
Click to shrink...


giphy.gif
 
Simba1

Simba1

Member
Dec 5, 2017
5,383
So we should expect little better battery life and in same time little higher clocks.
 
Darknight

Darknight

"I'd buy that for a dollar!"
Member
Oct 25, 2017
22,842
Elfotografoalocado said:
The security issue with the Switch is already plugged and has been for a year. Only the models sold before mid-2018 or thereabouts (Depending on the manufacture date) are vulnerable to the hardware exploit. Maybe there are other issues, but the hacking community is very focused on the hackable Switches.
Click to expand...
Click to shrink...

I know the old one was plugged, but I thought the hacking community has hinted that they have a couple more exploits that they haven't revealed yet. Certainly, not on the same caliber as the one from the older models, but still holes that can be exploited.
 
Elfotografoalocado

Elfotografoalocado

Banned
Oct 27, 2017
5,618
Spain
Executive Taxi said:
This is so underwhelming?
Click to expand...
Click to shrink...
There's no evidence that it's a shrunk Tegra X1.
There are two sources: one is the entry about the "Shield TV 2019" for Google developers that has the specs of the 2017 Shield TV as a placeholder. (Including a regular Tegra X1 with the same frequencies, no spec update in the only place where it says its still a Tegra X1) Another source states that the new Tegra chip (With no details in the architecture or number of cores) is in use in a new Shield TV.
So we really know nothing new about the chip, the news here is that they are making a new Shield TV around it, which should come as no surprise to anyone.
 
ByteCulture

ByteCulture

Banned
Nov 1, 2017
706
I guess i will wait for Switch Pro then?

Iam going to buy a new Switch at the end of the year with RPG Maker. :/
 
antonz

antonz

Banned
Oct 25, 2017
5,309
Nvidia is in fact very busy making variants all over the place. Even X3(Xavier) now has a Lower Power option with a 6 Core CPU and a 384 Core Volta GPU variant.
 
TheRaidenPT

TheRaidenPT

Editor-in-Chief, Hyped Pixels
Verified
Jun 11, 2018
5,949
Lisbon, Portugal
z0m3le said:
Well we know the max clock speed for the new X1 210b chip was 1.267GHz, if this is indeed the case, and the configuration is the same, the max performance a Switch "pro" would have is 648GFLOPs. I would also expect any vendor to not use max clock, so you'd probably end up with 1.152GHz or 590GFLOPs. I don't think we should take a conclusion from a 14 month old tweet so literally though, the clocks might be higher for Tegra 214 than 210b for instance. They are different profiles, and a 12nm Tegra X1/X2 chip should be able to reach 1.6GHz minimum, considering the 16nm version of such a GPU can hit 1.5GHz. It's worth noting that the 16nm and 12nm process is the same, it's actually a free upgrade offered by TSMC, and is only 12nm in performance metrics, based on TSMC's expectations, it is not physically shrunken.

I've been sitting on this for a little while, because the configuration of Mariko is unknown to us, we don't know what profile differences there are, we don't even know if the 210b profile is simply a faulty chip, so when binning the Mariko chip, they turn off part of the GPU that didn't pass testing, and ship those as Tegra T210b chips instead of T214 chips. Mariko might have more cuda cores, and 210b might just have them disabled in the Shield TV. Just one option of why we can't really be sure about what this means.
Click to expand...
Click to shrink...

Not much information to do an educated guess at this point
 

IamFlying

Alt Account
Banned
Apr 6, 2019
765
Pluto said:
I'm not getting it for Mario Maker only, if the new model is more future-proof I'd rather wait.

I have no idea what a 3DS to New 3DS situation is. :(
Click to expand...
Click to shrink...

Was the biggest update a handheld ever got. Doublecore to quad-core CPU and doubled RAM 128->256MB, but no GPU update . The new Switch will probably have not have such a "big" update, only a bit more RAM and a bit higher clocked CPU (and/or GPU).
 
Simba1

Simba1

Member
Dec 5, 2017
5,383
RockmanBN said:
Was hoping for a decent upgrade from the possibile Pro version because 480p on handheld with Crash is unplayable to me. So blurry.
Click to expand...
Click to shrink...

If I recall Crash runs at locked 480p, so probably that game wouldn't get upgrade in any case unless Activations itself don't release patch for higher resolution.
Games that would benefit most from any higher clocks would be those with dynamic resolutions, in case of those games we would have improvement without any patch for games.
 
ShinobiBk

ShinobiBk

One Winged Slayer
Member
Dec 28, 2017
10,121
IamFlying said:
Was the biggest update a handheld ever got. Doublecore to quad-core CPU and doubled RAM 128->256MB, but no GPU update . The new Switch will probably have not have such a "big" update, only a bit more RAM and a bit higher clocked CPU (and/or GPU).
Click to expand...
Click to shrink...

It's very odd. People always cite new 3DS as a way to temper expectations when it was very much a major upgrade, it's just the system itself wasn't really built to take advantage of it.
 
Zedark

Zedark

Member
Oct 25, 2017
14,719
The Netherlands
RockmanBN said:
So don't go expecting native 720p undocked and 1080p docked for games like DOOM, Crash, or Xenoblade 2?
Click to expand...
Click to shrink...
Considering XBC2 touched 364p at worst, native 720p throughout requires a straight 4x resolution boost, which was never a realistic expectation for a PRO model that comes out less than 3 years after the original system (PS4 PRO was a 2.3x boost, XB1X a bit over 4x after 4 years of launch difference and XB1 not being the most powerful version at launch, either).
 
Last edited:
SuperSah

SuperSah

Banned
Oct 26, 2017
4,079
Anyone who expects a real "Pro" Switch is kidding themselves. This is Nintendo, there is NO way they'll split their userbase.

I expect a lite version and a 3DS type deal with slighter better processing at best.
 

Deleted member 7130

User requested account closure
Banned
Oct 25, 2017
7,685
When I said I'd at least wait for a Switch XL people thought I was being a stupid casual or something
 
z0m3le

z0m3le

Member
Oct 25, 2017
5,418
Elfotografoalocado said:
There's already Pascal in mobile Tegra, its called Tegra Parker, and it doesn't reach 1.6GHZ, ever. It's capped at 1.3 at maximum, and in the 15W power mode it only reaches 1.12.

The security issue with the Switch is already plugged and has been for a year. Only the models sold before mid-2018 or thereabouts (Depending on the manufacture date) are vulnerable to the hardware exploit. Maybe there are other issues, but the hacking community is very focused on the hackable Switches.
Click to expand...
Click to shrink...
Nvidia has stated the 1.5TFLOPs GPU number for Parker, that requires a 1.5GHz GPU clock for the chip. I'm not talking about power consumption. I'm not interested in arguing with you about power consumption or what is possible with clocks. This is where I got the info: "NVIDIA quotes an FP16 rate of 1.5 TFLOPs for Parker, which implies a GPU clockspeed of around 1.5GHz. This is consistent with other Pascal-based GPUs in that NVIDIA seems to have invested most of their 16nm gains into ramping up clockspeeds rather than making for wider GPUs." What you are talking about is this here:
NEGfk3v.png


What you are looking at is a single die "apu" and with both CPUs with max clocks, the GPU is held to 1.302GHz, but that is not to say that the GPU cannot hit a higher profile. Tegra X2 is sold in many places with the specs list of 1.5TFLOPs (fp16) on the GPU, Switch won't use Denver at all, so the performance numbers in this chart is irrelevant.
 
You must log in or register to reply here.