[Bf-cycles] Blender 2.66

Brecht Van Lommel brechtvanlommel at pandora.be
Tue Dec 11 07:54:27 CET 2012


Hi,

On Tue, Dec 11, 2012 at 3:25 AM, David Black <db4tech at yahoo.co.uk> wrote:
> Viewport Render Toggle Switch:
> After trying some tests with a very dense particle system, I would like to
> ask your thoughts about a toggle switch for Cycles Viewport rendering,
> letting a user choose if Viewport rendering displays 'View' settings (as it
> does presently) or 'Render' settings (like the final F12 render).
>
> Currently it seems the only way to preview (render) my dense particle system
> with many child particles (heavily reduced 'View' settings for faster
> Viewport navigation) is by doing a full F12 render.
>
> Would a Viewport render 'View' / 'Render' toggle also be helpful for Cycles
> strand rendering Patch tests?

I'd like to have this working but there's some dependency graph
related inside Blender which make this difficult. So it will work at
some point but I don't know when yet.

> Viewport Half Resolution Cycles Render:
> One Idea I had while Cycles was in very early development, having a Viewport
> Shading 'Fast Rendered' option, below the 'Rendered' option.
> The 'Fast Rendered' option displaying a half resolution view (with possibly
> reduced light sample bounces for extra speed?), this would be helpful for
> those using laptops without access to GPGPU, also while using very high
> resolution displays, there would be no need to make a very small Viewport
> preview 'render' window to position lights in real time.
>
> Have tried some Viewport render tests, halving the resolution is about twice
> the performance (which I guess is logical).

This could be useful yes, would be good to add at some point.

> Couple of Questions to Finish:
> When Motion Blur is not enabled, why does the code still affect render
> speeds?  Render speeds have slowed a little, even while using optimal tile
> sizes, compared to before when the code was present but commented out.
> Please understand this is not a complaint, really pleased MB is there, just
> very interested to learn.

I tried very hard to avoid this, but I didn't find a solution.
Basically what happens is that the compiler can't optimize certain
things any more when you add a function call between two pieces of
code, even if it's behind an if() test. Variables have to be moved out
of registers and loaded back again. The only solution would be to
compile an MB and non-MB kernel. But for many features this could be
done, not sure where it ends then.

> And finally... Does the Viewport render now display using progressive
> refine, instead of the old progressive method?  I notice Tile size also
> greatly influences Viewport render speed even though tiles are not visible
> during rendering.

They're the same thing, it's just that previously you didn't have
control over the tile size.

Brecht.


More information about the Bf-cycles mailing list