• 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!)

We don't know if it's as slim as Switch 1. What we're looking at could be only part of the shell of the console, ie just the front, or just the back, and when combined, it could be thicker.
Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
S99O3Vr.gif

Is there some kind of ranking?
Do I at least have a spot in the top 10?
 
Thank you everyone who responded to my question and tried to explain Ram speed/quantity. I tried my best to follow and I'm still lost due to the math not being basic lol

If only Ram math was like: 10x200=2000, 12x120= 1440, therefore 10x200>12x120
Oh, I misunderstood your question! I think I can explain using simple math! RAM math isn't much more complicated than that, you're just missing some of the numbers

Code:
RAM speed = (the number of RAM modules)x (how much data can the RAM move in a single chunk) x (the number of chunks it can transfer a second)

PS4 had 8 GB of RAM, but it used half-gig modules. That's 16 total modules.
Switch 2 has 12 GB of RAM, but it uses 6 GB modules. That's 2 modules.

Code:
PS4 RAM speed = (16) x (how much data can the RAM move in a single chunk) x (the number of chunks it can transfer a second)
Switch 2 speed = (2) x (how much data can the RAM move in a single chunk) x (the number of chunks it can transfer a second)

Those modules are connected to a bus that carries the data around in chunks. The wider the bus, the more data in each chunk. PS4 uses a pretty narrow 16 bit bus, Switch 2 uses a very wide 64 bit bus.

Code:
PS4 RAM speed = (16) x (16 bits) x (the number of chunks it can transfer a second)
Switch 2 speed = (2) x (64 bits) x (the number of chunks it can transfer a second)

A bit is the smallest unit of data possible - a one or a zero. There are 8 bits in a byte, because computer engineers hate us, and want us to be miserable. We can simplify our math a little though. Each RAM module, times the size of the bus, is how much total data the device can move at once.

Code:
16 bits = 2 bytes
2 bytes x 16 modules = 32 bytes,
PS4 RAM speed = (32 bytes) x (the number of chunks it can transfer a second)

64 bits = 8 bytes
8 bytes x 2 modules = 16 bytes a transfer
Switch 2 speed = (16 bytes) x (the number of chunks it can transfer a second)

So how many times, per second, can the RAM move this tiny little chunks over the bus? The RAM industry has developed multiple different kinds of RAM optimized for various situations. Switch 2 uses LPDDR - "low power double data rate" - a type optimized for power efficiency. PS4 used GDDR - "graphics double data rate" - a type optimized for speed, but extra expensive and power hungry.

In a twist, 10 years later, despite being "speed optimized", the Switch 2 RAM is faster. 7500 "mega transfers per second." Mega transfer meaning a million transfers, because all those zeros are annoying to write down. This is convenient for us, becuase we can multiply "mega transfers" times "bytes" to get "megabytes per second." The PS4 can do 5500 MT/s.

Code:
PS4 RAM speed = (32 bytes) x (5500 MT/s)
              = 32 x 5500 MB/s
              =176000 MB/s
              = 176 GB/s

Switch 2 speed = ( 16 bytes)  x (7500 MT/s)
               = 16 x 7500 MB/s
               = 120000 MB/
               = 120 GB/s
 
How do people feel about 256GB of storage - taking into account the potential size of third party games?

Reasonable for the price, which I hope it to be available at (~400 euro). Of course more is better, but with SD Express on its way and that being likely the standard for many other handhelds, mobile devices, I suspect the pricing to be reasonable for storage expansion throughout its lifetime.

We'll see around Summer how much micro SD Express will cost and how many suppliers are available, but my expectations are ~256GB for 100 euro. So that doubles the storage and apart from the huge games like COD, RDR2, NBA. Most 3rd party games should hopefully be around ~80GB.

Looking at an 256GB SD Express card: AData Premier Extreme, it costs around ~100 USD.
 
fucking

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.


how do they no. no. i must not fall into the old ways...
 
Well, more bandwidth is always more better, even if you don't strictly need it.

The one thing which has me thinking this points more towards 4N is that Nvidia has an LPDDR5X memory controller that they could use for this; the one used on Grace. Grace taped out at around the same time as T239, and is manufactured on 4N. Admittedly it's not impossible that they separately developed a LPDDR5X controller for 8nm (or any other manufacturing process), but it does line up with my prior expectations on 4N and LPDDR5X.

Correct, but when margins are low, they are less likely to splurge. LPDDR5x would be a luxury for T239 on 8nm, but on 4nm that increased bandwidth is necessary to prevent a bottleneck. Nothing is proven here of course, but I do see it as another indicator for 4nm.
 
How do people feel about 256GB of storage - taking into account the potential size of third party games?
Hard to say. I'm just concerned about external cards. I'd like to be able to hit 1tb of storage or at least close to it. Whether the speeds of those cards can match ufs 3.1 is up in the air at this point, whether that even matters is also up in the air lol.
 
0
I hope we can catch more display shipments next month, It is the last (?) big piece missing to complete the puzzle.
We saw those 8" Innolux LCD panels at nvidia India (i think your finding maybe? edit: yup).

Only problem is we don't know whether those panels were for automotive usage, we couldn't tie those to T239 reliably.

Yes, the last missisng piece. LiC found a mention of touch screen in one of those NL-AMxx entries, but it had no other info (no part number or manufacturer name). We need clock speeds as well.
 
I think this is it?
That's probably it. Full credit, @Look over there - who I haven't seen in a while, I hope they're all right - they did the majority of the analysis a long time ago, I just did the last little bit of work and wrote it up.
 
Since I saw this brought up multiple times, what's the difference in simple terms between 4nm and 8nm? I kinda get the building process is smaller but I don't know anything more than that (if even right lol)
Mostly heat and energy.
Smaller node means it uses less of both for the same performance.
Important for a portable device.
 
Since I saw this brought up multiple times, what's the difference in simple terms between 4nm and 8nm? I kinda get the building process is smaller but I don't know anything more than that (if even right lol)
4nm is more power efficient than 8nm. More performance for the same wattage, or the same performance at lower wattage. Also heat.
 
Since I saw this brought up multiple times, what's the difference in simple terms between 4nm and 8nm? I kinda get the building process is smaller but I don't know anything more than that (if even right lol)
With everything else being equal, 8nm would draw power at about 2x more compared to 4nm.

4nm would be better for battery life. Can use smaller battery for same battery life as a 8nm one (but requiring bigger battery).
 
something crazy to think about. we're not far from this thread closure, what announcements would bring this thread to a close? Or maybe we'll have to wait for someone to open the console and see what's inside?
Eh, we're farther away than we think, I feel. I think the thread - for organizational purposes; it's gonna be a 3k+ page thread with continuous conversation come launch day, after all - may close a few months after the system launches. By then we'll have first impressions, some technical breakdown from Digital Foundry or our own members (depending on who's first lol), and we'll be able to discuss any sort of technical showcases and whatnot since some games will be out by then. It would give us some breathing room to actually... talk about the thing we've been speculating over. Look back on every little detail we got right, and what we got wrong. Geek out over raytraced Wave Racer or F-Zero or something. So on. I don't think it'd be right to just shut this thread down once the system launches, especially given how much history - for lack of a better term - this thread has.

From there, I can see two new paths/successors to this thread: first one being a new hardware speculation thread. Switch 2 Pro Deluxe Sonic & Knuckles rumors will start flying shortly after launch, if the original Switch is anything to go by. Though a lot of us will probably not be in a rush to start this whole process over again, either due to the sheer amount of time we've put into Switch 2 speculation or because we're too busy putting too many hours into our fancy new Switch 2s. The second way forward could be a "Switch 2 Tech Discussion Thread" that would carry on the super nerdy technical spirit of this thread. It would be a fun way to have a centralized spot for the nerdiest of the nerds to explode over how nice the RTAO implementation in the next Zelda game is.
 
Since I saw this brought up multiple times, what's the difference in simple terms between 4nm and 8nm? I kinda get the building process is smaller but I don't know anything more than that (if even right lol)
If you compare the launch Switch with the 2019/Lite/OLED model, the former uses a 20 nm process and the latter 16 nm. This allowed for shrinking the device (Lite), and better battery life for the same performance while drawing less energy. And when overclocking those newer devices, much better performance than the launch Switch while sustaining reasonable heat and battery life.

Instead of 20 and 14 now we're comparing two newer processes, 8 and 4. But the idea is the same, 4 provides the power efficiency advantage.
 
Last edited:
Mostly heat and energy.
Smaller node means it uses less of both for the same performance.
Important for a portable device.
4nm is more power efficient than 8nm. More performance for the same wattage, or the same performance at lower wattage. Also heat.
With everything else being equal, 8nm would draw power at about 2x more compared to 4nm.

4nm would be better for battery life. Can use smaller battery for same battery life as a 8nm one (but requiring bigger battery).
Thanks to all! Now I get it, so if they achieved a 4nm chip (pardon my wrong tech terms) that's pretty big right?
 
We saw those 8" Innolux LCD panels at nvidia India (i think your finding maybe? edit: yup).

Only problem is we don't know whether those panels were for automotive usage, we couldn't tie those to T239 reliably.

Yes, the last missisng piece. LiC found a mention of touch screen in one of those NL-AMxx entries, but it had no other info (no part number or manufacturer name). We need clock speeds as well.
Yup

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
Please give us Xenoblade 1 through 3 first with all the settings maxed out, it would be absolutely dire to get this beast of a machine but for those games to be trapped in their current form. Let the shackles be released.
Past the obvious “able to play next-gen games” aspect of the successor console, this is legit all I’m wanting out of the Switch 2. Give us enhanced Switch 1 games running at 1440p/4K (and 60fps patches if I can be greedy) (in other words, basically what Xbox did for their BC program) and I won’t even care what the price of the new console is. Games like the Xenoblades and Pikmin 4 are too beautiful to be running at sub-HD resolutions.
 
umm...No?
Respectfully, your whole post has nothing to do with the conversation we were having. The question was "will the core count difference matter in the quality of the ray tracing." The answer is objectively and clearly "yes." Will those disadvantages be overcome by advantages elsewhere? In some cases yes, which you spend an awfully long time trying to say yourself - ignoring my lengthy post on the subject, one step back in the conversation.

I don't think your technical analysis holds up, for the most part, but I also don't think it's worth diving into. If every ray traced port runs at the same resolution and fidelity as the PS5 version, you know what, I'll take the L.
 
If you compare the launch Switch with the 2019/Lite/OLED model, the former uses a 20 nm process and the latter 14 nm. This allowed for shrinking the device (Lite), and better battery life for the same performance while drawing less energy. And when overclocking those newer devices, much better performance than the launch Switch while sustaining reasonable heat and battery life.

Instead of 20 and 14 now we're comparing two newer processes, 8 and 4. But the idea is the same, 4 provides the power efficiency advantage.
OOOOH, so that's why. The "magic" about Mariko was simply a smaller nm process. Thanks!
 
0
the node thing seems like a big deal because it's not like the smaller process is only slightly more efficient...it's double. feels like it would be so much better for portable hybrid device given everything we know.
 
How do people feel about 256GB of storage - taking into account the potential size of third party games?
It's fine, about 4x OLED capacity and 8x OG base model.

Leaves room for the inevitable 512 GB premium OLED model down the line.

It will be enough to fit probably 1 AAA (or several Nintendo) and a few live service games for new owners which is what is im thinking they are aiming for with that capacity
 
Thanks to all! Now I get it, so if they achieved a 4nm chip (pardon my wrong tech terms) that's pretty big right?
Yeah, TSMC 4N would be everything we hoped for.

If you have a chance to, read this excellent writeup by Thraktor. From nvidia leaks, we know it's 12SMs. 8nm and 12SMs just doesn't seem to math (physic) well with each other, we find 4N likely.

And the reason why it's mainly debate between SEC8N vs TSMC 4N (and not other nodes between) is because nvidia products (in the current time frame) have overwhelmingly been fabbed on either SEC8N or TSMC 4N. The Switch 2 SoC, T239, is likely to be either one of those given what nvidia had their other products fabbed on.
 
Respectfully, your whole post has nothing to do with the conversation we were having. The question was "will the core count difference matter in the quality of the ray tracing." The answer is objectively and clearly "yes." Will those disadvantages be overcome by advantages elsewhere? In some cases yes, which you spend an awfully long time trying to say yourself - ignoring my lengthy post on the subject, one step back in the conversation.

I don't think your technical analysis holds up, for the most part, but I also don't think it's worth diving into. If every ray traced port runs at the same resolution and fidelity as the PS5 version, you know what, I'll take the L.
I was mainly talking about vs Series S.

Vs PS5 is where balancing the RT acceleration difference, DLSS, and Ray Reconstruction combined with other optimizations like proper utilization of INT32 and Mixed Precision with remaining Tensor Core resources would matter the absolute most.

Theoretically it's possible to overcome PS5's output (due to how much of a deficit PS5 takes on resources to do true RT) but it very much depends on full utilization of T239 to achieve in theory
 
0
Yup

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.

I'm just surprised that the super dense essay has more reactions than the post with the RAM and storage.

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
Well, more bandwidth is always more better, even if you don't strictly need it.

The one thing which has me thinking this points more towards 4N is that Nvidia has an LPDDR5X memory controller that they could use for this; the one used on Grace. Grace taped out at around the same time as T239, and is manufactured on 4N. Admittedly it's not impossible that they separately developed a LPDDR5X controller for 8nm (or any other manufacturing process), but it does line up with my prior expectations on 4N and LPDDR5X.
Do we have any proof that T239 is 4N? 4N seems much more cutting edge than Nintendo has used in decades. Going from 16nm to 4nm would be monumental
 
Respectfully, your whole post has nothing to do with the conversation we were having. The question was "will the core count difference matter in the quality of the ray tracing." The answer is objectively and clearly "yes." Will those disadvantages be overcome by advantages elsewhere? In some cases yes, which you spend an awfully long time trying to say yourself - ignoring my lengthy post on the subject, one step back in the conversation.

I don't think your technical analysis holds up, for the most part, but I also don't think it's worth diving into. If every ray traced port runs at the same resolution and fidelity as the PS5 version, you know what, I'll take the L.
If we go by oldpuck's analysis from yesterday, can we expect better visuals from ray tracing even though the number of cuda cores is less than the number of shaders in xss?
 
0
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