DavidGraham
Veteran
The whole PCB of the Switch 2 has now leaked with high res pictures, NVIDIA SoC confirmed, 12Gb LDDR5X SK Hynix confirmed, USB-C also confirmed.
Last edited:
Switch 2 Speculation
- Thread starter Goodtwin
- Start date
How "Switch 2" is this versus more "Switch Pro"? Anything on it making it a generational advance? I'm guessing DLSS will mean more direct PC ports making it kind-of SteamDeck-with-bonus-Nintendo-games option.
DavidGraham
Veteran
6x the GPU cores, 3x the amount of RAM, 2x the CPU cores. The GPU arch also goes from Maxwell to Ampere, which is a huge IPC gain + new features (Ray Tracing + Machine Learning + DX12U).
Last edited:
Charlietus
Regular
Also, going from 256 Maxwell cores at a low frequency to 1536 ampere cores at a higher frequency. That's a generational jump. Probably the last one as far as raw power goes for consoles.
Charlietus
Regular
Looks like Switch 2 is on the Samsung 8nm process. That's a shame.
Source: the signature on the chip says 'SNW8VF' - the 'SN' referring to Samsung.
Source: the signature on the chip says 'SNW8VF' - the 'SN' referring to Samsung.
Subtlesnake
Regular
Samsung has other processes besides 8nm, and ~200 mm^2 should be too small to fit the GPU + 8 core A78 + cache/IO on 8nm. The 12 core AGX Orin has a die size 448mm^2 on that process.
Charlietus
Regular
There is also an 8 in the signature (SNW8VF), so maybe something like Samsung W(?) 8nm VF(?).
Maybe some kind of modified 8nm process?
Subtlesnake
Regular
The "than is needed" is the important bit. Based on the Famiboards analysis, T239 is too big to be reasonably clocked down in portable mode on 8nm. So that suggests Nintendo have a different performance target, otherwise they would have designed a smaller chip. Unless there is a special variant of 8nm with different performance characteristics.
But nVidia hasnt any design on any other Samsung process. And why would Nintendo pay nVidia to port something to another Samsung process when they could go to TSMC 7nm DUV process?!
Subtlesnake
Regular
I guess only for cost reasons? But I am not the one pointing to SN in the chip signature. I am just making the point that 8nm doesn't fit either, if the chip is really ~200mm^2 and based on what we know about the power characteristics of the process.
Last edited:
DuckThor Evil
Legend
Subtlesnake
Regular
Well 0.6 of 200 gives 120, so that leaves 80 mm^2 for the CPU and residual IO, or 3.48 billion transistors. But the die size won't scale linearly with the number of SMs, so in practice you would expect to get less space for the CPU. I guess it might be possible after all, but it seems pretty tight.
Edit: Also the issue is not getting PS4 performance in a console form factor, which should be no issue, but getting it at <15W. (Even 12W for the whole SoC would be an upgrade since the Switch 1 consumed ~9W when docked).
So you expect Switch 2 to be more powerful than SteamDeck? While simultaneously being thinner and have more battery life?
The Steam Deck doesn't match the PS4 btw.
Larger die, lower clock rates. Best way to improve efficiency. This is the reason why 8nm makes sense. 8nm is so much cheaper than any other smaller process. And the whole eco system is optimized for games.
I think the Switch 2 will be more efficient but not more powerful in the handheld mode.
I think the Switch 2 will be more efficient but not more powerful in the handheld mode.
