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

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

Reader mode
niconiconick64 said:
But I'm still guessing:
7nm TSMC
Portable: 375-450MHz GPU, ~1 GHz CPU
Docked: 850-900 MHz GPU, ~1.5-1.7 GHz CPU
Just to keep my hopes in check.
Click to expand...
afaik the Switch keeps CPU frequency consistent between docked and portable, as game logic relies upon it. GPU speeds shift since the requirements vary depending on output resolution.
 
niconiconick64 said:
Hmmm... Now that I think about it, SoCs like the Snapdragon 888 (which utilizes A78 cores) and the cheaper 875 are nearly 2 years old and were manufactured under a 5nm node.
So definitely 5nm or at the very least, TSMC's 7nm could be a real possibility considering the aforementioned power efficiency concerns posted by other users ITT.

But I'm still guessing:
7nm TSMC
Portable: 375-450MHz GPU, ~1 GHz CPU
Docked: 850-900 MHz GPU, ~1.5-1.7 GHz CPU
Just to keep my hopes in check.
Click to expand...

The CPU is unlikely to be clocked differently between portable and docked play. They will want to keep the game logic the same when switching.
 
kvetcha said:
I can't recall off the top of my head, but is there another feasible ARM processor that can handle a 8-core, 1-cluster configuration?
Click to expand...
You know what, you're right. I saw this, thought "man, this confirms A78C" then immediately read Thraktor's comment and confused myself. This essentially confirms the C variant, yes, as base A78 has a max of 4 CPUs per cluster?
 
niconiconick64 said:
Hmmm... Now that I think about it, SoCs like the Snapdragon 888 (which utilizes A78 cores) and the cheaper 875 are nearly 2 years old and were manufactured under a 5nm node.
So definitely 5nm or at the very least, TSMC's 7nm could be a real possibility considering the aforementioned power efficiency concerns posted by other users ITT.
Click to expand...
I've been thinking 7nm, because that's what Ampere is built on, but now I'm thinking 5nm since we've confirmed octo-core. A78 was built for 5nm, and Lovelace is 5nm (I believe?) and is stated to have at least started as an Ampere refresh on a smaller node.

niconiconick64 said:
But I'm still guessing:
7nm TSMC
Portable: 375-450MHz GPU, ~1 GHz CPU
Docked: 850-900 MHz GPU, ~1.5-1.7 GHz CPU
Just to keep my hopes in check.
Click to expand...
Other than the fact that the CPU will likely stay clocked the same (as others have pointed out) I think this is a reasonable floor, and about where my head is at too.
 
kvetcha said:
afaik the Switch keeps CPU frequency consistent between docked and portable, as game logic relies upon it. GPU speeds shift since the requirements vary depending on output resolution.
Click to expand...
Gotdatmoneyy said:
The CPU is unlikely to be clocked differently between portable and docked play. They will want to keep the game logic the same when switching.
Click to expand...
unknown.png


Wasn't the Switch's CPU clocked well above 1GHz anyways for certain titles like BoTW?
 
Ojoloco said:
Click to expand...

Well shit, when it rains it pours. Inferring frames was the inevitable direction of this tech, but inferring frames and then using those frames as inputs to the upscaler sounds like some combination of magic and/or foolishness. You'd think you could only get one or the other.
 
0
niconiconick64 said:
Hmmm... Now that I think about it, SoCs like the Snapdragon 888 (which utilizes A78 cores) and the cheaper 875 are nearly 2 years old and were manufactured under a 5nm node.
So definitely 5nm or at the very least, TSMC's 7nm could be a real possibility considering the aforementioned power efficiency concerns posted by other users ITT.

But I'm still guessing:
7nm TSMC
Portable: 375-450MHz GPU, ~1 GHz CPU
Docked: 850-900 MHz GPU, ~1.5-1.7 GHz CPU
Just to keep my hopes in check.
Click to expand...
Besides the changing CPU that looks pretty reasonable.
niconiconick64 said:
unknown.png


Wasn't the Switch's CPU clocked well above 1GHz anyways for certain titles like BoTW?
Click to expand...
That's only for short bursts during loading, and it's done in both docked and undocked modes.
 
bloopland33 said:
So what the hell is lore.kernel.com? The Google search preview says "This is a listing of public inboxes, see the `mirror' link of each inbox for instructions on how to mirror all the data and code on this site."

Why are these inboxes public and on that site!? I don't get it.
Click to expand...

It's for the Kernel of Linux. Linux (base) is fully open source, so stuff like this has to be open too i'd say.
 
niconiconick64 said:
unknown.png


Wasn't the Switch's CPU clocked well above 1GHz anyways for certain titles like BoTW?
Click to expand...
The Switch lets games select these perf profiles per context. The "boost mode" perf is short lived and designed to be activated by games during loading screens and the like to speed up decompressing assets.
 
0
niconiconick64 said:
unknown.png


Wasn't the Switch's CPU clocked well above 1GHz anyways for certain titles like BoTW?
Click to expand...
Those profiles are used only to accelerate load times, not during typical use. The Switch CPU is clocked to 1020mhz essentially all the time, portable or docked.
 
0
Simba1 said:
Real question is why than Nintendo used 1GHz instead above 2GHz? There is huge difference between 1GHz and 2GHz.
Nintendo could easily push Switch A57 to run at least at 1.2GHz or 1.3GHz, but they didnt, they usually going with more lower clocks than device could run at.

I am willing to bet that Drake will not have 2GHz CPU clock (especially if it's really 8-core CPU). :)




Like I wrote, regardless of node I dont see Nintendo running 2GHz or above 2GHz with Drake hardware.
We talking about mobile devices that should use those clocks all the time (for 8 cores), while in same time worrying about heating and battery life,
and Nintendo usually don't like to push clocks even hardware could run higher clocks.
Click to expand...
Simba1 said:
Thats my point, because heating and power consumption/battery life they didnt run 2GHz or above,
but they could definitely run CPU above 1GHz (for instance 1.2-1.3GHz) without big difference.
Hack scene clocked Switch CPU even at 1.8GHz and Switch was used regularly and was working without problems,
yes little higher heating and less battery life but it was working without problems.

Like I wrote Nintendo usually going with lower clocks than device could run at,
point that Drake could run 8-core CPU at 2GHz doesnt mean automatically that Nintendo will do that.
Click to expand...
You’re comparing apples to oranges here. You’re comparing the first iteration of arm’s new design, which was the A57 that supported 64-bit instruction sets to the eighth generation of CPUs the arm has which support the 64-bit instruction, and the latter is on a much superior node. On top of that you’re comparing a CPU who is designed to not clock beyond 2.1 GHz to a CPU who is designed to clock no higher than 3.3 GHz. The comparison was already flawed to begin with. Yes, they are both arm-based cores, however the A57 is very ancient at this point and as soon as you start clocking it higher than its max or close to its max, You start wasting a lot of energy, you introduce throttling, a lot of heat, etc.

If Drake had a CPU clocked to 2GHz (not saying it will), it would not be comparable to the situation of the A57 and 2GHz.



To make it even clearer, a single A57 core clocked to 1.9 GHz consumes about 2W. A single a78 core clocked to 1.9 GHz consumes less than a quarter of that.
 
niconiconick64 said:
unknown.png


Wasn't the Switch's CPU clocked well above 1GHz anyways for certain titles like BoTW?
Click to expand...
There is a boost mode that maxed out the CPU and brought the GPU to extremely low performance to help loading games since the CPU is used in decompression. That mode would is the same if you had the Switch docked or mobile as well. For gameplay the CPU clocks never change change between playstyle.
 
Getting info on the CPU is great. I don't think it means anything for the release timing of the product we actually care about, though.

And man, is history ever repeating itself, with Nvidia announcing a new architecture and a bunch of features that the new Switch won't have. Only semi-ironically worried now for all the hot takes that an Ampere Switch is badly outdated and Nintendo should have used Ada, just like all the Maxwell vs. Pascal nonsense from 2016/2017.

Skittzo said:
Do you mean the codename itself has changed or is now referring to something else? That seems extremely unlikely.

Could the specs of T239 have changed since February? Maybe, but that's also fairly unlikely considering how far along it seemingly was. But possible.
Click to expand...
It's less about the specs changing since February and more about what was in the leak not necessarily having been up to date, between the lack of reliable date information, and the fact that it was a certain branch of the repository where it's not likely active development was happening directly.
 
oldpuck said:
I've been thinking 7nm, because that's what Ampere is built on, but now I'm thinking 5nm since we've confirmed octo-core. A78 was built for 5nm, and Lovelace is 5nm (I believe?) and is stated to have at least started as an Ampere refresh on a smaller node.


Other than the fact that the CPU will likely stay clocked the same (as others have pointed out) I think this is a reasonable floor, and about where my head is at too.
Click to expand...
4 nm?

FdG4e2VXwAAAv0G
 
bloopland33 said:
So what the hell is lore.kernel.com? The Google search preview says "This is a listing of public inboxes, see the `mirror' link of each inbox for instructions on how to mirror all the data and code on this site."

Why are these inboxes public and on that site!? I don't get it.
Click to expand...
kernel.org is the website used to coordinate the development of the Linux kernel, which is open source. Because of the nature of that - lots of companies collaborating on a shared, public project, or even some individual developers who are funded by multiple companies - there is a mailing list that is Quite Old, and very very busy, the Linux Kernel Mailing List, or LKML for short. LKML got so busty eventually you had to bust out sub mailing lists for people to coordinate work that then gets pushed to LKML. lore.kernel.org is the archive off all those emails, so that they're searchable, because actually subscribing to the maillist is an easy way to destroy your inbox.
 
kvetcha said:
Kinda sounds like DLSS3 will require an Ada GPU?
Click to expand...
Not necessarily.
www.globenewswire.com

NVIDIA Introduces DLSS 3 With Breakthrough AI-Powered Frame Generation for up to 4x Performance

DLSS 3 Generates Entire Frames in Real Time With AI, Benefiting GPU- and CPU-Limited Games and Apps...
www.globenewswire.com www.globenewswire.com
Since DLSS 3 builds on top of DLSS 2 integrations, game developers can quickly enable it in existing titles that already support DLSS 2 or NVIDIA Streamline.
Click to expand...
 
Dakhil said:
Not necessarily.
www.globenewswire.com

NVIDIA Introduces DLSS 3 With Breakthrough AI-Powered Frame Generation for up to 4x Performance

DLSS 3 Generates Entire Frames in Real Time With AI, Benefiting GPU- and CPU-Limited Games and Apps...
www.globenewswire.com www.globenewswire.com
Click to expand...
Based on this keynote it sounds like much of what DLSS3 does requires the new dedicated hardware in Ada GPUs.
"Powered by new fourth-generation Tensor Cores and a new Optical Flow Accelerator on GeForce RTX 40 Series GPUs, DLSS 3 is the latest iteration of the company’s critically acclaimed Deep Learning Super Sampling technology and introduces a new capability called Optical Multi Frame Generation."
Click to expand...
 
Holy hell at DLSS 3.0. There’s no way that can be replicated analytically; optical flow without deep learning just isn’t good enough.

EDIT: this actually may have been wrong! I didn’t realize that there was dedicated hardware to accelerate this on the GPUs.
 
Last edited:
kvetcha said:
Kinda sounds like DLSS3 will require an Ada GPU?
Click to expand...
Waiting for a better technical breakdown, but it looks like the big feature is the generation of new frames from old frame data, and that it uses natively rendered 4k frames to do it? In which case, you'd need a machine capable of doing native 4k at, say, 30fps, and then use this to get to 60 fps. Not sure it would apply to the Switch anyway

Edit: I stand fully corrected, NVidia's press release explicitly says they use the super sampled prior frames to generate new frames "Combining the DLSS-generated frames with the DLSS super-resolution frames enables DLSS 3 to reconstruct seven-eighths of the displayed pixels with AI" which is full on bonkers
 
oldpuck said:
Waiting for a better technical breakdown, but it looks like the big feature is the generation of new frames from old frame data, and that it uses natively rendered 4k frames to do it? In which case, you'd need a machine capable of doing native 4k at, say, 30fps, and then use this to get to 60 fps. Not sure it would apply to the Switch anyway
Click to expand...
Doesn't seem like it requires native 4k frames? Just old and new frame data.

Regardless, not really relevant to the Switch, as the capability seems locked to 4000-series cards.
 
0
Could this new Switch launch at the end of Nintendo's FY if it's going to begin assembly/production soon?
 
0
oldpuck said:
Waiting for a better technical breakdown, but it looks like the big feature is the generation of new frames from old frame data, and that it uses natively rendered 4k frames to do it? In which case, you'd need a machine capable of doing native 4k at, say, 30fps, and then use this to get to 60 fps. Not sure it would apply to the Switch anyway
Click to expand...
I'm assuming it needs to use native res. frames because it wouldn't be capable of doing both DLSS upscaling and then frame interpolation at the same time?
 
0
Yeah, they're explicitly using super sampled frames to generate new frames which is full on nutso.

Using the OFA solves one of the trickiest parts of DLSS which is supplying motion vectors.

At this rate you’ll eventually be able to plug in a single screenshot of Super Mario Bros and DLSS 4.0 will just spit out the entirety of Mario Odyssey
 
0
So Ultra Performance mode is dead for now in DLSS 3, right? It sounds like they are focusing on quarter resolution plus the frame interpolation to get that 7/8 number: 1 - (1/4) * (1/2) = 7/8. Not that it matters, because the tradeoff is obviously worth it if the frame generation is good.
 
Dakhil said:
www.globenewswire.com

NVIDIA Unveils DRIVE Thor — Centralized Car Computer Unifying Cluster, Infotainment, Automated Driving, and Parking in a Single, Cost-Saving System

Wielding 2,000 Teraflops of Performance, Platform Integrates Next-Gen GPU and Transformer Engine to Support AI Workloads for Safe and Secure Autonomous...
www.globenewswire.com www.globenewswire.com
Click to expand...

I look forward starting to speculate about Switch 2 Thor starting in 2025. I just need to know if i have to use "historical" Thor or Marvel Thor.
 
So with DLSS 3.0 providing up to a 4 times performance boost does that mean we can start speculating about the next console after Drake in 2029?

/s

Well not really but maybe joking a little.
 
developer.nvidia.com

Solving Entry-Level Edge AI Challenges with NVIDIA Jetson Orin Nano | NVIDIA Technical Blog

NVIDIA Jetson Orin Nano series system-on-modules (SoMs) deliver up to 80x the AI performance of NVIDIA Jetson Nano and set the new standard for entry-level edge AI.
developer.nvidia.com developer.nvidia.com
Jetson Orin Nano 4GBJetson Orin Nano 8GB
AI Performance20 Sparse TOPs | 10 Dense TOPs40 Sparse TOPs | 20 Dense TOPs
GPU512-core NVIDIA Ampere Architecture GPU with 16 Tensor Cores1024-core NVIDIA Ampere Architecture GPU with 32 Tensor Cores
GPU Max Frequency625 MHz
CPU6-core Arm Cortex-A78AE v8.2 64-bit CPU 1.5 MB L2 + 4 MB L3
CPU Max Frequency1.5 GHz
Memory4GB 64-bit LPDDR5 34 GB/s8GB 128-bit LPDDR5 68 GB/s
Storage–
(Supports external NVMe)
Video Encode 1080p30 supported by 1-2 CPU cores
Video Decode1x 4K60 (H.265) | 2x 4K30 (H.265) | 5x 1080p60 (H.265) | 11x 1080p30 (H.265)
CameraUp to 4 cameras (8 through virtual channels*) 8 lanes MIPI CSI-2 D-PHY 2.1 (up to 20 Gbps)
PCIe1 x4 + 3 x1 (PCIe Gen3, Root Port, & Endpoint)
USB3x USB 3.2 Gen2 (10 Gbps) 3x USB 2.0
Networking1x GbE
Display1x 4K30 multimode DisplayPort 1.2 (+MST)/e DisplayPort 1.4/HDMI 1.4*
Other I/O3x UART, 2x SPI, 2x I2S, 4x I2C, 1x CAN, DMIC and DSPK, PWM, GPIOs
Power5W – 10W7W – 15W
Mechanical69.6 mm x 45 mm 260-pin SO-DIMM connector
Price$199† $299†



Yeah…. I’m not seeing 8N
 
Terrell said:
I dunno, did they make a Shield with a Xavier chip? This also makes mention of T194 in the code sample, or is that just me reading too much into it?
Click to expand...
Thraktor said:
It could be a Shield product, or Jetson (although that would have some overlap with Jetson Orin NX in performance), or it could be a third party product, like an Android tablet or Chromebook or whatever. Hard to tell, but they definitely expect to use this in non-Nintendo products.
Click to expand...
I think T239 being used for a Jetson devkit is unlikely, considering Nvidia recently announced Jetson Orin Nano.
developer.nvidia.com

Solving Entry-Level Edge AI Challenges with NVIDIA Jetson Orin Nano | NVIDIA Technical Blog

NVIDIA Jetson Orin Nano series system-on-modules (SoMs) deliver up to 80x the AI performance of NVIDIA Jetson Nano and set the new standard for entry-level edge AI.
developer.nvidia.com developer.nvidia.com
 
oldpuck said:
Hey OldPuck, it's me again.
Oh hey, Anonymous Lurker, how is it going?
Click to expand...

I have updated this post with CPU info because having a giant update that was out of date the moment I posted seemed like it was going to do more harm than good.
 
0
ReddDreadtheLead said:
developer.nvidia.com

Solving Entry-Level Edge AI Challenges with NVIDIA Jetson Orin Nano | NVIDIA Technical Blog

NVIDIA Jetson Orin Nano series system-on-modules (SoMs) deliver up to 80x the AI performance of NVIDIA Jetson Nano and set the new standard for entry-level edge AI.
developer.nvidia.com developer.nvidia.com
Jetson Orin Nano 4GBJetson Orin Nano 8GB
AI Performance20 Sparse TOPs | 10 Dense TOPs40 Sparse TOPs | 20 Dense TOPs
GPU512-core NVIDIA Ampere Architecture GPU with 16 Tensor Cores1024-core NVIDIA Ampere Architecture GPU with 32 Tensor Cores
GPU Max Frequency625 MHz
CPU6-core Arm Cortex-A78AE v8.2 64-bit CPU 1.5 MB L2 + 4 MB L3
CPU Max Frequency1.5 GHz
Memory4GB 64-bit LPDDR5 34 GB/s8GB 128-bit LPDDR5 68 GB/s
Storage–
(Supports external NVMe)
Video Encode 1080p30 supported by 1-2 CPU cores
Video Decode1x 4K60 (H.265) | 2x 4K30 (H.265) | 5x 1080p60 (H.265) | 11x 1080p30 (H.265)
CameraUp to 4 cameras (8 through virtual channels*) 8 lanes MIPI CSI-2 D-PHY 2.1 (up to 20 Gbps)
PCIe1 x4 + 3 x1 (PCIe Gen3, Root Port, & Endpoint)
USB3x USB 3.2 Gen2 (10 Gbps) 3x USB 2.0
Networking1x GbE
Display1x 4K30 multimode DisplayPort 1.2 (+MST)/e DisplayPort 1.4/HDMI 1.4*
Other I/O3x UART, 2x SPI, 2x I2S, 4x I2C, 1x CAN, DMIC and DSPK, PWM, GPIOs
Power5W – 10W7W – 15W
Mechanical69.6 mm x 45 mm 260-pin SO-DIMM connector
Price$199†$299†



Yeah…. I’m not seeing 8N
Click to expand...
Any idea why there's such a large range for power consumption for these? I guess the amount they throttle?
 
oldpuck said:
kernel.org is the website used to coordinate the development of the Linux kernel, which is open source. Because of the nature of that - lots of companies collaborating on a shared, public project, or even some individual developers who are funded by multiple companies - there is a mailing list that is Quite Old, and very very busy, the Linux Kernel Mailing List, or LKML for short. LKML got so busty eventually you had to bust out sub mailing lists for people to coordinate work that then gets pushed to LKML. lore.kernel.org is the archive off all those emails, so that they're searchable, because actually subscribing to the maillist is an easy way to destroy your inbox.
Click to expand...
Thanks..........I think I understand. Kind of. Maybe? Haha.

I guess I'm still kinda stuck on why emails from an @nvidia email have to be public. I assume this was a mistake obviously - since it gave us info about T239 I wonder if there was some other channel that NVIDIA individual or whoever should have been using so it wasn't added to the archive. Unless the archive is a rule if you're dealing in Linux.
 
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