• 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.
  • General system instability
    🚧 We apologise for the recent server issues. The site may be unavaliable while we investigate the problem. 🚧

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

Reader mode
Thraktor said:
We know from a variety of sources that the T239 chip is Drake, ie the one being used in the new Switch model. It having "eight cores in a single cluster" means it has 8 CPU cores (obviously), but doesn't strictly speaking limit what those cores can be, as ARM's DynamicIQ supports clusters with mixes of different cores (eg 4x A78 and 4x A55). However, as the code would almost certainly have to be changed to support these hetergeneous clusters (ie different core types would run at different frequencies), and the commit here doesn't
Click to expand...

I think I can speak for most of the lurkers here when I say, Thraktor, there is no "obviously" when it comes to this stuff. Though after a year here I've gleaned a lot so it's not entirely gibberish anymore.

Hidden content is only available for registered users. Sharing it outside of Famiboards is subject to moderation.
 
Last edited:
Procheno said:
And also the 8 core thing is not for some other device that happens to use the same tech?
Click to expand...

In this instance the open-source Linux kernel running on T329 would be for another device like an Nvidia Shield or Chromebook, since Switch 2 will run a closed source OS, but the overall specs of T239 would not be different. It'd be 8 cores in both a Switch and a Shield. It's their advantage to reuse the chip like they did with Mariko, which was also used in the Shield TV +.
 
Z0m3le said:
I'll just make it simple. Nvidia was hacked, from that hack, we saw that Switch 2 will use T239 codenamed "Drake" (this is a Nvidia codename). This 8 core leak is from September 5th, just 2 weeks ago, and is real, this is the linux kernal, which also means that this is a chip in production right now, as os support for a CPU points to real hardware.

From the above, we can know that the specs are
1536 Cuda cores (Steam Deck has 512 cores for comparison) also 48 Tensor cores and 12 RT cores.
8 core A78C CPU (better IPC than Ryzen 2, but will have a lower clock than current gen consoles)
128bit memory bus, 88GB/s - 102GB/s for LPDDR5, which is the memory type based on Orin's relation to T239.
Click to expand...
wait what, do you have a source for the 8 core leak? Been out of the loop for a while.

Edit: nvm I’m up to speed.
 
Last edited:
Hermii said:
wait what, do you have a source for the 8 core leak? Been out of the loop for a while.
Click to expand...
Check a few posts beforehand! Someone found a Linux kernel update request from an Nvidia employee from just a week or two ago. Not even sure it qualifies as a leak. It's just...information.
 
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
 
Cheppito said:
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
Click to expand...
I think with the T210 the issue was that the big.LITTLE implementation did not allow for both clusters to be used at once. It was an either/or situation, so Nintendo just disabled the LITTLE cores entirely.

I imagine that's no longer an issue.
 
Cheppito said:
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
Click to expand...
No, as Thraktor points out, because the code isn't specifying more than one core type, it will just be a single cluster of 8 A78C cores. Otherwise multiple frequencies would be needed.
 
Cheppito said:
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
Click to expand...
I can't remember if it was ever confirmed or not but there was talk that they fused off those A53 cores physically, not just that they left them unused. That wouldn't be doable if they're all in the same cluster.

IIRC the A53 cores were barely used for any T210 product.
 
TheSpearGuy said:
Keep in mind that Nintendo will underclock this new device for more battery life
Click to expand...
There's a limit to how low they can go though, which is kind of the exciting part. Especially considering they'll want to ensure this can be BC with existing Switch games, so it's unlikely it can be clocked below current Switch clocks.
 
Skittzo said:
There's a limit to how low they can go though, which is kind of the exciting part. Especially considering they'll want to ensure this can be BC with existing Switch games, so it's unlikely it can be clocked below current Switch clocks.
Click to expand...
(not to mention Ampere can't go lower than 300Mhz before it hits the bottom of it's power-curve which is also pretty much where OG Switch is for its lower portable clock)
 
Z0m3le said:
Yeah, I disagree with the 2GHz number, we have heard in the past that Nintendo wanted to push A57 above 2GHz, but more importantly, we've seen them push A57 on 20nm to 1.78GHz, that is pretty close to 2GHz, so Nintendo does want the CPU performance, it's also worth noting that because of DLSS, the GPU's performance just isn't as important as the CPU's IMO.
Click to expand...

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). :)


mjayer said:
It can’t be regardless of node! No one just leaves untappped potential lying on the ground. If there is overhead to go higher for the best efficiency of the chip, it will.
Click to expand...

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.
 
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.
Click to expand...
Probably because CPU clocks remain constant between portable and docked profiles and so they had to carefully consider power consumption on their shitty 20nm node.
 
Cheppito said:
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
Click to expand...
No, they aren't doing the same. Last time around, that wasn't a choice by Nintendo - DynamiQ ARM CPUs became a thing just after the Switch's launch. Before that, the A57 and A53 clusters couldn't be used together, at the same time. They opted for the performance cores, which SHOULD tell everybody that they tried to make it as powerful as they could without heat issues, or making the portability element redundant, or product failure. The original T210 SoC couldn't use both clusters at once, either. But ARM CPUs of today can do that. The A78C variant exists in hexa-core and octa-core form ONLY, and information presented in this thread has reaffirms what I have said for quite a while.

What isn't clear here, is whether the A78C can do varying frequencies (for example, 4 cores at 2.8GHz, and the other 4 at 1.7GHz for performance and efficiency CPU core design within the single cluster), but I suspect that it'll land at an average of 2-2.2GHz per core.
 
0
Wow, what an incredible post from a new user. This got me super excited!

I guess the question now is, TSMC 6NM or TSMC 4NM? We know nvidia was facing fines for reduced 40 series gpu demand vs the capacity reserved. So which node do you think is more likely?

I am seriously entertaining the idea this SoC is manufactured on the same node as the 40 series GPUs. If this is the case a 2GHZ CPU Clock and that 1.3GHZ docked GPU clock seem very reasonable.

In fact I'm happy to stake that as my prediction for this device.

4nm.
8 A78C @ 2GHZ
12SM @ 460MHZ Handheld, 1.3GHZ Docked.
12GB LPDDR5 RAM @102GB/S
256GB EUFS Storage.

Will make the steam deck look like a toy, Let's go!

Oh damn I am going to have to buy a new Shield TV aren't I?
 
Notsowise said:
Guys, is it feasible that we see any new portable hardware from nvidea today? Something that may fit into the new switch?
Click to expand...
I could see them revealing T239/Drake for use in Laptops/a Sheild TV 2 as an extra thing with RTX 4000 (to help keep hype up considering the belated release schedule rumours.etc) but anything else than that would be pushing speculation a bit.
 
0
Skittzo said:
IIRC the A53 cores were barely used for any T210 product.
Click to expand...
As I understand it, they weren’t used in any; they were silicon errata and mention of them was eventually removed from Nvidia’s marketing for the chip to its partners and later technical documentation. Which is potentially why the X1 only ever saw one design win in a tablet, far fewer than its predecessor, the K1.

Fortunately, Nintendo was looking to make pretty much the perfect device for the final chip’s use case – something that needed consistent graphical performance and didn’t really need to throttle down into an ultra low-power state when not gaming – and they just so happened to be looking for an off-the-shelf chip they can use.
 
kvetcha said:
Probably because CPU clocks remain constant between portable and docked profiles and so they had to carefully consider power consumption on their shitty 20nm node.
Click to expand...

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.
 
For us, this was the most "happening" day in a long while.

Fot nVidia, it was:

flat,750x,075,f-pad,750x1000,f8f8f8.u4.jpg
 
kvetcha said:
PS4 is not going to be a problem at all. Drake should be close enough to Series S to enable ports.
Click to expand...
Yep. Modern gaming engines are highly scalable as far as graphics are concerned, and Drake has the advantage of being able to run at a lower resolution and use DLSS. The CPU was always going to be the larger sticking point as far as the potential for ports was concerned, so eight symmetric cores in encouraging.

Now we just need to know RAM.
 
LoneRanger said:
Octa Core CPU is the best possible outcome. Hopefully there is 1 for OS and 7 for developers.

But I would still as always said temper my expectation regarding clocks. My expect for CPU/GPU are the same Switch clocks. 1GHZ for CPU and 768/307 MHz for GPU in docked/portable modes. I know that would be a 8x jump in CPU + 6x jump in GPU + DLSS + new architecture + 3x RAM jump
Click to expand...
It would always be 1 OS core for devs as they have a pretty light OS as is.

If it was heavy then yes.

In any case, they can stretch to 1.9GHz for 7 cores and 1 OS core to 1.3-1.6GHz and consume ~3W just fine for the CPU


And this would mean that the device is only about half as strong as the series S with SMT On.

Simba1 said:
Like people said, clock is big factory.
Drake CPU clock will probably be somewhere around 1.5GHz, that should mean its noticeable stronger than PS4 Jaguar 1.6GHz CPU,
but noticeable weaker than Xbox Series S Zen 2 3.6GHz CPU (thats basically full desktop CPU).
Click to expand...

See above before this quote


(Series S is clocked to 3.4GHz with SMT on, 3.6Ghz with SMT OFF, and if it was 1.9 it would be about 20% weaker I think than the series S)
 
Cheppito said:
Very exciting news! I remember OG Switch has a custom T210 which didn't use half the cores the original T210 had (A53 cores me thinks). You think they're doing the same this time or maybe that's not happening because they're all in the same cluster?
Click to expand...
Aside from clock speeds, there was nothing custom about the tx1 in the switch.
 
Mr.Gamerson said:
Will the CPU overclocking for loading still be useful on this chip or will the CPU be fast enough at base clocks to allow for the max transfer speeds?
Click to expand...

Depends on what storage they use I imagine which is still something we have no idea of. There is quite the difference Between eMMC 5.1 and EUFS 3.1.

I imagine the clocks as is would easily saturate eMMC and microsd, but anything above that, who knows.
 
0
JoshuaJSlone said:
I mean, it's different hardware, so 100% it will not be what's there today? If you mean if the OS will be radically redesigned or offer new features, that's something else.
Click to expand...
They've been updating Switch OS consistently for years now with many features coming in the last year alone. With the fact that the OS actually supports 1440p and 4K, why on earth would they move to a new codebase when it's running the same games for the most part, at least at launch? Nintendo has been yearning to simplify their development pipeline for years. OLED Model, Switch and Switch Lite all need slightly different options set from the factory, but use the same OS and manual. I imagine this will be the case going forward.
 
0
TheSpearGuy said:
Yeah, but I don´t see it being on par with an Series S neither a PS4
Click to expand...
Given the amount of cores and architecture that we're actually extremely confident about here it would be very very difficult for this not to outperform the PS4. The CPU alone will run circles around the PS4's, even at current Switch clocks.


Series S yeah, there's no way it'll be up there but it should be close enough to receive ports, probably a good deal closer than Switch was to XB1.
 
Ojoloco said:
Now, we need to know the ram
Click to expand...
If you want to believe someone in this thread who claims to have contacts it's 12GB LPDDR5.

If you don't want to believe them it's still fairly likely anyway considering what modules are currently cheap and available.
 
ReddDreadtheLead said:
Just wanted to address the first part: is it? I mean there’s some geekbench that would outright delineate as three clusters but sees it as 1 processor 8 cores.

Like this for example:


gsmarena_001.jpg



Wouldn’t this also apply here?
Click to expand...

Well, we're getting into the difference between physical clusters and logical clusters. From the point of view of the physical hardware, DynamicIQ "combines the big and LITTLE CPUs into a single, fully-integrated cluster", but software may treat the cores as belonging to separate logical clusters (similar to how with multithreading, you will often see software report twice the number of CPU cores, as they're treating each thread as a logical core). In theory this code could be dealing with logical clusters instead of physical ones, but I don't suppose it matters much, as T234 only has one core type per cluster, and there don't seem to be sufficient code changes here to support multiple core types per cluster for T239.
 
Sigh I just wrote a big summary post too...

Cybergatuno said:
Hi guys,

First post here :cool:

It seems NVidia is adding support for Tegra239 into the Linux Kernel.


Well, this seems like a confirmation.
Click to expand...

Excellent basic detective work. Thank you, you've eliminated an entire class of arguments from this thread :)

Thraktor said:
Furthermore, the fact that T239 support is being added to the Linux kernel means that Nvidia plan to use it in products other than the new Switch. This isn't particularly surprising, as the Mariko chip (T210B01) was also created for Nintendo, but used in other products (Jetson Nano), so Nintendo don't seem to have any problem with their semi-custom parts being used elsewhere. It likely allows them to get a better deal from Nvidia.
Click to expand...
All excellent points, especially re: the cost. I wonder if this actually reduces the chance we see A78C, which are built for sustained performance, but might limit the flexibility of the device.

Thraktor said:
Lastly, code being updated to support T239 likely means that there is actually hardware to run this code on. That is, at the very least, engineering samples of T239 already exist, meaning full scale production of the chip is likely to start soon if it hasn't already.
Click to expand...

It looks like the maintainer here has a history of importing internal linux dev at NVidia and then being the "face" of merging those branches with mainline. At the point at which we're ready to get in tree, it seems highly likely chips are coming off the line, as internal teams are talking to each other.

Alovon11 said:
(not to mention Ampere can't go lower than 300Mhz before it hits the bottom of it's power-curve which is also pretty much where OG Switch is for its lower portable clock)
Click to expand...
You've mentioned this before - any source for this? Not doubting you just so I know?
 
oldpuck said:
All excellent points, especially re: the cost. I wonder if this actually reduces the chance we see A78C, which are built for sustained performance, but might limit the flexibility of the device.
Click to expand...
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?
 
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.
 
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