gaming

Switch 2: the tech specs that Nintendo and Nvidia are not sharing


Nintendo has rarely been forthcoming about the technical specifications of its consoles, giving only the minimum of details and – quite rightly, some might say – focusing on the games themselves. It makes sense, bearing in mind that the firm exited the console specs arms race with the Wii back in 2006. The pattern continues with Switch 2 and its core processor with only the Nvidia partnership confirmed and by extension, the fact that the new machine has both machine learning capabilities (hello, DLSS!) and hardware-accelerated ray tracing. So how much horsepower does the machine have? How does it compare to, say, PlayStation 4 or Steam Deck?

The truth is that we know plenty about the technological make-up of the machine – even without any official disclosures from Nintendo or Nvidia. The story begins on June 11, 2021 where noted PC leaker kopite7kimi discussed a chip called T239, apparently derived from the T234 processor Nvidia uses for the automotive industry and its own Jetson Nano dev kit. Since then, we know from our own sources that T239 may share the same GPU architecture as T234 – the Ampere architecture used by Nvidia for its RTX 30-series graphics cards – but everything else is all-new. The deep learning accelerator and ARM Cortex A78AE are gone, while the 2048 CUDA core GPU is slimmed down to 1536 cores. Meanwhile, a 256-bit memory interface drops down to 128-bit. In the process, the gigantic 455mm2 T234 processor drops down to around 200mm2 – much larger than Steam Deck’s, but fabricated on a less sophisticated, less transistor-dense process.

How do we know that these facts are true and apply to Switch 2? A combination of Linux disclosures for T239 combined with an Nvidia hack allowed sleuths online – principally on Reddit and the Famiboards forum – to piece together an almost complete spec list for the chip. Meanwhile, T239 itself formed part of Nvidia and Nintendo shipping manifests, all but confirming that this is indeed the Switch 2 processor, while confirmed technology like a file decompression accelerator (taking the burden from the CPU for handling assets coming in hot from storage) are very game-based technologies that would have little application for anything other than a games machine.

The latest DF Direct Weekly sees the team tackle what has – and what hasn’t – been disclosed in terms of the Nintendo Switch 2’s technical specifications.Watch on YouTube
  • 0:00:00 Introduction
  • 0:01:11 News 1: Nvidia reveals Switch 2 tech
  • 0:24:57 News 2: Switch 2 follow-up: display and streaming
  • 0:34:42 News 3: Switch 2 comparisons and highlights: Final Fantasy 7 Remake, Hogwarts Legacy, Elden Ring
  • 0:43:35 Metroid Prime 4, The Legend of Zelda: Breath of the Wild
  • 0:53:52 Fast Fusion, Hyrule Warriors, Star Wars Outlaws
  • 1:04:46 Supporter Q1: Is Switch 2 game and console pricing justified?
  • 1:15:07 Supporter Q2: Is the Switch 2 a Steam Deck killer?
  • 1:21:59 Supporter Q3: Why is there such a disparity between Metroid Prime 4 and many third party Switch 2 games?
  • 1:25:47 Supporter Q4: Will developers often target 40fps on Switch 2?
  • 1:32:25 Supporter Q5: What’s the Switch 2’s CPU performance?
  • 1:39:53 Supporter Q6: What processor node shrink will be needed for a Switch 2 Lite?

Controversy dogged T239 as kopite7kimi’s leak that the chip uses the same inefficient Samsung 8nm technology as the RTX 30-series cards seemed at odds with the need for a small, cool chip for a handheld. However, when a fully populated Switch 2 motherboard leaked, showing a circa 200mm2 processor clearly made by Samsung, virtually all doubt fell away. However, this led to doubts about the capabilities of the processor. Steam Deck is over three years old now, using a superior 7nm process for its 162mm2 processor. 7nm itself was used for both PS5 and Xbox Series consoles going back to 2020. The core process for Switch 2 is worse in 2025 – but does it matter?

The only missing part of the specs concerns the Switch 2’s clock speeds. With the chip so large, we should expect low frequencies and this leak – plausible but still unconfirmed to this day – corroborates that. Based on my own expectations from T239 (where I simulated approximate Switch 2 GPU performance by using a wider 2048-core RTX 2050 downclocked to 750MHz), I expected the circa 1GHz GPU clocks, while the 561MHz mobile mode looks sensible. Downclocking memory bandwidth from 102GB/s docked to 68GB/s for handheld play also mirrors what Nintendo did with the original Switch. The only question mark still remaining over these clocks concerns why the CPU frequency in handheld play (1101MHz) is faster than docked (998MHz).

With all of these specifications in mind, I have to say that there’s nothing from last week’s Nintendo Direct or the hands-on demos in the Paris and New York events that look at all surprising. Everything seems to slot in perfectly with expectations, barring a conspicuous lack of DLSS-supported games. Way back when, Activision’s Bobby Kotick described the machine as offering PS4 class power and by and large, that’s in the ballpark of what the third-party demos revealed. Elden Ring runs at 1080p at 30fps or thereabouts on PS4, and that’s what we saw with Switch 2 in last week’s Direct. Despite some briefing from Sega that Yakuza Zero runs at 4K60, all the gameplay assets we’ve seen look like the same 1080p60 as PS4. It’s the same with Final Fantasy 7 Remake: true, Switch 2 gets the Intergrade version but again, it’s running at 1080p and wobbly frame-pacing apart, runs at 30fps.

We’ve since confirmed that Switch 2’s T239 chip has no deep learning accelerator – DLSS processes on the GPU tensor cores only – but otherwise, our old report on the chip seems to be holding up.Watch on YouTube
Nvidia T234 Nvidia T239 (Switch 2) Nvidia RTX 2050 4GB
CPU Architecture 12x ARM A78AE 8x ARM A78C
CPU Clocks Max 1.43GHz 998MHz (docked), 1101MHz (mobile)*
GPU Architecture Ampere Ampere Ampere
CUDA Cores 2048 1536 2048
GPU Clocks Circa 1GHz 1GHz (docked), 561MHz (mobile)* Circa 1.235GHz
Memory/Interface 256-bit/LPDDR5 128-bit/LPDDR5 64-bit/GDDR6
Memory Bandwidth 204GB/s 102GB/s (docked), 68GB/s (mobile)* 96GB/s

* While eminently plausible, we’ve not managed to independently verify these spec points. We’ve included T234 here as the closest mobile component to the Switch 2 processor, while the laptop RTX 2050 is the closest consumer-level GPU to the Switch 2 graphics component.

However, similar to the way the original Switch offered GPU performance seemingly in the Xbox 360/Wii U class, we should expect more – simply by virtue of the fact that although the RTX 30-series Ampere architecture is now getting on for five years old, it supports all the modern features a developer would want: hardware-accelerated ray tracing, mesh shaders and machine learning for DLSS upscaling, just for starters. With 256GB of UFS storage backed by the file decompression hardware, developers should have lightning-fast access to data in a way that the PS4 never had. Cyberpunk 2077 will be an interesting exercise in seeing how Switch 2 handles demanding titles, but more than that, I can’t wait to see Star Wars Outlaws – where the footage shown in last week’s Direct looked perhaps too good to be true. Cyberpunk is a good litmus test, but remember it runs on PS4 and runs well on Steam Deck. Star Wars Outlaws is current-gen only and the game is simply ‘too big’ for the Deck.

We’ve been asked whether Switch 2 is a ‘Steam Deck killer’ and while we’ll only be able to make some kind of definitive analysis closer to review time, our answer for now is both yes and no. More specifically, our contention right now is that the docked Switch 2 delivers more than the Deck – but the jury is out on the mobile experience. Looking at third party games like Elden Ring and Final Fantasy 7, it’s hard to believe that the Deck could deliver native 1080p with the same features and performance Switch 2 is delivering.

However, mobile play may be another story. The Valve machine, running fully unlocked, has a 1.6GHz GPU and a 2.4GHz to 3.5GHz CPU, and I’ve seen it consuming up to 28W of power. Meanwhile, Switch 2 has a mooted 561MHz GPU clock and a 1.1GHz CPU, and based on a 20Wh battery with a minimum two hours battery life, it’ll be sipping a maximum 10W of power during play – almost one third of Steam Deck’s consumption. No matter how much more efficient the Switch 2 architecture may (or may not!) be, it’s hard to believe that the Nintendo machine in handheld mode can best the Deck in terms of raw horsepower. However, Nintendo’s machine has an ace in the hole: the advantages of a bespoke development and a to-the-metal graphics API. That said, let’s just say that I really, really want to see Star Wars Outlaws running on Switch 2. Getting that running to the standard seen in the Direct will be a challenge for docked play, let alone the more resource-constrained mobile mode.

Right now though, I’m wondering what has happened to Nvidia DLSS – conspicuously absent in the form we recognise within last week’s Nintendo Direct (with the possible exception of Hitman: World of Assassination – we’ve asked IO Interactive to confirm). In this week’s DF Direct, the number one question asked was what form – or version – of DLSS Nintendo Switch 2 supports. That’s difficult to ascertain with little actual evidence it’s in use in the games we saw, but the bottom line is this: a GPU using the Ampere architecture will have the requisite complement of tensor cores used to run ML workloads. There’s nothing stopping it running any version of DLSS from 2.0 upwards – even DLSS 4’s transformer model. However, DLSS is not a free lunch – it has a computational cost. It’s not so much if Switch 2 can run DLSS 4, it’s more like whether it is too expensive to do so and that’s a factor of how capable the GPU is.

So, we’ve seen what Switch 2 is capable of, and despite very selective disclosures from Nvidia and Nintendo, all of the leaks coalesce into something as close as we’re going to get to a confirmation of a trickle of information that began over four years ago. We may actually see those specs change over time though: the original Switch in handheld mode evolved from offering a 307MHz GPU clock to anything up to 460MHz over time as Nintendo offered more performance to developers in exchange for lower battery life. However, the core architecture is locked: out of date compared to more modern technology but clearly capable – and of course, the nature of these apparently dated components can only mean good things for Nintendo from a business perspective as it begins to map out the generation ahead.

First of all, with a main processor based on a process Nvidia hasn’t used for over two years now, there’s every chance that Nintendo secured a very good deal from Samsung in making the T239 processor – and this would be a major win in a world where the cost of silicon is only going up. Secondly, just like the Tegra X1 in the original Switch, Nintendo can also look forward a few years to a time where using an improved process technology becomes more affordable. This translates into a revised unit with much longer battery life and the opportunity to launch a smaller ‘lite’ model – and maybe even an OLED model if the technology becomes available.

For now though, it’s all about the launch games and we’ll report back with our own hands-on impressions in due course.





READ SOURCE

This website uses cookies. By continuing to use this site, you accept our use of cookies.  Learn more