bakaneko (nyan.delete@this.hyan.wan) on September 29, 2013 10:44 am wrote:
>
> You can be happy if your mobile GPU can paint
> 0.5-2 million triangles per second. The
> 30-90 million other triangles can't be used.

There's another axis entirely: quality.

You can be happy if your mobile GPU works. The quality of those GPU's (and their drivers) tend to be very very questionable. Remember all those complaints about Intel GPU drivers on PC's a few years ago? Or the constant comparisons of quality of the output from various vendors wrt anisotropic filtering etc?

The mobile GPU's are generally so bad that people don't even talk about the quality of output, because developers kvetch about their applications simply not even working. Never mind "real" OpenGL, even the ES3.0 support is just horrible.

We saw some of the horrors of mobile graphics thanks to Intel's incredibly broken choice of the last generation Atom integrated graphics. It never worked anywhere (just google the windows complaints). But outside of that known debacle, I don't think people realize that it wasn't that Intel picked a particular bad GPU - it's simply that mobile GPU's haven't cared about quality the way the PC GPU's were forced to by bad press.

And by "haven't cared about quality" I don't mean the fancy anisotropic texture filtering kind of "it looks better". I mean the "it works at all" kind of quality.

Here's a blog-posting one of the Linux GPU people pointed at from some developers that have to deal with more complex GPU use across different architectures and operating systems:

https://dolphin-emu.org/blog/2013/09/26/dolphin-emulator-and-opengl-drivers-hall-fameshame/

and I've tried to tell the Intel guys that they actually have an opportunity in trying to make quality not suck so much now that they use their own GPU. Imagine that? Intel being in the position of having better-working GPU drivers than most of the competition? Some heads must be exploding at just the thought.

The mobile GPU landscape is so new and limited that quality just hasn't been an issue (think PC GPU's ten-fifteen years ago), but I suspect that's starting to change. I'm surprised that nvidia hasn't been trying to make a bigger deal out of this.

Linus