| Darc Requiem said: Based on Nvidia's own slides. The Tegra X2 is roughly 50% more powerful 768GLOPS (X2) versus 512GLOPS (Tegra X1). It's manufactured on a smaller process. So it should use less power than a Tegra X1. The Tegra X2 also has 2 to 3 times the memory bandwidth depending on whether it's using LDDR4 (50GB/s) vs GDDR5 (80GB/s). The Tegra X1 had 25GB/s LDDR4. You'd hope Nintendo would opt for GDDR5 but you never know with Nintendo. Edit: The 1.5 TFLOP number people are touting is inaccurate that's in FP16. It's half that number 768GLOPs in FP32. FP32 is what the Xbox One and PS4 GPUs specs measured with. |
that last bit is also something I am always a bit tripped up with, wondering why the 1.5 can't be hit. I've always understood it was an unreasonable ideal (as measurements by manufacturers always are frankly), but never grasped the why.
Edit: Nevermind, figured it out. Curious that they measured at that vs FP32 or even FP24 (though I suppose that has to do with FP24 being uncommon these days.) I which case, I would say a good bit more rests on the extent of customization Nvidia is doing than I initially thought. Hmmm.
