DF Weekly: What does the Switch 2 motherboard leak say about system performance?
Should we be concerned about the Samsung-produced processor?
After a one week hiatus, Digital Foundry Direct Weekly returns! The truth is, we were hoping to rest the show for a second week but the barrage of Switch 2 leaks and news put paid to our plans to more casually ease ourselves into 2025. Kicking off the show is our reaction to the reveal of a fully populated Switch 2 mainboard, photographed on both sides, at various angles. The photos are almost certainly genuine - and it's the reveal of the handheld's main processor that has caused the most controversy with confirmation that it's Samsung handling production duties based on an Nvidia design.
Quite why this news is so important comes down to ongoing discussions since the Switch 2's T239 processor was first leaked. Renowned - and highly reliable - leaker kopite7kimi revealed way back in 2022 that Nintendo and Nvidia were developing a T239 processor for its next-gen handheld. Seemingly a cut-down version of the massive, hulking T234 used in the automotive industry, kopite7kimi believes it is fabricated on the Samsung 8nm process, as used for T234 and Nvidia's entire range of RTX 30-series consumer GPUs. At 455mm2, T234 is just too vast and power-hungry for a handheld and yet the motherboard shots reveal that it's clearly a Samsung processor within Switch 2. If it's another 8nm chip, this has serious implications for either performance or battery life - especially when the battery compartment looks much smaller than Steam Deck's.
I'd estimate the new SoC is around 200-220mm2 based on the size of the surrounding memory modules, believed to be in the region of 196mm2. This is much, much higher than Steam Deck LCD's 163mm2 and the OLED revision's 131mm2. While some believe that Samsung 8nm is not dense enough to contain the logic in a 200-220mm2 area, I'd beg to differ. Switch 2 apparently has 1536 Nvidia CUDA cores while Nvidia's RTX 3050 crams in 2560 of them into a 200mm2 chip. Cut back the CUDA cores, expand the footprint of the chip and I think it is quite feasible to fit Switch 2's CPU, GPU, media and file decompression logic onto the chip we've seen leaked. That said, there are theories that another Samsung process may be in effect, perhaps 5nm. Hands-on with final hardware, we may get some idea via power consumption metrics but ultimately, the whole 8nm question will only be categorically answered when the likes of TechInsights have thoroughly analysed the product.
- 0:00:00 Introduction
- 0:00:50 News 1: Switch 2 motherboard leaks - with possible 8nm SoC
- 0:27:47 News 2: Nintendo patent hints at Switch 2 upscaling
- 0:43:45 News 3: High quality GTA 6 trailer reveals new details
- 0:55:35 News 4: Wukong developer says Series S holding back Xbox port
- 1:07:41 News 5: AI update - new GPTs, game tech
- 1:19:38 Supporter Q1: Does it make sense to pair an AMD CPU with an Intel GPU for a budget build?
- 1:26:02 Supporter Q2: Will Sony ship games like Marvel's Spider-Man on Switch 2?
- 1:31:23 Supporter Q3: Are you saving some performance-challenged Switch games for Switch 2?
- 1:35:26 Supporter Q4: Is this the most exciting Q1 in gaming history?
Whichever way you slice it, the news of a Samsung chip does put the Switch 2 at an efficiency disadvantage up against Steam Deck and the vast array of Windows PC handhelds out there. These are based on fabrication technologies from TSMC - the most successful chip maker in the business right now. With a Samsung chip, clock speeds will be lower and power efficiency won't be as good. However, equally, I'd argue that Switch 2 has its own 'secret sauce' that none of these devices have: the power of a fixed platform.
Last year, the DF team took a look at the best tech from Switch 1 games, coming up with a range of extremely impressive first and third party offerings. This year, it is a whole decade since the Tegra X1 mobile processor was announced and yet, from 2017 onwards, that same processor - downclocked significantly from standards specs - delivered hit after hit on the Switch platform, to the point where we coined the 'impossible port' phrase when we first saw Doom 2016 running on it. I'd also point you towards Oliver Mackenzie's remarkable video comparing the existing Switch to iPhone 15 Pro. True, it's an actively cooled chip up against a thermally challenged, fanless design, but even so, the veteran processor holds up rather well, all things considered, bearing in mind its vintage.
There's no reason why a Switch 2 with a Samsung-produced processor won't do the same throughout its lifespan, particularly as the fundamental Nvidia technologies are so good. Eight ARM A78C cores pair with Nvidia's Ampere architecture to deliver machine learning features and even ray tracing. Expect loading times to be considerably reduced owing to the inclusion of a custom file decompression block.
Another advantage Switch 2 has over its predecessor is that T239 is specifically built for the purpose at hand - even if there are challenges with the process technology, they would have been factored into the design and thoroughly tested during the simulation phase before the first chip was manufactured. Ultimately, we need to trust that Nintendo and Nvidia know what they are doing - but this is still a better place to be compared to Switch 1, where Nintendo effectively had zero input into the chip design. And let's not forget the negative reaction/outrage when we revealed the original Switch's clock speeds: 150m unit sales later, everything turned out fine.
In this week's Direct, we also tackle the online discourse surrounding the idea that Switch 2 will offer performance in line with PS4 Pro (when docked, at least). I'm not entirely sure where this notion came from, but in a world where AMD's top-of-the-line Strix Point processor, running on the latest TSMC process, is nowhere near Pro's rasterisation performance, I'd say that this is wishful thinking - which often surrounds new console speculation. Think more along the lines of base PS4 graphics performance, augmented with modern GPU features and solid ray tracing support. I'd also expect Nvidia DLSS to be doing a lot of heavy lifting.
Speaking of which, this week's Direct also tackles Nintendo's upscaling patent which - on the face of it, at least - looks to be a thorough description of Nvidia DLSS, which is certainly an interesting take on the Nintendo/Nvidia partnership. However, as Alex Battaglia discusses, there's much more to it - it's a kind of grab-bag of various upscaling techniques for various scenarios, none of which seem to be particularly new or patent-worthy. However, there is one very interesting element to it that's worth discussion.
In my T239 simulation tests using a (very) low-power Nvidia laptop GPU, the computational cost of DLSS proved challenging. DLSS isn't a free lunch, processing time scales with output resolution. In my tests with Death Stranding, while DLSS upscaling from 720p to 1080p and even 1440p were viable for a 30fps game, 4K was off the table. DLSS cost was around 18ms to upscale to 4K - in excess of the 16.7ms frame-time to run at 60fps, and over half the frame-time budget for a 30fps experience. The patent is interesting in that it accepts that machine learning-based upscaling is challenging and suggests that the complexity of the AI model could scale according to target frame-rate.
I'd expect to see Switch 2 announced soon and it'll certainly be interesting to see third-party developers break cover with product announcements. If you watch the video above, you'll see that ballpark GPU performance can run a great many games at highly respectable quality levels - and I suspect that based on my projections, a vast array of PS4-era ports will appear in addition to Nintendo's first-party fare - and I wouldn't rule out ray tracing support in certain games. I'd also expect current-gen titles targeting 60fps to somehow find their way across to Switch 2, likely running at 30fps instead. The next few months should prove fascinating.
In the meantime, do enjoy this week's Direct-that-almost-never-was. There's a lot of great topics up for the discussion - the unearthing of a higher quality version of the GTA 6 trailer and what we've learned from it, Oliver's thoughts on the latest innovations in the AI technology space, plus reaction to the comments from Game Science that Black Myth: Wukong poses too many issues to run on Xbox Series S. There's a certain amount of logic that contradicts that statement based on the PC version, but at the same time, the PS5 version suggests that Game Science had issues with that port. If memory isn't an issue as the PC version suggests, why was texture quality degraded on PS5 when no such reductions were needed on 8GB PC graphics cards? It's a bit of a mystery.