LG "NUCLUN" SoC - Any info?

Much worse than I thought then

 

SoC vendors basically don't update their drivers in-between major OS versions. So whatever the phone comes out with right now is basically what you'll see within its foreseeable lifetime.

The other possibility is that the memory subsystem sucks and performance is suffering badly from that, but the fill-rate numbers seem to be ok so that's contradictory.

 

Besides Nebu's legitimate points, good Rogue/Android implementations I've seen are the A80T and the MT6595. Both Intel & LG's attempts look lacklustering.

 

Raw CPU bandwidth is about half of iPhone 6. Of course that doesn't tell us how GPU vs CPU requests are handled...

 

That's a problem of ARM's interconnects, basically a single cluster only has 6.4GB/s available to itself. I think I see that the GPU has two ports available to itself so that'd be 12.8GB/s and able to saturate the memory controller, but who knows what I'm missing. I'm really thinking it's the drivers.

 

The drivers would be a point if there wouldn't be a multitude of other platforms with Rogues with the same driver version and without the same lacklustering results (unless I'm missing something obvious here).

 

GFXBench's image quality results aren't even necessarily a measurement of the quality of the rendering occurring during any of the performance tests.

And if the difference between Odin's G6430 and Apple's says anything it's that characterizing the IQ of a graphics architecture from GFXBench's results isn't so meaningful, let alone trying to attribute it specifically to the inclusion of FP16 units in addition to its FP32 units (which obviously never made sense anyway.)