- Jul 27, 2016
So 14/16nm refresh, likely the same as what is in the Switch Lite... and uh a different "type" of NAND? I don't know whether that refers to MLC/TLC/QLC, overall capacity or even the amount of chips on the board. I suppose a NAND shrink can end up with different dimensions, but would seem an excessive declaration if they were just changing suppliers.
It would have to be on a node lower than 14nm/16nm.
The Switch Lite has a 3570mah battery which is 13.2 watt hours compared to the current Switch 4310mah 15.9 watt hour battery.
Nintendo says you can get four hours playtime of Zelda BOTW with Switch Lite compared to three hours with the current Switch;
that would mean the Switch Lite is using 3.3 watts per hour running BOTW, and let's just say the screen ,speakers and reading the cartridge add up to 1 watt(?), which means the Tegra SOC would be drawing 2.3 watts an hour.
The Jetson Tegra Xavier with 512 cores on 12nm gets 1.4 tflops FP32 and 11 tflops FP16 at 30 watts, which means at 2.3 watts it would get around 108 gflops FP32.
Zelda BOTW supposedly runs around 157 gflops on the current Switch.
So, the Nintendo Switch Lite has to have a chip that has 512 cores on 8nm or 7nm DUV.
I don't think Nvidia would die shrink Turing just for Nintendo , so Ampere maybe?
Nvidia is putting Ampere on 7nm EUV, but they likely would have taped it out on Samsung's 8nm LPU or TSMC's 7nm DUV first, in case EUV didn't pan out.
According to a Ubisoft developer they could get 60%+ of the processing they need from FP16 flops
If the Switch Lite when at full power and using FP16 flops for say 50% of the needed flops, possibly could run next gen games at 720p.
So the revision for the current Switch possibly could run next gen games at 1080p and a pro at 2k upscaling to 4k. ,.... or maybe not.