• Hey everyone, staff have documented a list of banned content and subject matter that we feel are not consistent with site values, and don't make sense to host discussion of on Famiboards. This list (and the relevant reasoning per item) is viewable here.
  • Do you have audio editing experience and want to help out with the Famiboards Discussion Club Podcast? If so, we're looking for help and would love to have you on the team! Just let us know in the Podcast Thread if you are interested!
  • General system instability
    🚧 We apologise for the recent server issues. The site may be unavaliable while we investigate the problem. 🚧

StarTopic Future Nintendo Hardware & Technology Speculation & Discussion |ST| (Read the staff posts before commenting!)

I'm just curious: how do we know the chips that'll be used for Dane so early already?
Was there any leak specifically about Nintendo using the Orin chip, or part of it is speculation due to it coming late 2022/early 2023?
Yeah since like June or so we've known that Dane is a cut-down Orin variant. From the above mentioned Nvidia leaker.
 
2TFLOPs@25W is high but it is roughly half the full chip power consumption at 50W.
If Switch 2 is using NX, most likely they will lower the clock speeds. Maybe 1.5 GHz for CPU and 80% of max GPU speeds.. I also don't see the need for ray tracing on a 2 tflop gpu hardware.

Man this is exciting. Especially knowing it's using a 12GB lpddr5 128 bit bud width. It really sounds perfect for switch 2!

Does anyone know the die size of the Orion NC btw?
 
Last edited:
If Switch 2 is using NX, most likely they will lower the clock speeds. Maybe 1.5 GHz for CPU and 80% of max GPU speeds.. I also don't see the need for ray tracing on a 2 tflop gpu hardware.
You may are describing the 15W power profile of this board. It may also be the max profile of the switch 2 chip on a shield TV with another downclock for switch 2 docked mode.
 
0
~

Anyway, in terms of video encode streams, AGX Orin supports H.264, H.265 (HEVC), and AV1.

...AV1 encode was not something I expected whatsoever. I guess that's a Lovelace feature brought back in here?
(since Ampere didn't update NVENC, Lovelace bringing an update would be quite plausible)
 
0
considering this is the 3rd version of Ampere, made for low wattage, I don't think there will be another. it's possible but it just seems like this was what they made to fit within the Switch's constraints. otherwise, I don't think its chip would be "T23x"
If nintendo is asking for a N7/N6 chip, they could use another ampere configuration closer to GA100. Today's Orin presentation is mostly an indication of the power requirements of A78/Ampere on 8 nm.

Orin NX is still consuming more W/TFLOPs than A12X which will be 5 years old at its release.
 
0
I haven't found the area size of the die or the soc size (8nm still?) but I found this on the page:

70mm x 45mm, Jetson Orin™ NX packs a punch with 5X the performance of NVIDIA Jetson Xavier™ NX. Its small size, superb performance, and power efficiency bring big performance to your next-gen products like drones and handheld devices.

jetson-orin-nx-webpage-module-all-1cC-P@2x.png


You guys also note the usage? Next gen products like drones and handheld devices. That's huge. It doesn't outright say gaming devices, but it

Is there any chance they would lower the RAM to 8GB for a switch version, while keeping the bandwidth the same? 12 might be overkill, unless they want to make a beefy OS or really take advantage of a suspension feature.

Will be interesting to see the clocks for 10, 15 and 25w clock speeds for performance mode.. They should release it.


Also, what happened to Orion S and the 5 and 10 watt mode? 🤔 I wonder if it's the nano next now.
 
Last edited:
It's theoretically possible to lower the RAM amount while keeping the same bus size/bandwidth, yes. It depends on what modules are offered by the respective vendor(s).
Samsung's LPDDR5 catalog still only has 64-bit 6 GB, huh.
Micron is sampling 64-bit 4 GB.
SK Hynix offers 16-bit 1 GB... I think? I'm just assuming that page uses uppercase B when it should be lowercase b, because otherwise it makes much less sense.

Oh, yea, going back to your question on Samsung's power consumption and speed gain claims for LPDDR5; my interpretation of "20% less power than LPDDR5" is "20% less power per bit transferred". So I guess LPDDR5X running at 4/3 the speed of LPDDR5 should consume about 16/15ths the power of LPDDR5 running at its respective max.
 
now that we know that Orin does have (some amount of) RT, it calls into mind the early insider news that there was no RT in the dev kits. outside of the "early kits just doesn't have RT enabled hardware", either there is still not RT planned, or Nintendo is gonna surprise them with RT in near final kits next year

there's still the question of just how much acceleration up-to-8 RT cores can bring. could we see Metro Exodus Enhanced on Dane?
 
I haven't found the area size of the die or the soc size (8nm still?) but I found this on the page:

70mm x 45mm, Jetson Orin™ NX packs a punch with 5X the performance of NVIDIA Jetson Xavier™ NX. Its small size, superb performance, and power efficiency bring big performance to your next-gen products like drones and handheld devices.

jetson-orin-nx-webpage-module-all-1cC-P@2x.png


You guys also note the usage? Next gen products like drones and handheld devices. That's huge. It doesn't outright say gaming devices, but it

Is there any chance they would lower the RAM to 8GB for a switch version, while keeping the bandwidth the same? 12 might be overkill, unless they want to make a beefy OS or really take advantage of a suspension feature.

Will be interesting to see the clocks for 10, 15 and 25w clock speeds for performance mode.. They should release it.


Also, what happened to Orion S and the 5 and 10 watt mode? 🤔 I wonder if it's the nano next now.
Orin S is still likely what Dane is, but they wouldn't be discussing it publicly anymore since Nintendo secured it as the chip for the next Switch. After they announce the next Switch Nvidia will likely use Orin S for new Shield products or something, if they make any.
 
A couple screenshots from the GTC Orin session:

This one gives us the relative size of Orin SoC
8H1Kos7.png


Note the "TW" mark on the SoC—does it mean Taiwan/TSMC?
tqPutwh.png


For comparison, the Orin X in Zhiji Motors' press photo bears a "KR" mark—Korea/Samsung?
20210927104810_CZZbMFxCID.png
Perfect size for the switch
 
Note the "TW" mark on the SoC—does it mean Taiwan/TSMC?
tqPutwh.png
AGX Orin's marking mentioning "TW" probably refers to AGX Orin being packaged in Taiwan. Of course, that doesn't necessarily mean that TSMC's N7 process node or more advanced was used to fabricate AGX Orin, since AGX Orin could still be fabricated using Samsung's 8N process node.

GA102's marking also mentions "TW", but GA102 was fabricated using Samsung's 8N process node.
930-default.jpg
 
Last edited:
AGX Orin's marking mentioning "TW" probably refers to AGX Orin being packaged in Taiwan. Of course, that doesn't necessarily mean that TSMC's N7 process node or more advanced was used to fabricate AGX Orin, since AGX Orin could still be fabricated using Samsung's 8N process node.

GA102's marking also mentions "TW", but GA102 was fabricated using Samsung's 8N process node.
930-default.jpg


I wonder if the final packaging step could be done at a different location than the SoC fab step.
 
It's theoretically possible to lower the RAM amount while keeping the same bus size/bandwidth, yes. It depends on what modules are offered by the respective vendor(s).
Samsung's LPDDR5 catalog still only has 64-bit 6 GB, huh.
Micron is sampling 64-bit 4 GB.
SK Hynix offers 16-bit 1 GB... I think? I'm just assuming that page uses uppercase B when it should be lowercase b, because otherwise it makes much less sense.

Oh, yea, going back to your question on Samsung's power consumption and speed gain claims for LPDDR5; my interpretation of "20% less power than LPDDR5" is "20% less power per bit transferred". So I guess LPDDR5X running at 4/3 the speed of LPDDR5 should consume about 16/15ths the power of LPDDR5 running at its respective max.
I'm thinking more from a cost perspective for Nvidia and Nintendo too.. Is it better to split NX into two, with one having 12GB and another 8GB, or having it all at 12GB? 8GB on its own is cheaper than 12GB, but making 2 variants could cost more money. Then again, Nvidia shield is 3GB while Switch is 4GB.

Perfect size for the switch
maybe its just me but the board looks a lot bigger than the switch TX1.. TX1 board for reference
R.14f7f6da59ee0ec13d35797484ace0d8
TX1 developer board
 
I'm thinking more from a cost perspective for Nvidia and Nintendo too.. Is it better to split NX into two, with one having 12GB and another 8GB, or having it all at 12GB? 8GB on its own is cheaper than 12GB, but making 2 variants could cost more money. Then again, Nvidia shield is 3GB while Switch is 4GB.


maybe its just me but the board looks a lot bigger than the switch TX1.. TX1 board for reference
R.14f7f6da59ee0ec13d35797484ace0d8
TX1 developer board
TX1 board is 87mm x 50mm

Orin NX is 70mm x 45mm
 
Orin S is still likely what Dane is, but they wouldn't be discussing it publicly anymore since Nintendo secured it as the chip for the next Switch. After they announce the next Switch Nvidia will likely use Orin S for new Shield products or something, if they make any.
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
 
I'm thinking more from a cost perspective for Nvidia and Nintendo too.. Is it better to split NX into two, with one having 12GB and another 8GB, or having it all at 12GB? 8GB on its own is cheaper than 12GB, but making 2 variants could cost more money. Then again, Nvidia shield is 3GB while Switch is 4GB.
Shouldn't be that much more expensive if the design itself isn't being changed, just the RAM chips being swapped? The price difference there is mainly the RAM itself, right?
(the Shield and the Switch shouldn't require a design change just for that difference in RAM size; for example, Samsung offers both 32-bit 1.5 GB and 2 GB options for LPDDR4/4X)
 
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
The Orin S was targeting 15w, and probably evolved into Orin NX. The 5w variant is meant for the windshield ADAS applications, and AFAIK doesn't have a name; it's supposed to come out in 2023, matching Nano Next's release timeframe.
 
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
Yeah it could be that Nano Next is bimmed Orin S.

I'm guessing it's cost effective to make a separate chip that has similar performance to Orin NX but without all of the additional hardware that is disabled already.
 
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
Yeah, it will be interesting to see just how much the Dane SoC diverges from the Orin NX as size-wise it could fit inside the Switch's form factor with a better cooler (or maybe the same as it had about 20 degrees of headroom even when overclocking Eristra)
 
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
The Jetson Orin NX die is significantly larger than the Tegra X1 die. The only hardware component Nvidia provides to Nintendo is the SoC die. And there are many components on Jetson AGX Orin (and by extension Jetson Orin NX) that are completely unnecessary for Nintendo's purposes, such as the Programmable Vision Accelerator (PVA), the Generic Timestamp Engine (GTE), the Sensor Processing Engine (SPE), etc., that can be removed to make Dane smaller.

AnBtv2z.png
 
I don’t want to derail and I’m sure this has been discussed. What are the potential options for sd cards for the next switch? They need to be faster and hold more memory. There are third party game files that are so big right now. What can Nintendo do to upgrade in that department.
 
I don’t want to derail and I’m sure this has been discussed. What are the potential options for sd cards for the next switch? They need to be faster and hold more memory. There are third party game files that are so big right now. What can Nintendo do to upgrade in that department.
I'm crossing my fingers for UFS storage... which would be both an internal sorage solution as well as card style similar to micro sd
but there are other options.
As well as just not upgrading it altogether
 
0
I don’t want to derail and I’m sure this has been discussed. What are the potential options for sd cards for the next switch? They need to be faster and hold more memory. There are third party game files that are so big right now. What can Nintendo do to upgrade in that department.
I think the best option for external flash storage would be to use UFS Cards 3.0, which seems to have comparable sequential speeds to UFS 2.1, and seems to probably have the best balance of performance, power consumption, and thermals.
 
I haven't found the area size of the die or the soc size (8nm still?) but I found this on the page:

70mm x 45mm, Jetson Orin™ NX packs a punch with 5X the performance of NVIDIA Jetson Xavier™ NX. Its small size, superb performance, and power efficiency bring big performance to your next-gen products like drones and handheld devices.

jetson-orin-nx-webpage-module-all-1cC-P@2x.png
Someone correct me if I'm calculating this wrong but I pixel counted this and I got 295x280 for the silver area, while the board is 928x600. So it roughly translates to ~31% of the width and ~47% of the height of the board, making it ~21mm x ~22mm for about ~460 mm2 of die size.

Assuming I didn't mess up, they will have to cut a lot of it to fit in the Switch form factor, hopefully a huge chunk of it are the unnecessary components for a console @Dakhil mentioned.
 
The aforementioned technical briefing for the Jetson AGX Orin mentions single-lane UFS support.
For single lane, eUFS 2.X offers 600 MB/s while eUFS 3.X offers 1450 MB/s, or so says wikipedia.
For UFS cards, 1.X and 2.0 are 600 MB/s per lane, while 3.0 is 1200 MB/s per lane.

Looking at Samsung's catalog for eUFS... they're all listed for a 2 lane interface. Would they work fine in the Jetson AGX Orin (but at single lane speeds, of course)? Or not?
 
The aforementioned technical briefing for the Jetson AGX Orin mentions single-lane UFS support.
For single lane, eUFS 2.X offers 600 MB/s while eUFS 3.X offers 1450 MB/s, or so says wikipedia.
For UFS cards, 1.X and 2.0 are 600 MB/s per lane, while 3.0 is 1200 MB/s per lane.

Looking at Samsung's catalog for eUFS... they're all listed for a 2 lane interface. Would they work fine in the Jetson AGX Orin (but at single lane speeds, of course)? Or not?
Honestly at that point just use NVME
 
Someone correct me if I'm calculating this wrong but I pixel counted this and I got 295x280 for the silver area, while the board is 928x600. So it roughly translates to ~31% of the width and ~47% of the height of the board, making it ~21mm x ~22mm for about ~460 mm2 of die size.

Assuming I didn't mess up, they will have to cut a lot of it to fit in the Switch form factor, hopefully a huge chunk of it are the unnecessary components for a console @Dakhil mentioned.
From what I understand Orin AGX and Orin NX use the same exact chip. Orin NX however has about half of the chip deactivated, so you can basically cut that area in half by three quarters right off the bat without changing anything about how Orin NX performs.

Then there's plenty of automotive elements that can be removed, so in essence we're likely looking at at least the GPU configuration of Dane in the active portion of Orin NX.

EDIT: re-mathed
 
Last edited:
A couple screenshots from the GTC Orin session:

This one gives us the relative size of Orin SoC
8H1Kos7.png


Note the "TW" mark on the SoC—does it mean Taiwan/TSMC?
tqPutwh.png


For comparison, the Orin X in Zhiji Motors' press photo bears a "KR" mark—Korea/Samsung?
20210927104810_CZZbMFxCID.png
Just like @Z0mble mentioned on the discord, the chipset that have been presented today seems to be the original 17BTr Orin chip made on 8 nm. Orin X 'seems' to be another beast with 21BTr which could be made on 7 nm as said by this Chinese car maker.
if the Orion NX board is really similar in size to the TX1s, I wonder what's then point of it all to make an S variant at 5-10 watts using the same node? Would it be binned or a cut down version of Orion NX I wonder? I wouldn't be surprised if the Orion S ends up being the Nano next.
Xavier already had the credit card board size but the nature of the chip being a binned massive chipset is the reason why it won't see it in a low margin gaming device.
 
0
From what I understand Orin AGX and Orin NX use the same exact chip. Orin NX however has about half of the chip deactivated, so you can basically cut that area in half right off the bat without changing anything about how Orin NX performs.

Then there's plenty of automotive elements that can be removed, so in essence we're likely looking at at least the GPU configuration of Dane in the active portion of Orin NX.
wouldn't it be closer to ¼ the size?
 
now that we know that Orin does have (some amount of) RT, it calls into mind the early insider news that there was no RT in the dev kits. outside of the "early kits just doesn't have RT enabled hardware", either there is still not RT planned, or Nintendo is gonna surprise them with RT in near final kits next year

there's still the question of just how much acceleration up-to-8 RT cores can bring. could we see Metro Exodus Enhanced on Dane?
If it Dane ends up having RT cores it might simply just be the case that early dev kits used stand in cards that don‘t have RT cores and Nintendo just didn’t tell devs that final hardware would have them.
 
0
A couple screenshots from the GTC Orin session:

This one gives us the relative size of Orin SoC
8H1Kos7.png


Note the "TW" mark on the SoC—does it mean Taiwan/TSMC?
tqPutwh.png


For comparison, the Orin X in Zhiji Motors' press photo bears a "KR" mark—Korea/Samsung?
20210927104810_CZZbMFxCID.png

R.14f7f6da59ee0ec13d35797484ace0d8
TX1 developer board
38476bff2f0c481deddc07b546b49d6e03c114af

And to complete the comparison, here is the switch. I don't know if this is mariko or erista
 
Someone correct me if I'm calculating this wrong but I pixel counted this and I got 295x280 for the silver area, while the board is 928x600. So it roughly translates to ~31% of the width and ~47% of the height of the board, making it ~21mm x ~22mm for about ~460 mm2 of die size.

Assuming I didn't mess up, they will have to cut a lot of it to fit in the Switch form factor, hopefully a huge chunk of it are the unnecessary components for a console @Dakhil mentioned.
That sounds about right, in that it is a large chip. But it should be noted that NX is supposed to be the same size as full fat ORIN I believe, and it is just some features turned off.
ODNX02-A2 corresponds to the 20 nm** Tegra X1.

** → a marketing nomenclature used by all foundry companies
Thank you for clarification.
 
Power draw/heat is probably still too high, no? eUFS is 2 or 3 orders of magnitude more power efficient when active, IIRC.


True, and the Jetson AGX Orin apparently uses eMMC 5.1 by default.
Well at least it would at the minimum hit the 250mb/s cap easily if they stuck with eMMC.

And if the Steam Deck can use NVME, I think Dane can use a lower-speed NVME drive or a "Proprietary" NVME solution like Microsoft did with the Series S|X
 
Please read this staff post before posting.

Furthermore, according to this follow-up post, all off-topic chat will be moderated.
Last edited:


Back
Top Bottom