• 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.

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

Reader mode
LiC said:
I really don't want to get into process node posting, but either this chart's numbers are wrong, or the entire discussion where "Orin's power consumption is bad on 8nm and we need to move to at least 5nm for it to be better" has been foundationally flawed. It's not clear to me which node Orin's 8N is based on, but let's assume it's 8LPP for now. According to this chart, 8LPU offers a whopping 60% reduction in power consumption over 8LPP. That already makes it better than 5LPP! 5LPP would have 46.8% the power consumption of 10LPP, as you calculated, while 8LPU would have only 36% of 10LPP. 8LPA would be even better at 30.6% of 10LPP.

According to the other Samsung chart in this post, 8LPU is the direct evolution of 8LPP. So if T239 needed to move off 8N/8LPP to improve power consumption, why would they make the more expensive and difficult jump to branch of the family with 5LPP, even though 5LPP is less efficient than the directly available 8LPU? If 8N is actually based on 8LPU instead (which this article claims is the case), then this would make even less than zero sense, since moving from 8N/8LPU to 5LPP would actually result in a less power efficient chip than if they had just stuck with what Orin was already using.

If we believe the chart, the first possible node that would result in power savings compared to 8LPU would be 4LPP, at 25.2% of 10LPP's power consumption. Although, that's if you even trust that number, since it's based on Samsung's claim from sometime last year about a node that didn't exist then and possibly still doesn't exist today.

So I'm wondering if this chart is just totally wrong. It seems to be based entirely on Samsung's marketing claims and I don't see an indication that any testing was done to verify it.

I feel the need to state explicitly that even if the numbers are all accurate, none of this means a worse performing node is going to be used. It would just mean the 5nm rumor is wrong, and things go back to being between Samsung 8nm (though potentially a significantly better 8nm than Orin uses) and TSMC. And with that, I will try to again start ignoring process node discussions because I really just don't care.
Click to expand...

Yeah, these node comparisons are very difficult, as there are so many variables that you can't just say X node consumes Z% less than Y node and be done with it. Diagrams like the one you quoted just muddy things further, as they're almost certainly combining measurements made in different ways at different times on different circuits at different points of the voltage curve, and trying to align them all together is a quick path to madness.

However, in the case of comparing Samsung's processes, there is one slide they have shown which is at least like-for-like across multiple manufacturing processes:

4475af18972bd40753f3f2486c899e510fb3091b-2.jpg


This is from a presentation in July 2021 covered by Anandtech. It's interesting because rather than the usual numbers provided completely out of context, they're specifically measuring the performance of a sub-block of the ARM A57 core. This doesn't necessarily have any relationship to the scaling we would see with T239's hardware on different nodes (aside from the co-incidence of using a sub-block from TX1's CPU), but it's at least a like-for-like comparison using a real-world CPU design. Frustratingly it doesn't show the exact processes, but it shows approximately a 30% reduction in power from Samsung's 8nm process to their 5nm process, which is about what I'd expect. It also shows about a 20% increase in clocks, presumably at ISO-power.

Again, this can't be directly extrapolated to something like T239, and the power savings can vary a lot depending on where you are on the voltage curve, but it's probably the least-worst point of reference if anyone is curious about the difference between Samsung's 8nm and 5nm foundry offerings.
 
Does the Steam Deck's SSD even run at fast speeds or is it throttled to not drain too much power and produce too much heat.

Is it possible to do an SSD that actually runs pretty fast (~600 MB/s) on mobile hardware that uses <=11W?
 
Concernt said:
Possibly, but that doesn't mean they would limit those in TV mode.
Click to expand...
Well, there are connected limitations. More power in TV mode means more heat, which means a bigger fan and heat pipe, higher cost to ship, and heavier to hold, less room for battery, etcetera. And I have argued quite strongly that I think Nintendo will try to do what they did with the original Switch, which is match performance ratios to pixel ratios, then tweak from there as performance/battery life issues are found. It is both a sane starting point for the design and the likely natural end point of developing games and hardware simultaneously.

No judgement of some folk's desires/expectations of the device, but my (amateur) engineering mind can't make 1080p screen/4TF docked/OG Switch battery life/no significant form factor changes work, purely from a realm of physics. LPDDR5 has on paper greater efficiency than LPDDR4, but that power efficiency comes from long idle periods on the memory controller, which just don't happen in a gaming console. OLED screens are more power efficient on paper, but in practice, there are no savings on actual video games when comparing the power draw of the OLED vs the Redbox Switch.

With T239, Nintendo has a custom chip from the beginning, and they're stepping into the same environment that pushed Microsoft to make a Series S - they just don't believe that a cost reduced version of the base system is going to be available within the lifetime of the device. T239 will need to serve all of Nintendo's needs across the lifetime of the system and it's likely revisions.

We know Nintendo wasn't happy with Erista's battery life, and moved to customize it as soon as possible. I think we should look to Mariko more than Erista for Nintendo's battery life goals. I think we can look to the Switch for Nintendo's performance targets too.

Switch was 20% of a PS4. Nintendo knows that their device lives and dies on exclusives and their "unique way to play" and the Switch's Unique Way to Play was portable mode. Nintendo enabled a number of impossible ports by betting on a simple, flat architecture, with more modern features, and then just selling a ton of units to drive the economics to their side.

Nintendo doesn't need 4TF to make their next gen exclusives. 2.3TF docked would already be a massive 6x jump, plus DLSS and RT to add on top. Plenty for next-gen Mario, and put it in a slightly better league relative to a PS5 than the Switch was to a PS4. A 1TF handheld mode and a 720p screen would put it within spitting distance of the Steam Deck, while offering a much better feature set, and (with the power of ARM) a much more comfortable handheld experience with better battery life. And that performance gap would be very easy for developers to handle, unlike the Series S/X gap.

Note that this 2.25x gap is actually much larger than the current 1.7x gap between docked and handheld. I'm arguing for a larger gap. And I'm not arguing that these perf numbers represent some kind of ceiling. Nor does it represent my personal priorities. I just think in the game of "Fantasy Console Manufacturer" it reflects what, at least to me, seem like Nintendo's priorities as well as the current market, the realities of software development, and the laws of physics.

I expect the final numbers to reflect 1000s of manhours of hardware and software developers working in tandem to get all the details right, even more so than the original Switch. And I expect them to reflect at least one major, unaccounted for difference that we don't know about yet. And I of course expect to have gotten several things wrong!*

tl;dr: I think that REDACTED will have a bigger gap between handheld and docked mode than Switch, but I think for lots of reasons those two are still deeply connected. I don't see any major tech on the horizon that is likely to improve handheld battery life significantly that Nintendo/Nvidia could bet on, and I think the OG Switch battery life is something Nintendo would like to avoid repeating.

*One example might be a frequency scaling solution. I've speculated about this heavily before, but with the extreme range of CPU power required by cross-gen games versus some 9th gen exclusives, Nintendo might get a lot of mileage out of allowing more GPU perf in exchange for dropped CPU perf, or vice versa.
 
tl;dr: I think that REDACTED will have a bigger gap between handheld and docked mode than Switch, but I think for lots of reasons those two are still deeply connected. I don't see any major tech on the horizon that is likely to improve handheld battery life significantly that Nintendo/Nvidia could bet on, and I think the OG Switch battery life is something Nintendo would like to avoid repeating.
Click to expand...
@oldpuck i hope you are rightr, one of my biggest gripe is the narrowed gap between docked/portable performance on Switch because of the higher clock tier unlocked for certain games.
 
0
Thraktor said:
Yeah, these node comparisons are very difficult, as there are so many variables that you can't just say X node consumes Z% less than Y node and be done with it. Diagrams like the one you quoted just muddy things further, as they're almost certainly combining measurements made in different ways at different times on different circuits at different points of the voltage curve, and trying to align them all together is a quick path to madness.

However, in the case of comparing Samsung's processes, there is one slide they have shown which is at least like-for-like across multiple manufacturing processes:
Click to expand...
The 27% TDP improvement from SEC 8nm to SEC 5nm in this specific example happens to be right on the money for the gap between SEC 8nm and TSMC 7nm for Ampere. It also matches ARMs own rough numbers of A78 on various nodes.

Which matches with the general consideration of SEC 5LPP as their "catch up" to TSMC 7nm.
Thraktor said:
Again, this can't be directly extrapolated to something like T239, and the power savings can vary a lot depending on where you are on the voltage curve, but it's probably the least-worst point of reference if anyone is curious about the difference between Samsung's 8nm and 5nm foundry offerings.
Click to expand...
Take the Jetson Power Tool, plug in Switch clocks and T239 specs, set the load to medium, and apply a 30% power savings and you get roughly the Switch's power draw. One of the reasons I've been thinking TSMC 7nm for so long. If Sammy can offer that level of gain over 8nm, for better than TSMC cost, on a long lived node, then that is a very comfortable place to be in.
 
ItWasMeantToBe19 said:
Is it possible to do an SSD that actually runs pretty fast (~600 MB/s) on mobile hardware that uses <=11W?
Click to expand...
There's UFS, which is what replaced eMMC on high end phones.

Looking at Wikipedia, single lane UFS 2.X, 3.X and 4.X are respectively 600, 1450 and 2900 MB/s.

TBH, I forgot about UFS 4 when I last replied you, because it launched last year and I just unconsciously crossed the possibility. I still find it unlikely for now, but it's a possibility if Nintendo really goes all-in.
 
My Tulpa said:
Oh sure, if there is some interesting gameplay design they could think of that can only work by using the tensor cores for unique AI gameplay stuff, or using the RT cores for ray trace lighting for fundamental gameplay...there of course will be some intersting, nich-y Nintendo games that might run on Drake only.

I expect that.

I dont see why releasing their big games on the OLED will be "hampering" their development while the Drake model option is on the market.

The next big 3D Zelda or next big 3D Mario or next big Pokemon etc that will rely on huge raytracing/AI gameplay integration that make the gameplay different than anything before...is like years and years away.



This is the best time for Nintendo to start slowly introducing DLSS and ray tracing techniques into their game developments. Since this is the future of their hybrid type gaming. Do it alongside a still very successful system with high engagement and much life left.

I don't see what "lesser" kind of SoC they could had that could comfortably do DLSS and raytracing at very low clocks and very low power thresholds...than the apparent SoC they and Nvidia have come up with?

This SoC design seems to be the best option if you want to take a Switch game and use DLSS to make it instead run at a locked resolution and locked framerate that matches the screen resolution and refresh rate you are playing on...with some left over to sprinkle in some higher graphics IQ as well.

I mean...what are people expecting this thing to do besides mostly this?
Click to expand...
DLSS is a technique for trying to lighten the rendering load. If the load is already sufficiently light because the game is designed to run on much weaker hardware (which, importantly, needs to run the game without any hardware accelerated upscaler), then you're not really gaining much. Similarly, any effects done with RT will have to be duplicated with more conventional raster methods on the original Switch, because a tiny Maxwell GPU isn't going to be able to run RT effects at any usable speed (even a giant Pascal GPU has a ton of trouble with them, this is why RT cores exist).

Trying to introduce these techniques slowly in way that still allows for games to run acceptably on Switch is probably the worst way to do it, since you're increasing the development effort while simultaneously putting significant limits on how much you can actually use or benefit from them. They're much better suited for exclusive games.
 
BlueSpirit20 said:
I'm not sure how "insiders" can come to the forum and complain about YouTubers making content from what is being posted in the forum, which is only being made because these "insiders" are looking for attention themselves!!
Click to expand...
Bet this didn't catch on like you hoped it would lol
 
Reviews of the Steam Deck tend towards there not being much difference between SD card and the NVMe drive, which points to it being throttled to hell and back.
Just as well anyway; if you take a random ass PCIe gen 3 drive (that is not SK Hynix's Gold P31*), your average sequential read perf/watt is no higher than somewhere in the 600's.
For reference: Samsung claims that its eMMC can achieve sequential read of 330 MB/s in 0.5 watts, which normalizes to 660 MB/s:1 watt. That's right, it's very possible that a random gen 3 NVMe drive is less efficient than eMMC.

*The Gold P31 is probably the most efficient gen 3 NVMe drive with a sequential read of about 1 or 1.1 GB/s:1 watt.

As for eUFS's power draw...
Uh, I suppose there's this post I made almost a year ago. My personal thought? eUFS would be lovely.
 
ninjas definitely paid that guy a visit. he's shook!

maybe he means it's delayed due to other products changing process? is Atlan/Thor of any significance?
 
ItWasMeantToBe19 said:
How much would 256 GBs of UFS 4.0 storage cost for Nintendo to add in the Switch 2.
Click to expand...
I don't think that there are any teardowns of phones with 4.0 yet?
Looking back on my posts (here and here), 256 GB of eUFS 3.0/3.1 has been in the mid to high $20 range according to a few TechInsights teardowns over the past few years. To compare, 64 GB of eMMC should be below $10 by now.
 
0
Sometimes I think about how ridiculous it would be if everything bad happened to actually be true.

OLED was originally a Pro, downgraded last minute. Switch Drake was going to be the next 'Pro' but was fully cancelled, not coming anymore ever. Nintendo doesn't know yet what their next console will be, they're still figuring things out after the previous two plans fell through. Likewise they have no plans for games for the rest of the year and are actually going to try and just coast until the Switch's 10 year mark because they saw people on Twitter say it's too soon for a new Switch and that it's doing just fine.

It would truly be the darkest timeline.

Edit:. The next console is the Switch U. It's a Switch with 5G capabilties, but only plays Switch games natively and all the new fancy games are exclusive to the cloud.
 
Thraktor said:
I've had a look for photos of the 2019 Switch model's motherboard, but haven't been able to find any that show enough detail to identify whether individual pins are wired up. iFixit posted a comparison photo between the two boards, but only of one side of the board, and the USB-C hardware (and traces) are on the other side, unfortunately. I think it's a fairly safe bet that it doesn't have all four lanes wired up, though, because the "external_display_full_dp_lanes" flag is false for IcosaMariko, and in both the Icosa and Aula cases the value of this flag correctly matches up to the underlying hardware.

I had forgotten that the 11 developers article came out after the OLED model was announced, although I would still expect that the conversations may have happened over many months, it's definitely possible that there was come cross-over with Drake development, particularly if he asked any sources again after the OLED announcement to confirm. I suppose the thing I can't get my head around is that in 2021, over two years before the launch of [redacted], there would be so many dev kits in the wild that 11 different developers would be willing to talk to a single journalist about it. You'd need dozens and dozens of developers having dev kits in hand to the point where 11 of them would be willing to spill the beans to a single journalist, and that doesn't make much sense to me for a T239-based console over two years away.

I could definitely see a handful of third party devs knowing about and having dev kits for [redacted] at that point, but it would really be limited to teams making exclusive software for it, and maybe a few other third parties that Nintendo has a good relationship with and is looking for feedback from. It's probably a single-digit number of third parties at that point, and they're ones who are probably much less likely to leak. Zynga, for example, definitely doesn't fit in either of those categories and there's no way they'd need over two years to port a mobile game to a T239-based system.

I can definitely see there being a mix of Aula and [redacted] devs in his reporting. I believe his first reference to third parties working on a 4K-capable Switch was in this article in August 2020, citing a 2021 launch. It doesn't seem plausible that anyone could have expected a T239-based console to launch in 2021, even if dev kits were out there back then, but it does line up with a 4K Aula. If Nintendo did plan a 4K Aula, then there's every reason to expect that dev kits would have been in third parties' hands by that point, particularly as the hardware required for them was plentiful. By the 2021 reporting some of the people he was talking to may have been talking about [redacted], and things got muddled between the two.

Regarding the 4K becoming a talking point, I expect that if the console was able to output 4K this would have absolutely been stated in the tech sheet, and even though I wouldn't expect games to render at native 4K, I didn't mean to say there wouldn't be games that render at somewhat higher than 1080p and leverage the 4K output (I mean, very few if any PS4 Pro games rendered in native 4K, but it didn't stop it from being the main selling point). I also think you're over-estimating the technical expertise of both mainstream journalists and many game developers. Most people didn't even know that the 2019 revision had a die-shrunk SoC (I've regularly seen references to it having a "bigger battery"), and from the point of view of the people working on it, I could absolutely see them saying that Aula had a new SoC, because Mariko was a new chip that allowed for higher performance over the original Switch. Whether it was already in use in the 2019 revision isn't something they necessarily knew or cared about.

Also, regarding dev kits, we know that ADEV development kits for Aula exist. Software, SDK and tools differences would have been relatively minor, as they're largely configuration changes in terms of supported clock speeds and output resolution, so the "cancellation" would have been a software update that removes the additional performance modes and 4K output, and leaves devs with a relatively boring Switch OLED model dev kit, so there's not much to show.



Yeah, I could definitely see there being a cross-contamination with early reports of [redacted]. I'm not really sure at what point the motherboard would have been finalised, but certainly there's a point where it's not worth going back and removing a handful of traces. I'd be curious if anyone with experience in electronics would have any idea of what the timeline is for finalising PCB designs for a device like this, as it might give us an idea of how late they were still considering 4K support.
Click to expand...
ADEV exists, but I think we can make a decently accurate guess, based on when rumors suddenly started talking about it, that it didn't really become generally available until summer 2021, close to the announcement of the hardware itself. There's probably a scenario where Bloomberg's sources were a mix of devs on the earliest inner circle track for Drake and the wider, but still early access, track for ADEV (which might have still had 4k output at that point), but I'm not sure that really materially changes the scenario much outside of explaining a few things. I still wouldn't discount the scenario where this played out mostly internally to Nintendo back in 2019, since early reports of the red box Switch seemed to hype it up as more than what we got, but that's not necessarily mutually exclusive with a scenario where the 4k output specifically manages to survive until pretty close to launch.

Regardless, none of this would contribute to any reports of cancellations in 2022 beyond providing some background information for rumors to form from.
 
Also forgive my ignorance, but is T239 considered part of the Tegra line? I haven't heard of anything being referred to as "Tegra" post-X2.

So I mean... if that person is right and Tegra has been dropped, could that mean Mariko has ended production in preparation for Drake to ramp up or..?
 
chocolate_supra said:
Also forgive my ignorance, but is T239 considered part of the Tegra line? I haven't heard of anything being referred to as "Tegra" post-X2.

So I mean... if that person is right and Tegra has been dropped, does that mean Mariko has ended production in preparation for Drake to ramp up or..?
Click to expand...
The T in T239 stands for Tegra

The person is not right for many reasons. Orin exists
 
I feel like with Nintendo's small game file sizes, UFS 4.0 with a dedicated decompresser feels like it could actually dramatically reduce load times.

Hope it happens.
 
0
Boxle said:
Sometimes I think about how ridiculous it would be if everything bad happened to actually be true.

OLED was originally a Pro, downgraded last minute. Switch Drake was going to be the next 'Pro' but was fully cancelled, not coming anymore ever. Nintendo doesn't know yet what their next console will be, they're still figuring things out after the previous two plans fell through. Likewise they have no plans for games for the rest of the year and are actually going to try and just coast until the Switch's 10 year mark because they saw people on Twitter say it's too soon for a new Switch and that it's doing just fine.

It would truly be the darkest timeline.
Click to expand...
If Nintendo made a major business decisions based on Twitter impressions and/or didn't have multiple backup plans for new hardware then they should just give up honestly. But the chance of them being that unprepared is 0
 
oldpuck said:
Well, there are connected limitations. More power in TV mode means more heat, which means a bigger fan and heat pipe, higher cost to ship, and heavier to hold, less room for battery, etcetera. And I have argued quite strongly that I think Nintendo will try to do what they did with the original Switch, which is match performance ratios to pixel ratios, then tweak from there as performance/battery life issues are found. It is both a sane starting point for the design and the likely natural end point of developing games and hardware simultaneously.

No judgement of some folk's desires/expectations of the device, but my (amateur) engineering mind can't make 1080p screen/4TF docked/OG Switch battery life/no significant form factor changes work, purely from a realm of physics. LPDDR5 has on paper greater efficiency than LPDDR4, but that power efficiency comes from long idle periods on the memory controller, which just don't happen in a gaming console. OLED screens are more power efficient on paper, but in practice, there are no savings on actual video games when comparing the power draw of the OLED vs the Redbox Switch.

With T239, Nintendo has a custom chip from the beginning, and they're stepping into the same environment that pushed Microsoft to make a Series S - they just don't believe that a cost reduced version of the base system is going to be available within the lifetime of the device. T239 will need to serve all of Nintendo's needs across the lifetime of the system and it's likely revisions.

We know Nintendo wasn't happy with Erista's battery life, and moved to customize it as soon as possible. I think we should look to Mariko more than Erista for Nintendo's battery life goals. I think we can look to the Switch for Nintendo's performance targets too.

Switch was 20% of a PS4. Nintendo knows that their device lives and dies on exclusives and their "unique way to play" and the Switch's Unique Way to Play was portable mode. Nintendo enabled a number of impossible ports by betting on a simple, flat architecture, with more modern features, and then just selling a ton of units to drive the economics to their side.

Nintendo doesn't need 4TF to make their next gen exclusives. 2.3TF docked would already be a massive 6x jump, plus DLSS and RT to add on top. Plenty for next-gen Mario, and put it in a slightly better league relative to a PS5 than the Switch was to a PS4. A 1TF handheld mode and a 720p screen would put it within spitting distance of the Steam Deck, while offering a much better feature set, and (with the power of ARM) a much more comfortable handheld experience with better battery life. And that performance gap would be very easy for developers to handle, unlike the Series S/X gap.

Note that this 2.25x gap is actually much larger than the current 1.7x gap between docked and handheld. I'm arguing for a larger gap. And I'm not arguing that these perf numbers represent some kind of ceiling. Nor does it represent my personal priorities. I just think in the game of "Fantasy Console Manufacturer" it reflects what, at least to me, seem like Nintendo's priorities as well as the current market, the realities of software development, and the laws of physics.

I expect the final numbers to reflect 1000s of manhours of hardware and software developers working in tandem to get all the details right, even more so than the original Switch. And I expect them to reflect at least one major, unaccounted for difference that we don't know about yet. And I of course expect to have gotten several things wrong!*

tl;dr: I think that REDACTED will have a bigger gap between handheld and docked mode than Switch, but I think for lots of reasons those two are still deeply connected. I don't see any major tech on the horizon that is likely to improve handheld battery life significantly that Nintendo/Nvidia could bet on, and I think the OG Switch battery life is something Nintendo would like to avoid repeating.

*One example might be a frequency scaling solution. I've speculated about this heavily before, but with the extreme range of CPU power required by cross-gen games versus some 9th gen exclusives, Nintendo might get a lot of mileage out of allowing more GPU perf in exchange for dropped CPU perf, or vice versa.
Click to expand...
Forgive me for not having a reply with quite as much detail, I don't mean any disrespect or anything I'm just super tired. To put it in as many words, I'd say I'm optimistic on performance, though I don't expect 4TF, I think sub-2TF would be pessimistic. That said, my expectations for the CPU are solidly "OK, could be better."
 
Reader mode
You must log in or register to reply here.
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