
Originally Posted by
Kosmos992k
Nintendo Switch is almost 100% certain to use an nVidia Tegra X2 System on a Chip (SoC) design - like the nVidia Tegra X1 based Shield tablet already does.
Here is the good news The Tegra X2 uses the same GPU core architecture as the top tier nVidia graphics cards, however it has a lower clock speed and has far fewer GPU cores.
Talking of GPU cores, let's take a look at the PS4 & PS4 Pro (I'll also make mention of the Xbox One). The PS4Pro has the latest and greatest AMD GPU architecture (Polaris 10/11) - which is roughly equivalent to the Pascal architecture in X2. PS4 uses an earlier GCN design from AMD and has 18 Compute units with 64 cores a piece for 1,152 'cores' giving about 1.84 TFLOPs (@800MHz). The PS4Pro by contrast has twice that number - 36 Compute Units with 64 cores per unit, for 2304 'cores' giving 4.2TFLOPs (@911MHz).
Let's talk nVidia the Tegra X2 GPU has 256 cores, a full bore GTX1080 has 2560 cores, that's 10 times the number, and has a much higher base clock speed than the X2 can manage. In short, the X2 inside the Switch has less than a tenth of the power of a GTX1080 GPU.
The GTX1080 peaks at about 9 TFLOPS (@1.6GHz). So your Nintendo Switch with about 1/10 of the cores will hit about 900 GFLOPs, but we know it will clock lower, dropping performance closer to 750 GFLOPS - which is the actual projected rating of the Tegra X2 for Floating Point math. This is not PS4 Pro beating, and it's not really on par with the original PS4 which has 1.84 TFLOPS performance (Xbox One has about 1.3TFLOPs).
Incidentally, the data bus in the Tegra X2 is only 128 bits wide which is half that of the PS4/PS4Pro resulting in much lower memory bandwidth on the X2 vs the APUs in Xbox One, PS4 and PS4 Pro.
So, let's review.
Theoretical max GPU performance;
Switch 750 GFLOPS (0.75 TFLOP)
Xbox One 1.3 TFLOPS
PS4 1.84 TFLOPS
PS4PRO 4.2TFLOPS
Switch data bus = 128bit
XBox One/PS4/PS4PRO = 256 bit