• 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
oldpuck said:
That's what most general purposes OSes do, and I assume Horizon (the switch OS) does the same. This is one fo the reasons that the Switch was so nice to port to - the CPUs all behave the same, so devs don't have to do things like "this CPU runs a faster clock, but this has a larger cache, so I need to put physics on this CPU and then optimize for cache misses..."

When you're expecting to run on 3 cores, then you probably set your thread pool to a small number of threads. It can be much faster to manually have your threads jump between multiple tasks, than make a thread for each task and have the OS try and schedule them for you. So running Switch games on lots of cores is less likely to give you the perf boost of the same number of cores but better per-core clock speed. Though that is not a universal truth, and once you're talking about both clock AND core increases, the tradeoffs become less obvious.

Also, I'm not actually sure how devs request performance profiles from the hardware. @davec00ke might be right that to access better clock speeds might always require a patch, though potentially a trivial one.
Click to expand...
If I'm interpreting Switchbrew correctly, games don't set the clocks directly, they pick them from a list (specifically this one). I don't see anything preventing the OS from interpreting these pretty liberally when it knows a game is running in BC mode.
ReddDreadtheLead said:
A game would only recognize the amount of threads it was coded for afaik.

So BoTW would only see 3 threads/cores

Not 5 or 7.

XBox games (and presumably PS5 too) also disable their newer features and set a clock for the BC titles until they get patched to make use of the newer features. 3.8GHz in the Series X case.


i wonder if Nintendo will give devs options for 3, 5 and 7 cores though…. With their own set speeds.

would be interesting to see what devs opt for…
Click to expand...
It depends on how the game is programmed. Certainly, console games are a lot more likely to make hard assumptions about the number of available threads than most other types of software, but it's by no means a guarantee.
 
ReddDreadtheLead said:
A game would only recognize the amount of threads it was coded for afaik.

So BoTW would only see 3 threads/cores

Not 5 or 7.

XBox games (and presumably PS5 too) also disable their newer features and set a clock for the BC titles until they get patched to make use of the newer features. 3.8GHz in the Series X case.


i wonder if Nintendo will give devs options for 3, 5 and 7 cores though…. With their own set speeds.

would be interesting to see what devs opt for…
Click to expand...
It will interesting how last gen xbone/ps4 and even switch ports (simultaneously releasing at the same time as switch 2) will be handled on Switch 2. Especially if Switch 2 gets 8 cores...

I imagine if we get ps4 and switch ports for both switch and switch 2, only 3 cores will be used for both switch and switch 2 consoles to make developing games easier (vs 3 cores and 7 cores) Since A78s are 3x more performant per clock per GHz than a57s and jaguars, something like 1.2-1.5Ghz should theoretically have no problem with CPU bottlenecked ps4 games. 1Ghz will be pretty close too. But it we get switch 2 exclusive games, then 7 usable cores should be expected for most games.
 
0
I think ideally Nintendo would give some control to the developers. Let them choose if they want 4 cores at 2ghz and 3 cores at 1.2ghz or 7 cores at 1.65ghz (for example). This is presuming 1 core is reserved for the OS.
 
Off the shelf, the version of the DynamIQ Shared Unit used with the A78C is limited to 8 MB of L3 cache, yes.
Of course, a little over half a year later, along with the rest of the cores introduced with ARM v9, the DSU got redesigned* for a new max of 16 MB.
And who knows, Nvidia may do their own thing and may not necessarily be limited to 8 MB.

*interesting that ARM went into more detail with describing the structure of this DSU-110, as opposed to the previous version(s). For the older versions, I assume that the L3 cache is all one chunk? Or multiple segments that make up one chunk? The DSU as introduced with the A75 was limited to 1/2/4 MB of L3 cache and was able to be partitioned into a max of 4 groups.
There's like no detail about the L3 cache in the version of the DSU packaged with A78C beyond 'max is now 8 MB'.
For the DSU-110, it's explicitly described as being divided up into a max of 8 slices, kinda sorta (but not exactly...) like how Intel CPUs have their L3 cache divided into slices.
 
Nintendo and Nvidia will have to go with 8 cores, or at least 6. In practical terms, the Switch is a 3-core system because they reserve a whole core for the OS. This is an important point to considerate.

As others said before, losing 1 core out of 4 is 25%, but 1 out of 8 is 12.5%. The impact on performance is too much for a modern system that has to keep up with technological advances for at least the next 5-7 years.

And while it’s true that single core performance keeps being a relevant part of gaming performance, most modern games benefit a lot more from multicore cpus compared to games from the start of the PS4/XBO generation. Eventually devs will end up taking advantage of having access to 2 powerful consoles with SMT (8c/16t) available.

So apart from the core count, we have to consider the thread count of the CPU. According to Arm’s website the A78C doesn’t support SMT:
High-performance computing with support for up to eight big-core-only cluster and up to 8MB L3 cache running eight threads in parallel
Click to expand...
(@ReddDreadtheLead it seems it’s not possible to have 16MB of L3 unless they go for more cores). Edit: just saw that Dakhil and Look over there answered you already.

Let’s do a little bit of light research about CPU performance:

In this video by LTT:

They show that 4 cores keep giving a competent performance in gaming, and that it should be enough for the near future. But they kinda flunked their metodology by only testing 4c/8t CPUs, so in reality they show that 8 threads is the minimal floor for future-proofing the device.

Hardware Unboxed arrives at a similar conclusion by testing the 4c/8t i3 10105F against stronger CPUs with disabled cores (and all of them have HT). 8 threads is the minimum going forward. Big caches help too.



According to the Steam Survey 4c and 6c have almost equal marketshare (no word on # of threads), and 4c have been declining in % during the pandemic. Probably most 4c/Xt are older systems that are getting replaced, but it’s just speculation on my part.
Sadly there aren’t a lot of recent videos comparing 2c vs 4c vs +6c; this next video is a bit older (from 2017) and at least show a more varied selection of games:


The bench has caveats: they use a Core i5-8600K (6c/6t) and disable pairs of cores to equalize clocks, but they decided to do a CPU benchmark with a GTX 1060 and all settings on High/Ultra… 🤦🏻‍♂️
We can still learn a few things despite the GTX 1060 bottlenecking hard most of the time:
  • In all games the 2c CPU just can’t keep up.
  • In most games the 6c CPU is not pushed as hard as the 4c despite the GPU bottleneck.
  • In those cases the 6c CPU is at 70-80% utilization while the 4c is at >90%.
  • The 6c losing 1 core for the OS (16,7%) shouldn’t compromise performance that much.
  • The 4c CPU losing 1 core (25%) would definitely lose a lot of performance because it is at it’s limits.
So, TLDR, I think Nintendo and Nvidia’s next system won’t have 2 cores, shouldn’t have 4 cores, could have 6 cores, and should have 8 cores (and 8 threads). And it probably won’t have more than 8 cores.
 
Falcatops said:
Nintendo and Nvidia will have to go with 8 cores, or at least 6. In practical terms, the Switch is a 3-core system because they reserve a whole core for the OS. This is an important point to considerate.

As others said before, losing 1 core out of 4 is 25%, but 1 out of 8 is 12.5%. The impact on performance is too much for a modern system that has to keep up with technological advances for at least the next 5-7 years.

And while it’s true that single core performance keeps being a relevant part of gaming performance, most modern games benefit a lot more from multicore cpus compared to games from the start of the PS4/XBO generation. Eventually devs will end up taking advantage of having access to 2 powerful consoles with SMT (8c/16t) available.

So apart from the core count, we have to consider the thread count of the CPU. According to Arm’s website the A78C doesn’t support SMT:

(@ReddDreadtheLead it seems it’s not possible to have 16MB of L3 unless they go for more cores). Edit: just saw that Dakhil and Look over there answered you already.

Let’s do a little bit of light research about CPU performance:

In this video by LTT:

They show that 4 cores keep giving a competent performance in gaming, and that it should be enough for the near future. But they kinda flunked their metodology by only testing 4c/8t CPUs, so in reality they show that 8 threads is the minimal floor for future-proofing the device.

Hardware Unboxed arrives at a similar conclusion by testing the 4c/8t i3 10105F against stronger CPUs with disabled cores (and all of them have HT). 8 threads is the minimum going forward. Big caches help too.



According to the Steam Survey 4c and 6c have almost equal marketshare (no word on # of threads), and 4c have been declining in % during the pandemic. Probably most 4c/Xt are older systems that are getting replaced, but it’s just speculation on my part.
Sadly there aren’t a lot of recent videos comparing 2c vs 4c vs +6c; this next video is a bit older (from 2017) and at least show a more varied selection of games:


The bench has caveats: they use a Core i5-8600K (6c/6t) and disable pairs of cores to equalize clocks, but they decided to do a CPU benchmark with a GTX 1060 and all settings on High/Ultra… 🤦🏻‍♂️
We can still learn a few things despite the GTX 1060 bottlenecking hard most of the time:
  • In all games the 2c CPU just can’t keep up.
  • In most games the 6c CPU is not pushed as hard as the 4c despite the GPU bottleneck.
  • In those cases the 6c CPU is at 70-80% utilization while the 4c is at >90%.
  • The 6c losing 1 core for the OS (16,7%) shouldn’t compromise performance that much.
  • The 4c CPU losing 1 core (25%) would definitely lose a lot of performance because it is at it’s limits.
So, TLDR, I think Nintendo and Nvidia’s next system won’t have 2 cores, shouldn’t have 4 cores, could have 6 cores, and should have 8 cores (and 8 threads). And it probably won’t have more than 8 cores.
Click to expand...

Too bad it won't have 16 threads? Interesting how different ARM and AMD CPU architecture are. Not exactly sure how the threading works, but since A78s are as performant as the cpus on steam deck and current gen per core per Ghz, multi tasks threading gives amd the advantage in that case? Someone correct me/clarify please.
 
Falcatops said:
Nintendo and Nvidia will have to go with 8 cores, or at least 6. In practical terms, the Switch is a 3-core system because they reserve a whole core for the OS. This is an important point to considerate.

As others said before, losing 1 core out of 4 is 25%, but 1 out of 8 is 12.5%. The impact on performance is too much for a modern system that has to keep up with technological advances for at least the next 5-7 years.

And while it’s true that single core performance keeps being a relevant part of gaming performance, most modern games benefit a lot more from multicore cpus compared to games from the start of the PS4/XBO generation. Eventually devs will end up taking advantage of having access to 2 powerful consoles with SMT (8c/16t) available.

So apart from the core count, we have to consider the thread count of the CPU. According to Arm’s website the A78C doesn’t support SMT:

(@ReddDreadtheLead it seems it’s not possible to have 16MB of L3 unless they go for more cores). Edit: just saw that Dakhil and Look over there answered you already.

Let’s do a little bit of light research about CPU performance:

In this video by LTT:

They show that 4 cores keep giving a competent performance in gaming, and that it should be enough for the near future. But they kinda flunked their metodology by only testing 4c/8t CPUs, so in reality they show that 8 threads is the minimal floor for future-proofing the device.

Hardware Unboxed arrives at a similar conclusion by testing the 4c/8t i3 10105F against stronger CPUs with disabled cores (and all of them have HT). 8 threads is the minimum going forward. Big caches help too.



According to the Steam Survey 4c and 6c have almost equal marketshare (no word on # of threads), and 4c have been declining in % during the pandemic. Probably most 4c/Xt are older systems that are getting replaced, but it’s just speculation on my part.
Sadly there aren’t a lot of recent videos comparing 2c vs 4c vs +6c; this next video is a bit older (from 2017) and at least show a more varied selection of games:


The bench has caveats: they use a Core i5-8600K (6c/6t) and disable pairs of cores to equalize clocks, but they decided to do a CPU benchmark with a GTX 1060 and all settings on High/Ultra… 🤦🏻‍♂️
We can still learn a few things despite the GTX 1060 bottlenecking hard most of the time:
  • In all games the 2c CPU just can’t keep up.
  • In most games the 6c CPU is not pushed as hard as the 4c despite the GPU bottleneck.
  • In those cases the 6c CPU is at 70-80% utilization while the 4c is at >90%.
  • The 6c losing 1 core for the OS (16,7%) shouldn’t compromise performance that much.
  • The 4c CPU losing 1 core (25%) would definitely lose a lot of performance because it is at it’s limits.
So, TLDR, I think Nintendo and Nvidia’s next system won’t have 2 cores, shouldn’t have 4 cores, could have 6 cores, and should have 8 cores (and 8 threads). And it probably won’t have more than 8 cores.
Click to expand...

try 1 core and 1 thread!

 
Falcatops said:
Nintendo and Nvidia will have to go with 8 cores, or at least 6. In practical terms, the Switch is a 3-core system because they reserve a whole core for the OS. This is an important point to considerate.
Click to expand...
While I think more cores is better for exclusive development, I think if this device is truly intended as revision, they will match the number of cores, and simply provide faster clocks. Otherwise you're paying for 4 cores physically that <10 games can use.
 
ShadowFox08 said:
Too bad it won't have 16 threads? Interesting how different ARM and AMD CPU architecture are. Not exactly sure how the threading works, but since A78s are as performant as the cpus on steam deck and current gen per core per Ghz, multi tasks threading gives amd the advantage in that case? Someone correct me/clarify please.
Click to expand...
It’s simply ARM vs x86_64 CPUs

ARM for the most part doesn’t do SMT, x86_64 CPUs these days for the most part almost never lack SMT. It’s special cases that have equal core to thread count.

Falcatops said:
Nintendo and Nvidia will have to go with 8 cores, or at least 6. In practical terms, the Switch is a 3-core system because they reserve a whole core for the OS. This is an important point to considerate.

As others said before, losing 1 core out of 4 is 25%, but 1 out of 8 is 12.5%. The impact on performance is too much for a modern system that has to keep up with technological advances for at least the next 5-7 years.

And while it’s true that single core performance keeps being a relevant part of gaming performance, most modern games benefit a lot more from multicore cpus compared to games from the start of the PS4/XBO generation. Eventually devs will end up taking advantage of having access to 2 powerful consoles with SMT (8c/16t) available.

So apart from the core count, we have to consider the thread count of the CPU. According to Arm’s website the A78C doesn’t support SMT:

(@ReddDreadtheLead it seems it’s not possible to have 16MB of L3 unless they go for more cores). Edit: just saw that Dakhil and Look over there answered you already.

Let’s do a little bit of light research about CPU performance:

In this video by LTT:

They show that 4 cores keep giving a competent performance in gaming, and that it should be enough for the near future. But they kinda flunked their metodology by only testing 4c/8t CPUs, so in reality they show that 8 threads is the minimal floor for future-proofing the device.

Hardware Unboxed arrives at a similar conclusion by testing the 4c/8t i3 10105F against stronger CPUs with disabled cores (and all of them have HT). 8 threads is the minimum going forward. Big caches help too.



According to the Steam Survey 4c and 6c have almost equal marketshare (no word on # of threads), and 4c have been declining in % during the pandemic. Probably most 4c/Xt are older systems that are getting replaced, but it’s just speculation on my part.
Sadly there aren’t a lot of recent videos comparing 2c vs 4c vs +6c; this next video is a bit older (from 2017) and at least show a more varied selection of games:


The bench has caveats: they use a Core i5-8600K (6c/6t) and disable pairs of cores to equalize clocks, but they decided to do a CPU benchmark with a GTX 1060 and all settings on High/Ultra… 🤦🏻‍♂️
We can still learn a few things despite the GTX 1060 bottlenecking hard most of the time:
  • In all games the 2c CPU just can’t keep up.
  • In most games the 6c CPU is not pushed as hard as the 4c despite the GPU bottleneck.
  • In those cases the 6c CPU is at 70-80% utilization while the 4c is at >90%.
  • The 6c losing 1 core for the OS (16,7%) shouldn’t compromise performance that much.
  • The 4c CPU losing 1 core (25%) would definitely lose a lot of performance because it is at it’s limits.
So, TLDR, I think Nintendo and Nvidia’s next system won’t have 2 cores, shouldn’t have 4 cores, could have 6 cores, and should have 8 cores (and 8 threads). And it probably won’t have more than 8 cores.
Click to expand...

A 4c/8T is roughly equivalent to what a 6C/6T CPU is in terms of perf. A 8C/16T is roughly equivalent to a 10c/10T CPU. The scaling vs a cpu that doesn’t do smt is smaller the larger it is, but admittedly it depends on the application that’s being processed. Some take SMT better and can give you a 50% perf uplift, others do a 5-10% uplift in performance. Others get a 35% uplift in performance.

And I don’t think this comparison really makes sense in the context of what the switch is. While for anyone else, especially those with x86 CPUs, should opt for more threads and cores as they target the audience that would make proper use of this.

We have to way what they really need and want.


oldpuck said:
While I think more cores is better for exclusive development, I think if this device is truly intended as revision, they will match the number of cores, and simply provide faster clocks. Otherwise you're paying for 4 cores physically that <10 games can use.
Click to expand...
I don’t really think Nintendo will not upgrade the CPU core count. They’ve been willing to upgrade the core count when the CPU becomes a concern from the Wii to Wii U, even if they didn’t change it really from the GCN. Or 2 cores to 4 core with the 3DS that was meant for another purpose.

Even if few games utilize it.
 
ShadowFox08 said:
Too bad it won't have 16 threads? Interesting how different ARM and AMD CPU architecture are. Not exactly sure how the threading works, but since A78s are as performant as the cpus on steam deck and current gen per core per Ghz, multi tasks threading gives amd the advantage in that case? Someone correct me/clarify please.
Click to expand...
SMT (Simultaneous Multi-Threading) can be advantageous, depending on the workload.

Conceptually, the idea of multiple threads per core came from the observation that a single thread doesn't necessarily utilize all of a core's resources. So with the cost of a small bit of silicon and power, you can have the core run another thread to try to maximize your usage of resources. Well, when I say small, it's from the perspective of the bigger x86 cores; AMD's Zen and Intel's Core cores. ARM's cores as well as Intel's Atom cores don't use SMT/HT (HyperThreading's the Intel version).

But remember, you're splitting up resources. You're not going over 100%. A 1*X core, 2*X thread CPU cannot match up against 2*X core, 2*X thread (of equal architecture and clock) in a workload asking for 2*X threads, unless the single thread utilization is utterly abysmal.

Also, each workload asks for a different number of threads. Sometimes you need so few threads such that SMT makes practically no difference. Other times, it can be pretty nice to have.
I say this in the context of software used by the general audience in the present: most games are still not super high thread count workloads. A lot of the stuff done today that does ask for tons of threads are typically not run on dedicated gaming devices.

Can it change eventually with the PS5/Series generation? Sure. But that's probably stuff of high enough complexity to not expect a 2022-2023 era mobile device to comfortably handle.
(new desktops that you want to last a while on the other hand...)
 
0
Skittzo said:
.
Click to expand...
Look over there said:
Off the shelf, the version of the DynamIQ Shared Unit used with the A78C is limited to 8 MB of L3 cache, yes.
Of course, a little over half a year later, along with the rest of the cores introduced with ARM v9, the DSU got redesigned* for a new max of 16 MB.
And who knows, Nvidia may do their own thing and may not necessarily be limited to 8 MB.
Click to expand...
if they do their own thing, aren’t they more likely to cut rather than add, as is typical for game consoles?
 
That could happen, yes, though I'm not sure if that can be generalized to be typical for consoles?

Looking at the Jetson AGX Orin... 3 separate 4xA78 clusters with 2 MB of L3 each, so for that at least, it seems that Nvidia opted for the middle option (out of 1, 2, and 4 MB for a given cluster). But there is that 4 MB sitting on top as a cache for both CPU and GPU (so, L4 from the CPU's perspective). So... who knows.
 
Look over there said:
That could happen, yes, though I'm not sure if that can be generalized to be typical for consoles?

Looking at the Jetson AGX Orin... 3 separate 4xA78 clusters with 2 MB of L3 each, so for that at least, it seems that Nvidia opted for the middle option (out of 1, 2, and 4 MB for a given cluster). But there is that 4 MB sitting on top as a cache for both CPU and GPU (so, L4 from the CPU's perspective). So... who knows.
Click to expand...
Pretty sure ps5 and series got significantly less cache compared to off the shelf zen 2.
 
0
That is true*
Standard chiplet-based desktop Zen 2 CPUs (that is, without integrated graphics) have 16 MB of L3 cache per CCX.
Monolithic Zen 2 CPUs with integrated graphics (or, what AMD refers to as APUs) have 4 MB of L3 cache per CCX. They're the desktop SKUs that end with a G, as well as all the laptop SKUs, IIRC.

If you're looking through the lens of regular chiplet-based Zen 2, yes, the PS5 and Series have reduced L3 cache.
If you're looking through the lens of Zen 2 APUs, the PS5 and Series look like... well, just another Zen 2 APU, cache-wise.
 
Thanks for saying what SMT means. I can guess from the context that it isn't about Shin Megami Tensei, but having it spelled out is nice. lol

Kind of going OT now, but it doesn't actually matter how Nintendo positions the next hardware, what matters is what it is. Some games have Remaster in their title despite being remakes, yet the title doesn't make them any less of a remake. (On a side note, I wish the industry would stick to one definition, because the current naming scheme just muddies the waters. Then you get takes like "Skyward Sword HD is a lazy remake" or "Link's Awakening for Switch is just a port". It doesn't help that FFVIIR, and to a lesser extent RE2, seem to have changed the definition of remake, while people also conveniently ignore that the former is releasing in parts.)
The illegal leaks hint at the interior being completely different, and these discussions seem to further prove that, something which usually isn't done with a revision, and wasn't done with the PS4 Pro, the revision the Switch Pro was named after. (Curious that a PS naming scheme was chosen and not a Nintendo one.) I knew that if the Pro wouldn't pan out, the narrative would switch to "Oops, I guess it's the successor now" in an attempt to backpedal and that was basically confirmed once the same people pedalling the Switch Pro for years started saying things like "It doesn't matter whether it's the Switch Pro or Switch 2". Of course it matters. I'm definitely not going to spend money to buy the same console again, but I'd buy a successor. Since someone has mentioned it on here, but I doubt all the people who mainly play ACNH or haven't touched their Switch much since they stopped playing it, would go out in droves to buy a better Switch, just because their island chucks here and there.
Third party exclusives also gain a different meaning depending on what it is. In the case of a revision the question would be if third parties would neglect such a massive install base. Even if we were to assume that the Switch's install base is highly inflated, because everyone owns two, that'd be a potential install base of 50+ million. I could see Nintendo trying to give incentive to some third parties to help add value to the revision, perhaps even helping with distribution etc., or third parties porting their cloud-only games to it, but the fact also remains that the PS4 Pro and XBSX didn't light the sales charts on fire, losing out to the cheaper slim models, from what I've heard. In the case of a successor, exclusive third party games can only get a "No shit, Sherlock".
Since people like to say that it depends on "marketing", if the Switch was marketed as a Wii U revision, you'd just accept that without question?
I've been thinking for a while now that a successor around 2023 with a cross-gen period for a year or more would make sense, but I'm also usually baffled by the people, who can't wait to spend money on a new console. lol
 
0
Hermii said:
if they do their own thing, aren’t they more likely to cut rather than add, as is typical for game consoles?
Click to expand...
Look over there said:
That could happen, yes, though I'm not sure if that can be generalized to be typical for consoles?

Looking at the Jetson AGX Orin... 3 separate 4xA78 clusters with 2 MB of L3 each, so for that at least, it seems that Nvidia opted for the middle option (out of 1, 2, and 4 MB for a given cluster). But there is that 4 MB sitting on top as a cache for both CPU and GPU (so, L4 from the CPU's perspective). So... who knows.
Click to expand...
The reason for those cuts on the consoles is probably not really applicable for PCs really. I think Consoles go by a perf-per-dollar while PC is perf-per-watt model. They make necessary cuts to reduce cost and increase yield for larger mass manufacturing of the die, as that is what you get for several years. As the platform gets older, the parts depreciate and becomes cheaper to produce and thus cheaper for a consumer.

With a PC part, if it’s binned it is still a sellable product and they can sell you the whole thing in some fashion or form, without necessitating the need to discard said silicon.

Previous consoles which (off the top of my head) that made cuts besides the series and PS5, were the 360 and PS3. They cut out the OoOE, and were In-order, but added other features to them. 360 had that extra 10MB of memory and the vector unit, PS3 had the infamous CELL processor which was brilliant in concept but poor in its execution.

PS5 and SX/S cuts are more likely related to not being too big to make it difficult to produce (would have been around 400mm^2 I think).

With an ARM CPU which are significantly smaller than the Zen2 CPUs, this is less of a concern with making a viable product with the die size than said Zen2 based product, but it is still a concern nonetheless and can increase cost. ORIN is like a 400-500mm^2 die.

And has that noticeable more robust caching system (plus the other stuff).
 
0
CypressFX said:
Nate any update on your NvN2 issue in regards to the podcast?

Do you have maybe some new info about release time frame or is there no new update on this?
Click to expand...

We need to help him find that last confirmation/puzzle piece =)
 
Last edited:
0
#Salt said:
Could the new Switch use UFS 4.0, which is set to start production by Q3 this year. Or would it be too new/expensive to use. https://www.gsmarena.com/samsung_an...eds_and_better_powerefficiency-news-54184.php
Click to expand...
Personally, I don't think the switch will even use UFS 3. Likely just UFS 2.1 or EMMC.

The original switch had EMMC that could operate at 350MB/s, the cartridges reportedly ran at 25MB/s and the SD Card up to 130MB/s and yet loading from different media made little difference in load times because the system was CPU / Memory bandwidth starved.

Whilst those problems could potentially be rectified or reduced with the Drake model, there will still be an upper limit to how fast data can get pushed from storage and into memory on that platform. They also won't go too exotic with the internal storage if the cartridges and external storage can't keep up. Too large a performance delta between each storage medium will take away from the switch experience.

So IMO, UFS 2.1 or eMMC is most likely.
 
0
I wouldn't call UFS 3.0 "exotic"

the best course of action is still to allow embedded memory to reach its peak speeds. making digital and physical equal hurts everyone, from developers to customers
 
0
Agree they won't arbitrarily limit the speed of embedded memory but it will come down to cost vs benefit as well.

Let's say we get UFS card as external storage, 500MB/s, the cartridges some how manage to get up to 250MB/s and we get UFS 2.1 256GB at around 800MB/s, Nintendo tests it and in real world performance loading games the cartridges load maybe two seconds slower than UFS card and UFS 2.1 is a second faster.

Nintendo then tries UFS 3 and it loads another second faster on average, but the cost of UFS 3 means they can only go with 128GB storage at the price point they are targeting.

Will they go with UFS 2.1 or 3? Which will the consumer care about more? The small decrease in load times or twice the storage?

These are the kinds of things they will take into consideration when building a product for mass market. If the price difference between UFS 3 isn't much they may just go UFS 3 for the sake of it, but if its big enough that they could increase storage at a given price point using UFS 2.1 I see them using the latter.
 
Last edited:
At this rate I feel like 2022 is off the table with more and more certainty as the days pass. I know people got burned last year and people took a lot of heat so they may be more hesitant to speak this year, but I still feel like we’d be hearing something from others if this was a 2022 device. And I don’t think Mochizuki is deterred by blowback, he’s a business journalist, he doesn’t care what Twitter/forums think.

I’m not sure how far to apply this logic to Q1 2023…I’ve been pretty firmly on the “this FY” train but I have way less clue when we’d be hearing things if this was a March 2023 release. Maybe not yet. I still think in the next couple months if we’re in the same spot that doesn’t bode well. We’ll be 7-8 months out at that point.
 
bloopland33 said:
At this rate I feel like 2022 is off the table with more and more certainty as the days pass. I know people got burned last year and people took a lot of heat so they may be more hesitant to speak this year, but I still feel like we’d be hearing something from others if this was a 2022 device. And I don’t think Mochizuki is deterred by blowback, he’s a business journalist, he doesn’t care what Twitter/forums think.

I’m not sure how far to apply this logic to Q1 2023…I’ve been pretty firmly on the “this FY” train but I have way less clue when we’d be hearing things if this was a March 2023 release. Maybe not yet. I still think in the next couple months if we’re in the same spot that doesn’t bode well. We’ll be 7-8 months out at that point.
Click to expand...
If it’s August with no new real movement going into fall you can comfortably say it’s not gonna be this FY. That still gives us June or July for more info, but it’s coming down to the wire.
 
bloopland33 said:
At this rate I feel like 2022 is off the table with more and more certainty as the days pass. I know people got burned last year and people took a lot of heat so they may be more hesitant to speak this year, but I still feel like we’d be hearing something from others if this was a 2022 device. And I don’t think Mochizuki is deterred by blowback, he’s a business journalist, he doesn’t care what Twitter/forums think.

I’m not sure how far to apply this logic to Q1 2023…I’ve been pretty firmly on the “this FY” train but I have way less clue when we’d be hearing things if this was a March 2023 release. Maybe not yet. I still think in the next couple months if we’re in the same spot that doesn’t bode well. We’ll be 7-8 months out at that point.
Click to expand...
I'm in the same boat; I didn't even clue in until just now that we're already two thirds of the way through this month. Insiders were speculating that if we didn't hear anything about a potential tape out by the end of May, this device is certain to miss the 2022 window.

Granted, could just as well be the info doesn't leak at all and we're none the wiser, but a new console releasing alongside the flagship Zelda title in spring 2023 to showcase it's capabilities just makes too much sense even for Nintendo to turn a blind eye
 
ShaunSwitch said:
Agree they won't arbitrarily limit the speed of embedded memory but it will come down to cost vs benefit as well.

Let's say we get UFS card as external storage, 500MB/s, the cartridges some how manage to get up to 250MB/s and we get UFS 2.1 256GB at around 800MB/s, Nintendo tests it and in real world performance loading games the cartridges load maybe two seconds slower than UFS card and UFS 2.1 is a second faster.

Nintendo then tries UFS 3 and it loads another second faster on average, but the cost of UFS 3 means they can only go with 128GB storage at the price point they are targeting.

Will they go with UFS 2.1 or 3? Which will the consumer care about more? The small decrease in load times or twice the storage?

These are the kinds of things they will take into consideration when building a product for mass market. If the price difference between UFS 3 isn't much they may just go UFS 3 for the sake of it, but if its big enough that they could increase storage at a given price point using UFS 2.1 I see them using the latter.
Click to expand...
Although Mouser may not be the best source, I only had time for a quick search. Mouser is selling the Micron 256GB UFS v3.1 for $27.34, and v2.1 for $24.86; the cost difference is minor. Granted that the small cost different gets amplified at scale, but Nintendo might be able to negotiate a deal with their volume. Also, Orin already supports UFS 3.0, and it seems unlikely Drake would be downgraded (that is, if they choose to use UFS).
 
If I remember correctly, there were some articles during the middle and end of May last year with rumours of a new Switch model with an OLED screen or something like that, but it also had the other rumours like a beefier system. I think articles towards the end of May probably mentioned a better kickstand too but I'm not sure.

There was something in March too so maybe the stuff in May was just corroboration.
 
0
I'm not expecting 2022 at all and will think simultaneous release with botw 2 is a bit more likely, if not too predictable. But of course, anything goes.

With Switch not slowing down significantly in sales though, why would Nintendo release it this Fall opposed to Spring of the following year? What would they benefit from a Q4 release?. I can list a few reasons why from Nintendo's perspective why it could be better to push back.

But if Bayonetta 3 releases this year, that and Scarlett/violet are the perfect games to showcase them.

if Nintendo does go for a March-Spring 2023 release. Perhaps after Scarlet/Violet get released is the most strategic.. 🤔
 
Last edited:
fwd-bwd said:
Although Mouser may not be the best source, I only had time for a quick search. Mouser is selling the Micron 256GB UFS v3.1 for $27.34, and v2.1 for $24.86; the cost difference is minor. Granted that the small cost different gets amplified at scale, but Nintendo might be able to negotiate a deal with their volume. Also, Orin already supports UFS 3.0, and it seems unlikely Drake would be downgraded (that is, if they choose to use UFS).
Click to expand...
Thanks for this. I was trying to find some price differences between the two but was having no luck.

The difference in price could still be the difference between breaking even and making a small profit on each switch so I think it will come down to whether the CPU and memory setup can take advantage of the faster read speeds.

One other factor I hadn't considered was power draw, Samsung claims UFS 3 uses less power than 2.1, I imagine using UFS 3 at UFS 2.1 performance levels would reduce power consumption even further, so they may go UFS 3 for that reason too.

Storage is the one thing I keep coming back to and can't decide one way or the other.

I can see five possible combinations.

EMMC + Micro SD Cards.
UFS 2.1 + UFS Card
UFS 2.1 + NVME
UFS 3.0 + UFS Card
UFS 3.0 + NVME

I really don't know out of all of those which they will go for. I can't see micro sd being paired with faster storage as it could cause issues with developers and cause differences in load times between media.

Don't think they will use NVME, despite widespread adoption I don't think NVME is such a friendly format for their target demographic. Plus power draw. I don't think Nintendo wants its users to take a screwdriver to their switches either.

UFS Card will require Nintendo to partner with vendors as they are pretty uncommon, but actually pretty affordable and definitely a more user friendly option. This format is closer in speed to UFS 2.1 but could go either way.

The fact Orin has support for UFS 3 makes me feel its a more likely option but not ruling out UFS 2.1.
 
Last edited:
ShaunSwitch said:
Thanks for this. I was trying to find some price differences between the two but was having no luck.

The difference in price could still be the difference between breaking even and making a small profit on each switch so I think it will come down to whether the CPU and memory setup can take advantage of the faster read speeds.

One other factor I hadn't considered was power draw, Samsung claims UFS 3 uses less power than 2.1, I imagine using UFS 3 at UFS 2.1 performance levels would reduce power consumption even further, so they may go UFS 3 for that reason too.

Storage is the one thing I keep coming back to and can't decide one way or the other.

I can see five possible combinations.

EMMC + Micro SD Cards.
UFS 2.1 + UFS Card
UFS 2.1 + NVME
UFS 3.0 + UFS Card
UFS 3.0 + NVME

I really don't know out of all of those which they will go for. I can't see micro sd being paired with faster storage as it could cause issues with developers and cause huge differences in load times between media.

Don't think they will use NVME, despite widespread adoption I don't think NVME is such a friendly format for their target demographic. Plus power draw. I don't think Nintendo wants its users to take a screwdriver to their switches either.

UFS Card will require Nintendo to partner with vendors as they are pretty uncommon, but actually pretty affordable and definitely a more user friendly option. This format is closer in speed to UFS 2.1 but could go either way.

The fact Orin has support for UFS 3 makes me feel its a more likely option but not ruling out UFS 2.1.
Click to expand...
I think Nintendo will stick with eMMC.
 
A couple of AR-related patents from Nintendo where published today that where filed back in Nov 2021.

United States Patent Application 20220152493
United States Patent Application 20220157029

The patent drawings demonstrate the claims using smartphone/tablet-like devices, but I guess the math behind the patents could run on anything with a processor and a camera.

I don't even know if it's something already being used in a mobile app by Nintendo (I'm out of the loop in that area), and I don't find them too interesting tbh, but I'm just posting them here because AR is being brought-up occasionally.

I haven't gone through them in detail, but from what I gather one talks about creating "virtual doors" and the other one placing "virtual objects", in the environment.

The second one could be easily mistaken as NINTENDOGS for smartphones, but yeah, it is what it is.

screenshot2022-05-191sdjpj.jpg
screenshot2022-05-191u7jtq.jpg
screenshot2022-05-1915skl1.jpg
 
Last edited:
davec00ke said:
I think Nintendo will stick with eMMC.
Click to expand...
In this instance, I think you may be right.

I am reading some steam deck tests people have done using Microsd vs EMMC vs NVME and there is very little difference in game load times, if any at all.

The main benefit from using solid state media is a reduction in access latency and all three formats have that benefit.

The bottleneck to load times is more likely to be CPU and memory bandwidth so more expensive storage options may not make sense.
 
ShaunSwitch said:
In this instance, I think you may be right.

I am reading some steam deck tests people have done using Microsd vs EMMC vs NVME and there is very little difference in game load times, if any at all.

The main benefit from using solid state media is a reduction in access latency and all three formats have that benefit.

The bottleneck to load times is more likely to be CPU and memory bandwidth so more expensive storage options may not make sense.
Click to expand...
there will be the question of when games are starting to stream assets from storage. can emmc handle that at a reasonable rate? I doubt it, without major cuts to the asset quality
 
Errrm... UFS 3.0 has been around since 2019. Version 3.1 since 2020. I lament that Nintendo discourse has hurt some of you, and conditioned you to scrape through the barrel when forming expectations, but this is beyond silly. Ever since 1536 CUDA Cores showed up in the leaks, it appears that a new copiate is needed for those who can't bring themselves to believe what we know to be real, what came from the horse's mouth, because the "cheap, behind the times, because Nintendo" narrative doesn't fit. In the last few pages, we've gone back to UFS 2.1 when Orin supports 3.0, quad-core processors when the Orin chip with 1024 CUDA Cores has an octa-core one, and we still expect this thing to perform DLSS and raytracing competently with older, weaker, and less efficient parts. Make it make sense. No bottlenecks at all :rolleyes:. Come On... I just find it wild as hell, tbqh.
 
Last edited:
UFS 3.0 is gonna be a big deal. Should close the gap between the SSD’s and Switch load speeds.
 
I, too, think they will use emmc and micro sd again.
Of course everything will depend on the final price of the new components.
 
0
ReddDreadtheLead said:
The reason for those cuts on the consoles is probably not really applicable for PCs really. I think Consoles go by a perf-per-dollar while PC is perf-per-watt model. They make necessary cuts to reduce cost and increase yield for larger mass manufacturing of the die, as that is what you get for several years. As the platform gets older, the parts depreciate and becomes cheaper to produce and thus cheaper for a consumer.

With a PC part, if it’s binned it is still a sellable product and they can sell you the whole thing in some fashion or form, without necessitating the need to discard said silicon.

Previous consoles which (off the top of my head) that made cuts besides the series and PS5, were the 360 and PS3. They cut out the OoOE, and were In-order, but added other features to them. 360 had that extra 10MB of memory and the vector unit, PS3 had the infamous CELL processor which was brilliant in concept but poor in its execution.

PS5 and SX/S cuts are more likely related to not being too big to make it difficult to produce (would have been around 400mm^2 I think).

With an ARM CPU which are significantly smaller than the Zen2 CPUs, this is less of a concern with making a viable product with the die size than said Zen2 based product, but it is still a concern nonetheless and can increase cost. ORIN is like a 400-500mm^2 die.

And has that noticeable more robust caching system (plus the other stuff).
Click to expand...
For the PS5/Series, I think that it goes back to the difference between chiplet and monolithic Zen 2, since the difference is consistent there.
And for the difference in cache there, what I hear is that the higher cache for chiplet Zen 2 is to attempt to reduce the frequency of going off die/chiplet, since that would have worse latency than it would for monolithic designs. Sounds sensible enough. And in turn the monolithic designs have less L3 cache to save some money.
This disparity is seen again with Zen 3; the chiplet versions have 32 MB per CCX while monolithic versions have 16 per CCX.

I'm not as familiar with PS360, so thanks for that example.
 
0
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