GMA500 vs. G3 software render

[Adam Jackson]

On 5/3/12 7:36 AM, Adam Williamson wrote:
> On Thu, 2012-05-03 at 13:21 +0200, Adam Pribyl wrote:
>> While I consider the latest development on gma500, present in many Atom
>> base netbooks, a great success and would like to say thank you to
>> developer(s),

When you say "latest development", what exactly are you referring to?

>> G3 software render negates all this for gnome shell as this
>> is extremely slow on Atom. With the fallback mode the G3 is working very
>> well on gma500. But the obvious question is, beside the questionable
>> future of acceleration on gma500 - what is the future of the fallback? Is
>> there still a list of devices that should use fallback "by default"? If
>> yes, then I would nominate the gma500 for it, as this is really horrible
>> experience, making many users scared, just after live CD/USB boot.

At least some parts of upstream Gnome really want fallback to go away. 
They will probably get their way eventually.  You would be wise not to 
consider fallback mode a permanent thing.

> Well, the problem there is that the determining factor in how fast
> software rendering is is, obviously, not the graphics card - the whole
> point of software rendering is to *bypass* the graphics card - but the
> CPU.

It's slightly more complicated than that, but, sure.

> I don't know if there's a plan to blacklist systems with sufficiently
> 'bad' CPU performance from using software rendering, or if ajax is of
> the opinion that just about anything should be able to run Shell
> acceptably via software rendering and if a system doesn't, there's a bug
> that needs fixing...ajax, is there an overview here?

I don't currently have such a plan.  I could, I guess.

My personal opinion is that Core and up are pretty tolerable and 
anything less kind of isn't but also already kind of wasn't.  But I no 
longer really have any idea whether my threshold for acceptable 
interactivity matches that of anyone else.  Nor do I own any of the 
non-free Intel kit - on the grounds of not rewarding Intel's duplicity - 
so I have no real idea what it's like there either.

So if we want to blacklist low performers, okay, that's a thing we can 
do I suppose.  Where do we draw the line?

- ajax