Watch On
A brand new video from the ever effusive Geekerwan has damaged down the small print of the silicon on the coronary heart of the Nintendo Change 2. He picked up a bricked console motherboard for 1,000 Yuan (approx. $140) from a Chinese language eBay analogue, Xianyu, and has completely gone to city on the chip.
Like, above and past.
I completely suggest you watch the video, as a result of in addition to being extremely detailed on the extent they went to with a purpose to get the complete particulars of the chip—together with slicing it down the center with a really skinny ion beam with a purpose to measure a cross-section to establish the method node—he is super-engaging and all the time obsessed with what he’d overlaying. It is a pleasure, truthfully.
Anyhow, the principle takeaway is that sure, that is the Nvidia T239 chip that was broadly rumoured to be the chip powering the brand new Nintendo handheld, and you’ll see from the metallic layer inscription that this was a processor taped out in 2021. Geekerwan notes that this four-year hole is a very very long time between tape out and launch, and certain hints at a Change 2 that was speculated to be launched some time again and was postponed for no matter motive.
It is Nintendo; the explanation could possibly be something.
As a result of it is an already damaged chip—they tried powering it on, however acquired nowhere—there was nothing stopping them from ripping the Nvidia chip open to see what was happening underneath the brand’d heatspreader. Lapping the highest meant they might get an awesome die shot, highlighting that whereas this does certainly home Ampere-generation Nvidia {hardware}, “this Ampere isn’t the opposite Ampere.”
There are particular variations between an RTX 30-series die and the format of the GPU part within the T239 chip. Geekerwan notes that the SM design really seems to be far nearer to the format of an Ada SM than an Ampere one, which might counsel that there are some parts shared between the GPU generations on this chip.
There are 12 SMs on this board, with 128 CUDA cores in every, for a complete of 1536 cores. That is what the rumours had already stated, in order that’s not a shock right here.
One other factor which appears to straddle two generations is the method node itself. Theoretically, it is a Samsung 8N node, however their examination suggests this can be a manufacturing course of that shares options of each Samsung’s 10 nm and eight nm nodes. That is, nonetheless, a customized course of, in order that does make sense.
With the GPU elements unusually wrapped round it, there may be the CPU phase. That is made up of eight Arm A78C cores, every with 256 kB L2 cache, with a shared 4 MB of L3 within the center.
In addition to ripping the wrapping layers off this silicon current, Geekerwan has additionally made an try at making an attempt to dial into the potential efficiency of the Change 2’s core. With a equally Ampere-based GPU, the GA107 of the RTX 2050 laptop computer chip is the closest match, although that comes with a far larger TGP and reminiscence and core clocks.
With a view to try to get a bead on gaming efficiency, they hobbled their RTX 2050 laptop computer as finest they might to match the rumoured handheld frequency of the Change 2—421 MHz—and halved the reminiscence clock. To get that low, they wanted to make use of inside instruments to convey the goal TGP and frequencies proper down.
Taking 3DMark Metal Nomad Mild as their GPU take a look at, you’ll be able to see that in handheld mode, the Change 2 really has an identical degree of efficiency because the Steam Deck, with the docked mode simulating 53% larger efficiency than Valve’s traditional handheld.
Although it is value noting that even in docked mode, it could’t contact the Radeon 780M, which powers many of the remainder of the hand held gaming PCs in the marketplace.
One factor they do word, nonetheless, is that whereas there’s a important distinction between the unique Change chip and the theoretical efficiency of the brand new one, it is not the 10x that Nvidia has been touting. At the least not with out DLSS, and possibly even Body Gen. Keep in mind that we stated there have been similarities between the Ampere GPU within the T239 and the Ada design?
Geekerwan then goes on to point out a simulated Cyberpunk 2077 run on the anticipated DLSS ranges {that a} Change 2 model, in each handheld and docked mode, would possibly ship. And, effectively, it is not dangerous, particularly once you assume it’s going to be on a comparatively wee display screen. He did activate path tracing, too, however the much less stated about that the higher.
The identical goes for Monster Hunter Wilds the place he says, “even if you happen to run 720p, low-definition, super-performance DLSS in dock mode, it turns into a shitty wilderness.” Although Black Delusion Wukong does not look dangerous, and the newest CoD additionally runs fairly effectively on this simulated {hardware}.
As you would possibly count on, it is the CPU part that’s approach behind the GPU half. The Arm cores run at 1.1 GHz, and in a Geekbench 6 evaluation, that places the simulated efficiency 140% behind the Steam Deck APU’s single-core efficiency and 52% behind on multi-core metrics.
It is a actually attention-grabbing breakdown of the Change 2’s potential efficiency, and an awesome take a look at the precise silicon, too. And, once more, I say go watch the video, this man’s nice.