JEMC said:
Why would Nintendo pay for the redesign of the X1 on a smaller node when they can get the same result, and cheaper, using a Tegra X2 runing at lower frequencies? And why would Nintendo pay for a custom designed chip, specially given that they went for an off-the-shelf product for the Switch, when they can get the same or very similar results using a Tegra X2 at full capacity (1.5 TFlops FP16 at 15W)? |
They could use the X2 for the slimmer Switch, but it doesn't look like it from the leaked Switch OS hack, there's something called a Tegra Mariko chip which has a model number T214 (Tegra X1 is T210), the Tegra X2 has a completely different model number.
1.5 TFLOPS at FP16 is only half that at actual FP32, which is what most devs use, so 750 GFLOPS at 15 watts is alright, but Nvidia can likely do better than that for 2020. 15 watts may be a bit too hot as well, the current Switch runs at 11 watts docked.
I don't think Nintendo necessarily wanted to go with an off the shelf part, it may have been a decision based on time/cost, but it's also had some headaches for them like hackers being able to easily exploit the X1. For future Switch iterations I believe they will use custom chips but chips that are based off of existing designs Nvidia has.
Last edited by Soundwave - on 10 November 2018