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

[Look over there]

I kinda want to avoid increasing the CPU power budget by much though, since we're eating an automatic increase from RAM due to the utilization of the expected increase in memory bandwidth. Like maybe up to another watt or so in handheld?

 

[ShadowFox08]

I kinda want to avoid increasing the CPU power budget by much though, since we're eating an automatic increase from RAM due to the utilization of the expected increase in memory bandwidth. Like maybe up to another watt or so in handheld?

yeah I know what you mean.. CPU will likely be a major bottleneck again for current ports though that are CPU heavy, unless they close the gap a bit. 4 A78s is not enough, especially when only 3 will be for games.. Nintendo needs to do better and double that, which would give Drake 2.33x more cores to work with (7 vs 3) for games. And they should raise the clock speeds by 40-50% more if they can help it. Anything helps.

But yeah it's likely already decided anyway, with a small chance of increase near launch. Whether we get the node size first or CPU and GOU clocks, the node has been chosen and will decide how high the clockspeeds will be as Nintendo will balance it with a similar power draw/heat dissipation/respectable battery life to the switch most likely.

 

[Deleted member 887]

A critical problem with 420MHz on Drake's GPU as the "High End" as you seemingly imply...Ampere's power curve bottoms out at 300MHz.


The system as depicted in that context would have pretty much no change between Portable and Docked modes...

How so? 420Mhz in portable and 1Ghz in docked isn't different? This is the same ratio as the current Switch? What am I missing?

no. you'd be running as limited as the PS4/XBO and in a world where the competition is running unfettered Zen 2s, you can almost absolutely rule out ports. even 6-cores could be stretching it, but at least you can clock higher

Sorry, by "reasonable" I was only asking in terms of power draw, not as in whether or not it would be a reasonable dev platform.

 

[Alovon11]

How so? 420Mhz in portable and 1Ghz in docked isn't different? This is the same ratio as the current Switch? What am I missing?


Sorry, by "reasonable" I was only asking in terms of power draw, not as in whether or not it would be a reasonable dev platform.

Oh, I thought you were taking 400MHz outright as you didn't mention how docked mode were perform there, my B.

 

[Z0m3le]

How so? 420Mhz in portable and 1Ghz in docked isn't different? This is the same ratio as the current Switch? What am I missing?


Sorry, by "reasonable" I was only asking in terms of power draw, not as in whether or not it would be a reasonable dev platform.

Remember, the OS cpu core can be downclocked while in game, so 6 or 8 cores makes more sense, especially because they don't add much to the cost as they fit in one package thanks to A78C, which was designed for games specifically.

Also, minimum portable clock is likely 460MHz just to cover any issues from Switch compatibility, you'd be looking at a very small power hit to support the additional 40mhz. That's 1.41tflops.

 

[FishyFish]

So is it possible the next Switch could be a little thicker? I ask because really the current Switch's joy cons are a bit on the small side for me. Also a thicker Switch would probably allow for slightly better cooling, would it not?

 

[Hermii]

I have a sick day and I have to look at Nvidia docs for a work project. TL;DR Do we have any reason to believe that Nintendo will go with an 8-core CPU config other than "that would be nice?"

Nope.

 

[Hermii]

Nintendo? Most definitely not, but I am sure Nvidia does as AMD is their rival.

I don’t think Nvidia does either honestly.

They just want Nintendo to keep buying tech from them. As long as both companies are happy with what they’re buying/ selling the SD doesn’t factor into anything.

 

[ITSMILNER]

Are people still anticipating an announcement of this device in September?

 

[Hermii]

Are people still anticipating an announcement of this device in September?

I don’t know who you are referring to by “people”, probably just one or two people.

It’s not some commonly held belief.

 

[Mobius]

Are people still anticipating an announcement of this device in September?

Unlikely if this thing is set to release in April or later, would Nintendo want to lose hardware sales for +6 months?

 

[ITSMILNER]

I don’t know who you are referring to by “people”, probably just one or two people.

It’s not some commonly held belief.

Wasn’t one of the details from the Funcle that they expected a September announcement? I might be misremembering

 

[Hermii]

Wasn’t one of the details from the Funcle that they expected a September announcement? I might be misremembering

That wasn’t the factory uncle, it was the other uncle that predicted splatoon oled. He could have been referring to a direct.

 

[Thugstas]

That wasn’t the factory uncle, it was the other uncle that predicted splatoon oled. He could have been referring to a direct.

He speculated bout September reveal if I’m correct

 

[Mobius]

He speculated bout September reveal if I’m correct

yes just speculation, nothing to take seriously

 

[Simba1]

Talking about CPU, CPU will probably this time be even bigger bottleneck than compared Switch to XB1/PS4,
Drake will have much stronger CPU (that would again be restrictected by heating and battery life with lower clocks) on other hand this time XS/PS5 have full strong desktop CPU.

 

[SiG]

Talking about CPU, CPU will probably this time be even bigger bottleneck than compared Switch to XB1/PS4,
Drake will have much stronger CPU (that would again be restrictected by heating and battery life with lower clocks) on other hand this time XS/PS5 have full strong desktop CPU.

I mean it really depends how far the SoC market will be pushed.

Even with Apple's M1/M2 chips, they could only go so far as to try to match full fat desktop with descrete GPU workloads. Nvidia SoCs could have a chance, but is still mainly dependent on what's the most powerful chip ARM has to offer.

Again, some folks are expecting ARM chips to go toe-to-toe with a desktop CPU. It doesn't quite work like that. It's just incredibly unrealistic of an expectation to say "They have to or bust".

 

[Skittzo]

So is it possible the next Switch could be a little thicker? I ask because really the current Switch's joy cons are a bit on the small side for me. Also a thicker Switch would probably allow for slightly better cooling, would it not?

Even if the main unit is thicker there's currently nothing suggesting any change in the joycons.

 

[Simba1]

I mean it really depends how far the SoC market will be pushed.

Even with Apple's M1/M2 chips, they could only go so far as to try to match full fat desktop with descrete GPU workloads. Nvidia SoCs could have a chance, but is still mainly dependent on what's the most powerful chip ARM has to offer.

Again, some folks are expecting ARM chips to go toe-to-toe with a desktop CPU. It doesn't quite work like that. It's just incredibly unrealistic of an expectation to say "They have to or bust".

Offcourse, but again mobile CPU performance highly depending from heat and battery life and thats why we expecting around 1.5GHz CPU clock at most for Drake, while you dont have such-a restriction in stationary hardware (I mean you have heating but thats not big problem in stationary hardware with bigger and better cooling).

Yes, at end we are comparing around 5-10W CPU with around 50-90W CPU.
So I disagree when someone says Drake will offer similar power/performance like Xbox Series S, even counting DLSS, someone even wrote that Drake will be somewhere betwine Xbox Series S and PS5.
IMO its much safer comparison something like around PS4 Pro in docked mode with DLSS and that would be very strong for mobile hardware,
also people should have on mind that games still didnt start really pushing PS5/XSX hardware because there are still over 90% of cross gen games.

I am sure Nintendo and exclusive games will perform and look great on Drake, but current gen port games will again be worse than for instance Xbox Series S version of games.

 

[SiG]

Offcourse, but again mobile CPU performance highly depending from heat and battery life and thats why we expecting around 1.5GHz CPU clock at most for Drake, while you dont have such-a restriction in stationary hardware (I mean you have heating but thats not big problem in stationary hardware with bigger and better cooling).

Yes, at end we are comparing around 5-10W CPU with around 50-90W CPU.
So I disagree when someone says Drake will offer similar power/performance like Xbox Series S, even counting DLSS, someone even wrote that Drake will be somewhere betwine Xbox Series S and PS5.
IMO better comparison is PS4 Pro, also people should have on mind that current gen still didnt start really pushing PS5/XSX hardware because there are still 90% of cross gen games.

I am sure Nintendo and exclusive games will perform and look great on Drake, but current gen port games will again be worse than for instance Xbox Series S version of games.

My problem is this assumption that they have to match/close the gap between the XBox Series S and PS5 in order to gain support/prove they have a viable product.

 

[Simba1]

My problem is this assumption that they have to match/close the gap between the XBox Series S and PS5 in order to gain support/prove they have a viable product.

Problem is that people generally want that Drake match current gen in power/performance (at least for XSS),
forgetting that we comparing full home consoles that basically have full PC components with mobile hardware.

 

[Erimgard]

Problem is that people generally want that Drake match current gen in power/performance (at least for XSS),
forgetting that we comparing full home consoles that basically have full PC components with mobile hardware.

I just want pretty first-party games with a stable frame rate tbh

 

[Stinky Horse]

CPU will be whatever gives the best performance for ~2 Watts or less, it's not a question of having to match another company's machine.

 

[ReddDreadtheLead]

Talking about CPU, CPU will probably this time be even bigger bottleneck than compared Switch to XB1/PS4,
Drake will have much stronger CPU (that would again be restrictected by heating and battery life with lower clocks) on other hand this time XS/PS5 have full strong desktop CPU.

Mm, probably not/not really, at least not the way it’s being made to be in theory.

If they can run on these older machines Desktop machines that have competent CPUs but aren’t as strong as the PS5/XBS, they can probably run good enough on a system like Drake (assuming 8 cores).

Reminder that the consoles have those CPUs with a good amount of headroom as reassurance. The amount that will actually push those to the limit is like only 2 studios that are third party.


One is Rockstar and the other will probably be CDPR.


probably

Drake would just be more at its limit and actually be pushed but the others wouldn’t be breaking a sweat as they aren’t being taxed so badly in that area. Not to mention, the GPU that it has, Ampere, has some uniqueness that is also present in RDNA2 in which it reduces some necessity for the CPU to do the same task.

ie, it leaves it more to the GPU to do it.

It’s GPU is also very good for things like physics so if it came down to it, it could be moved over to save CPU resources.

If it came down to it.


There’s also the thing that CPU performance doesn’t quite scale linearly. It can be 2x as strong on paper or 4x, but only net you 10-15 extra frames (for example purposes) in a desktop environment, imagine for a console that gets specific levels of optimization done to it?

I’m not saying that there won’t be issues, but it may not be as extreme as it’s being painted out to be here.

Yes, at end we are comparing around 5-10W CPU with around 50-90W CPU.

sigh

I’ve seen this type of comparison across this thread, the last thread and the thread before it.

It’s like hearing someone say that shampoo and conditioner are the exact same thing and that conditioner is worse because it doesn’t clean your hair when it isn’t even the goal of conditioner.


I’m not necessarily saying that you are wrong but more that the comparison is very flawed

 

[Terrell]

And honestly, when Horizon OS is barely sipping on the CPU and RAM compared to what you see with other hardware makers and what it takes to run their OSs in the background of games, raw CPU performance isn't an adequate metric to measure against in the first place.

 

[Stinky Horse]

Idk what Rockstar games people are expecting that would require more than an Xbox One X level CPU.

 

[PedroNavajas]

Thinking about it, 6 cores makes sense. 4 cores would perhaps be too little, since one ought to be dedicated to the OS. And I would take less higher clocked cores over many slow clocks. 8 cores at 1Ghz would be worse than 4-5 at 1.6GHz.

8 cores at +2Ghz is unrealistic.

 

[ReddDreadtheLead]

Idk what Rockstar games people are expecting that would require more than an Xbox One X level CPU.

GTA6 at 120FPS and no less of course.

 

[PedroNavajas]

Idk what Rockstar games people are expecting that would require more than an Xbox One X level CPU.

PS4 and XB1 had terrible CPUs. 4x2Ghz A78 would be stronger than that.

 

[mjayer]

Problem is that people generally want that Drake match current gen in power/performance (at least for XSS),
forgetting that we comparing full home consoles that basically have full PC components with mobile hardware.

This “it’s mobile so it can’t compare” line of reasoning isn’t the whole truth, imo. It’s a hybrid after all. Hence the very name. They can take it beyond “mobile” when docked, it does have a fan. I realize there are limitations because it’s not using desktop hardware but that gap is not the same, especially when it’s custom developed for it’s purpose this time, and it’s not an off the shelf mobile chip. Markio was just a node change right? This is the first time they are developing a new chip for a successful platform that features the hybrid nature and optimizing for that.

 

[ReddDreadtheLead]

PS4 and XB1 had terrible CPUs. 4x2Ghz A78 would be stronger than that.

It would only be up to PS4Pro/XB1X.

 

[PedroNavajas]

It would only be up to PS4Pro/XB1X.

Those were just slightly overclocked from base.

 

[ShaunSwitch]

Thinking about it, 6 cores makes sense. 4 cores would perhaps be too little, since one ought to be dedicated to the OS. And I would take less higher clocked cores over many slow clocks. 8 cores at 1Ghz would be worse than 4-5 at 1.6GHz.

8 cores at +2Ghz is unrealistic.

I think it depends on what node the chip is on.

If it's Samsung 8nm I'm expecting 6*A78C at a modest clock, maybe 1.4ghz, though I wouldn't rule out 8 cores at 1GHZ either.

If it does end up being a TSMC 5nm part then 8 CPU Cores is more likely IMO with maybe a 1.6ghz clock speed, I wouldn't even rule out 12 A78 cores on 5nm though I see it as unlikely as it would have to be 2 clusters of 6 and would complicate the chip design for little benefit.

 

[ILikeFeet]

CPU will be whatever gives the best performance for ~2 Watts or less, it's not a question of having to match another company's machine.

it also needs to make third parties happy. if the gap is just too large, then you can't conduct business

 

[ITSMILNER]

That wasn’t the factory uncle, it was the other uncle that predicted splatoon oled. He could have been referring to a direct.

Ahh ok, that makes more sense seeing as the device isn’t due for a good few months yet.

 

[Deleted member 1324]

NINJAD

 

[Deleted member 887]

I think it depends on what node the chip is on.

If it's Samsung 8nm I'm expecting 6*A78C at a modest clock, maybe 1.4ghz, though I wouldn't rule out 8 cores at 1GHZ either.

This was what I was trying to figure out when I set the CPU talk off by accident

I just don’t see how 8 cores 1Ghz fits in the power budget knowing what we know about the GPU, and knowing what we know about Orin.

I keep being tempted to just buy an AGX devkit and clock the DLA/PVA down to nothing, clock the GPU to what we expect and see the power draw but that’s a hefty amount of money to throw at a single post on a forum.

Remember, the OS cpu core can be downclocked while in game, so 6 or 8 cores makes more sense, especially because they don't add much to the cost as they fit in one package thanks to A78C, which was designed for games specifically.

Sure, you can get power savings by down clocking one core, but is it 3 TPC’s of savings? My goal was just to see how much headroom there was on 8nm not definitively state how many cores had to be in the device.

A78C doesn’t offer any power savings over A78, in fact the opposite. I don’t think swapping it in improves the power budget situation

Also, minimum portable clock is likely 460MHz just to cover any issues from Switch compatibility, you'd be looking at a very small power hit to support the additional 40mhz. That's 1.41tflops.

Smart, yeah that makes sense.

 

[ShaunSwitch]

This was what I was trying to figure out when I set the CPU talk off by accident

I just don’t see how 8 cores 1Ghz fits in the power budget knowing what we know about the GPU, and knowing what we know about Orin.

I am in a similar situation. Looking at the 1536 core GPU I don't see how they can provide a CPU that would do it justice on samsung 8nm. Maybe Nintendo goes with a super modern battery? Battery densities apparently improve by 5%-8% per year, even at the top end that's a 48% bigger battery. But even if its got an extra 50% power budget, how are they going to cool this thing in the same form factor as OG switch? Just doesn't make sense to me.

Has to be TSMC 5nm at this point.

 

[ILikeFeet]

Has to be TSMC 5nm at this point.

Nvidia could have gotten a good deal on Samsung 5nm. that's not off the table yet

 

[Deleted member 887]

Working backwards from 4LPP to get to 8LPP, take the 3 Ghz@1 watt, then go 3/(1.11*1.11*1.09) ~= 2.23 Ghz@1 watt.
Given that, my guess for how much power 1.7 Ghz would require would then be...
2.23/1.7 ~= 1.31 (ratio of the two frequencies)
1.31^2 ~= 1.73 (the relative increase in power to go from 1.7 Ghz to ~2.23 Ghz)
1/1.73 ~= 0.58
So, at best, 0.58 watts to get a single A78 to 1.7 ghz on Samsung's 8 nm node. More likely to be a bit higher than that. Probably fair to round it up to 0.6 watts?

If we stick your handwavy math into MY handwavy math, what do we get?



This is the preset power profiles on the Orin NX 8GB. For those asking why I was using the AGX as my base - I looked at both to try an isolate the power costs of the TPCs, which is how I started, but the AGX had a profile that looked most "Drake like". But, this power profile conveniently sits at 15 watts and has CPU clocks at about 1.7Ghz.

The Orin devices have AE cores. Let us assume that the power draw for a single AE core is the same two C cores. So the CPU config here looks exactly like what we're talking about - 8 CPU cores, running at 1.7 GHz. Running your math in, the CPU cores are consuming 4.8W of the 15W available.

This config has the PVA Core off (though some of the ancillary PVA hardware is still available), so for our handwavy estimates lets just assume the PVA cost is negligible. Two single speed DLA cores run in the range of 4-6W, so let's call it 5W for simplicity.

That leaves 5.2W to run 2TPCs @ 620Mhz. Now, obviously, there is still overhead to be cut here. The NX/AGX kits here are using LPDDR5 just like we expect Drake to use, but we haven't factored that out. Let's handwave some more and say that, overall, there is 10% power savings to be had just factoring out IO pins and the like from these dev kits.

More handwaving - let us assume that the differences between Orin Ampere and Drake Ampere are power neutral.

440Mhz seems like a floor for the GPU in handheld mode, to match various seemingly corroborated rumors about the performance of the device. Assuming that, at these clocks, TPCs scale linearly with power, and knowing that Drake has 6 TPCs...

5.2 * 0.9 * (440/620) * 3 = ~9.96W for Drake's GPU alone, in handheld mode, on Samsung 8nm.

If we assume that @Look over there 's CPU numbers are wildly optimistic, ironically, we can just about make it work if CPU's are consuming a full watt per core - but that doesn't pass the smell test on ARM efficiency or Ampere efficiency.

I've long generally assumed that Drake was 8nm, because I didn't see a reason to assume otherwise, and that we would see very low clocks to compensate power-wise. My assumption has also been that this would be positioned as a "Pro" device and that Nintendo would be able to squeeze a "Successor" device out of the same investment by some combination of die shrinking Drake and/or riding other efficiency gains to let them up the clocks. But I think I have to finally concede that doesn't track.

 

[Stinky Horse]

Full on Automotive Orin is intended to be run as though it's a 6 core CPU, with the other 6 running the same instructions paired in lockstep for safety/redundancy. Something to consider.

 

[Z0m3le]

Going live with the NES Transmission podcast feel free to join/ask questions

 

[RennanNT]

Also, minimum portable clock is likely 460MHz just to cover any issues from Switch compatibility, you'd be looking at a very small power hit to support the additional 40mhz. That's 1.41tflops.

I don't think BC requirements means much for Drake games. They could still limit them to 384 or even 307 MHz while allowing BC games to use 460 MHz (but with lower CPU/RAM clocks to compensate and hit the target battery life).

 

[Deleted member 887]

I am in a similar situation. Looking at the 1536 core GPU I don't see how they can provide a CPU that would do it justice on samsung 8nm. Maybe Nintendo goes with a super modern battery? Battery densities apparently improve by 5%-8% per year, even at the top end that's a 48% bigger battery. But even if its got an extra 50% power budget, how are they going to cool this thing in the same form factor as OG switch? Just doesn't make sense to me.

Has to be TSMC 5nm at this point.

Can it not be TSMC 7nm? Again, not my area of expertise, but Ampere has only ever used TSMC 7nm and Samsung 8nm, yes? And from a power perspective, you don't need huge gains to get to where we need to go. 30% better power efficiency across the board, and we're comfortably sub 10W for the SoC. Drake supposedly has some backported Lovelace features, but Lovelace and Hopper are supposedly TSMC N4.

 

[Dakhil]

Nvidia could have gotten a good deal on Samsung 5nm. that's not off the table yet

Although that's true, the question I have is are there any future products from Nvidia that Nvidia plans to fabricate using one of Samsung's 5 nm** process nodes?

Nvidia generally secures capacity for a process node to fabricate more than one product, with TSMC's 20 nm** process node being the sole exception with the Tegra X1 so far.

** → a marketing nomenclature used by all foundry companies

---

Can it not be TSMC 7nm? Again, not my area of expertise, but Ampere has only ever used TSMC 7nm and Samsung 8nm, yes? And from a power perspective, you don't need huge gains to get to where we need to go. 30% better power efficiency across the board, and we're comfortably sub 10W for the SoC. Drake supposedly has some backported Lovelace features, but Lovelace and Hopper are supposedly TSMC N4.

I think TSMC's N6 process node is a possibility, especially since TSMC's N6 process node is an evolution of TSMC's N7 process node, and TSMC has been shifting customers from TSMC's N7 process node to TSMC's N6 process node. And Nvidia's still using TSMC's 7 nm** process node for fabricating datacentre chips (Quantum-2, BlueField-3, and ConnectX-7).

And just a correction, Hopper (and Ada is rumoured to) be fabricated using TSMC's 4N process node, which is apparently based on TSMC's N5P process node.

** → a marketing nomenclature used by all foundry companies

 

[Brubi]

First of all hey all and sry for my english^^
@Dakhil My dream would be like in the stream said, that Nin goes all in aka SKU for 499 Dollar.
And that they use TSMC 4nm or 3nm, it exist a rumour that Nvidia got a lot of the 4nm and wanted to cut theme off but TSMC rejected that?

And was not there a rumour or just speculating that Nin was not happy about the System itself?

 

[Dakhil]

Hey everyone. Didn’t get much info as we weren’t alone a lot but I did ask about the drive. He said that Drake uses the same kind of drive as the current Switch but that it could change as there’s still a while before release. That’s the last I’m going to ask as he seemed annoyed when I brought it up so I’m going to leave it there.

I hope you’re all having a good weekend!

Do you mean the internal flash storage or the Game Cards when mentioning "drive"?

---

First of all hey all and sry for my english^^
@Dakhil My dream would be like in the stream said, that Nin goes all in aka SKU for 499 Dollar.
And that they use TSMC 4nm or 3nm, it exist a rumour that Nvidia got a lot of the 4nm and wanted to cut theme off but TSMC rejected that?

And was not there a rumour or just speculating that Nin was not happy about the System itself?

Your English is perfectly understandable.

I think TSMC's N3 process node and/or TSMC's N3E process node is/are completely out of the question, considering that Apple's most definitely paying huge amounts of money to have temporary, exclusive access to TSMC's N3 process node and/or TSMC's N3E process node before other companies do.
And yes, there was a rumour from DigiTimes about Nvidia securing too much capacity for TSMC's 4N process node. And I think there are grains of truth to that rumour since TSMC has warned customers of excessive capacity.

And I don't believe I've heard any rumours about Nintendo not being happy with Nintendo's new hardware.

 

[Deleted member 1324]

Do you mean the internal flash storage or the Game Cards when mentioning "drive"?

---

Your English is perfectly understandable.

I think TSMC's N3 process node and/or TSMC's N3E process node is/are completely out of the question, considering that Apple's most definitely paying huge amounts of money to have temporary, exclusive access to TSMC's N3 process node and/or TSMC's N3E process node before other companies do.
And yes, there was a rumour from DigiTimes about Nvidia securing too much capacity for TSMC's 4N process node. And I think there are grains of truth to that rumour since TSMC has warned customers of excessive capacity.

And I don't believe I've heard any rumours about Nintendo not being happy with Nintendo's new hardware.

Console drive.

 

[ShaunSwitch]

Can it not be TSMC 7nm? Again, not my area of expertise, but Ampere has only ever used TSMC 7nm and Samsung 8nm, yes? And from a power perspective, you don't need huge gains to get to where we need to go. 30% better power efficiency across the board, and we're comfortably sub 10W for the SoC. Drake supposedly has some backported Lovelace features, but Lovelace and Hopper are supposedly TSMC N4.

I think it's possible. As Dakhil said maybe 6nm if that is what TSMC are moving their customers to.

If there is anything to those rumours that nvidia purchased too much capacity for ADA then my gut is telling me it will be on that node.

 

[Dakhil]

Console drive.

Thank you for the clarification.

Basically, devkits for Nintendo's new hardware are using eMMC 5.1 for the internal flash storage, which is the same type of internal flash storage used on the Nintendo Switch, Nintendo Switch Lite, and the OLED model, according to rumours.