Re: [Flightgear-devel] Volumetric Clouds

[Stuart B. <stu...@gm...>]

Hi Fahim,

On Mon, Feb 9, 2026 at 4:34 PM Fahim Dalvi wrote:

> Hi Stuart,
> > On 8 Feb 2026, at 5:41 PM, Stuart Buchanan wrote:
> >
> > Fahim - what GPU do you have?  It would be useful to have an AMD card as
> another datapoint.
>
> I’m running Apple’s M4 Max, so its their custom CPU+GPU chip.
>

Even better :).   I'm delighted that it is working at all.

Out of interest, how does this compare to WS2.0 non-HDR framerates in
2020.3 or 2024.1.
One of our hopes in moving to the core profile, HDR and WS3.0, is that it
will provide better
like-for-like framerates, particularly for platforms that are not desktop
Nvidia/AMD GPUs.


> Baseline (next): 80+ fps
> International Standard Atmosphere, no clouds: 33 fps (up from 14 fps)
> Core high pressure region (some clouds): 44 fps (up from 28 fps)
> Thunderstorm (a lot of clouds): 12 fps (up from 4 fps)
>
> Its still a bit strange why no clouds is worse than some clouds, but
> probably not as critical to debug right away.
>

That is a very interesting result.

The raycasting is relatively expensive.  It basically goes along a ray step
by step from the eye point until it hits something.
Rather than use a fixed distance for each step, we use a Signed Distance
Field (SDF) so at each step we go forward the
maximum distance possible before we hit something.  That should mean that
the "no clouds" case is very fast indeed because
it should just do a single step before it finishes.  That it isn't points
to a bug in my code that should be easy to fix.

Great work Stuart (and thanks Fernando!)
>

It's been a fun project.  I've learned a huge amount about raycasting,
shaders, HDR etc.  One of the great things about
FlightGear is the broad opportunity to learn something new.   I know I
mention it regularly, but I would encourage
folks just to jump in and get their hands dirty

-Stuart