Soundwave said:
Getting back to the actual topic, I'm going to take a shot at what I think the Switch 2 specs are based on the Nvidia leak from a year+ ago and merge that with the news from CVG that The Matrix Awakens demo is running on said chip. So just as a quick recap ... (you look Google this stuff yourself if you want more detail on it).
About 18+ months ago a very, very reliable Nvidia leaker named Kopite (has leaked multiple Nvidia GPU things that were 100% correct) basically said the next-gen Switch chipset would was designated as Tegra T239. Tegra T239 didn't really exist at that point, it wasn't an officially announced product. Few months later an Nvidia employee accidentually lets it slip that a Tegra T239 indeed exists.
Then there was a data hack of Nvidia, and sure enough the Tegra T239 is in Nvidia files and people were able to glean a bunch of data about the chip. For starters the big thing was the graphics API for this Tegra T239 chip is listed in the Nvidia data as NVN2, which is a new graphics API. The Nvidia graphics API for the Switch 1 (and only the Switch) is NVN1. So there you go. But from the leak also basically shows the CPU and GPU configuration of this chip as follows
Tegra T239 codename: Drake
8-core ARM A78 (A78AE?) CPU cores
12SMs, 1536 CUDA cores (6x increase over Switch 1)
DLSS and RTX (ray tracing) hardware support
Graphics API is known as "NVN2" (Switch 1 API is "NVN").
This is from Nvidia's own Linux kernel data leak.
This is from Famiboards, but a rough outline of what the Tegra T239 would theoretically output at different clock speeds (thanks to Z0m3le)
Portable
400MHz ->1.228TFLOPs
500MHz ->1.536TFLOPs
600MHz ->1.843TFLOPs
Docked
768MHz (Switch docked clock) ->2.36TFLOPs
1GHz -> 3.07TFLOPs
1.152GHz -> 3.539TFLOPs
1.3GHz -> 3.993TFLOPs
1.5GHz -> 4.6TFLOPs
1.8GHz -> 5.53TFLOPs
2GHz -> 6.14TFLOPs
The current Tegra X1 clocks at 1 GHz for the top end ... but because the 20nm version of the chip was inefficient and ran a bit hot way back in 2015, Nintendo downclocked that to 768 MHz docked. The Tegra X1 in Nvidia's Shield miniconsole can still output at the full 1000 MHz. When the Tegra X1 was die shrunk to 16nm, the high end clock speed of the GPU increased again to 1.27 GHz max for the Mariko Switch/Lite models. Steam Deck by comparison clocks its GPU at 1600 MHz (1.6GHz).
So if this Matrix Awakens report is true, then I am guessing the Tegra T239 is running at 1 GHz docked. That was the original clock speed of the OG Switch and is still lower clocked than the more modern Mariko Tegra X1 in Switch Mariko/OLED + Lite. So it's not an absurdly high clock. It's a lot lower than Steam Deck's GPU clock, it's a good deal lower than the Mariko 16nm Tegra X1's top end too.
But at 1 GHz docked you're getting 3 teraflops performance ... 3 teraflops is knocking on the door of the XBox Series S (4 teraflops) and when you add DLSS on top of that, could a chip like that run the Matrix Awakens demo at a quality that is worth demoing? With native res at 600p or something DLSSed to 1440p lets say? I would say yeah. That sounds actually right on.
|
Z0m3le that guy is a clown. He was the one that hyped FP16 to heavens and said switch with FP16 would basically match xbox or exceed it in some cases and everyone believed him lol. right when i saw his name i stopped reading. About street fighter 2 gif thats just looks like sightly different color use and comparing 30fps-60fps is a much bigger jump, night and day actually if you run side by side.
Last edited by zeldaring - on 10 September 2023
