[QGIS-Developer] QGIS 3D and Graphics Card Requirements

Sun Oct 15 05:08:10 PDT 2023

Hi,

my 2 cents from a 3D-interested (not yet experienced) user:
1. doing GIS-stuff on a notebook, no matter how performant it is, for me
is a PITA.
Clumsy keyboards, overpriced hardware. GIS-work without a large or 2
screens is useless -> so whats the point of a mobile computer then?

2. Had a 10 year old mid-range gaming computer so far (i7 4770k, 16GGB
RAM, 2GB Graphics card).
For any usage in 2D this was still more than sufficient.
Native QGIS 3D crashed.
Sniffed a bit into Blender, which had much better performance, but the
workflow to use it efficiently with real world data is still unclear to me.

3. As my pockets are full, I am in the process to order a NVIDIA RTX
4090 and will be reporting back about the finding.

4. What I would wish for the 3D-future of QGIS: Somehow wire into the
Blender 3D capabilities.
Usecase: For planning processes of construction (Windmills, solar farms,
Buildings, bridges etc.):
(Semi-)realistic landscapes based on geodata and aerial images, but
pimped up with 3D-model bushes, trees etc based on semi-automatic
classification of aerial images and underlying geodata.

Seeing such processes in "reality" will bring much better understand for
planners, authorities, population etc. than the best crafted 2D maps
ever will.

Cheers,
Bernd

Am 15.10.23 um 13:08 schrieb Martin Dobias via QGIS-Developer:
> Hi Hannes
>
>
>     __
>
>     the two ridiculously expensive and high-end graphics cards mentioned
>     by Luke sound like complete overkill. Surely reading and preparing
>     the geodata for 3D display (from data providers, via storage, on the
>     CPU?) would be the bigger performance/usability issue of the 3D
>     views, right? I am sure that for reasonable reasons QGIS would choke
>     waaaay earlier than what those cards would be capable of. The
>     potential benefit of those cards for QGIS 3D definitely is not worth
>     paying *1200-2500€* extra.
>
> I have not had a chance to try QGIS 3D with such high-end cards yet, but
> you're right, users would be probably hitting performance issues
> elsewhere when working with large amounts of data (but this also depends
> on data types, formats, styling etc.)
>
>     Martin/Jean, is there any change that you could provide a more
>     detailed specification of the hardware and software requirements for
>     QGIS 3D? It comes up again and again and I've just been searching
>     again without success. I step up to add it to the QGIS documentation
>     in early November.
>
> Okay, great, let me try... there are very few firm requirements for QGIS
> 3D, so it's hard to answer some of the questions exactly.
>
>     Esri provides for example infos on the required OpenGL version and
>     extensions on
>     https://pro.arcgis.com/en/pro-app/latest/get-started/arcgis-pro-system-requirements.htm <https://pro.arcgis.com/en/pro-app/latest/get-started/arcgis-pro-system-requirements.htm>
>
>     E.g.: What is the minimum amount of GPU RAM necessary? How does it
>     scale with more complex 3D scenes and what happens if there is not
>     enough RAM?
>
> There's no strict minimum really, but I would say 1 GB GPU memory should
> be enough for many use cases. When dealing with large amounts of 3D data
> - e.g. point clouds or tiled scenes (3D tiles), it won't hurt to have
> more memory, although with default settings QGIS probably would not go
> over 2 GB GPU memory usage. All layers use tiling, so with increased
> distance of 3D data from camera, there's less detail displayed, which
> limits GPU memory usage and GPU processing power needed.
>
> Each map layer gets up 500 MB GPU memory, but only few that are memory
> intense (like point clouds or tiled scene) would normally use all of it.
> When a map layer gets to the limit, it will unload data (that are
> currently not being displayed) from GPU memory to get under the 500 MB
> limit. In the QGIS release 3.34 it will be possible to configure the
> per-layer limit (and QGIS 3D views will show a warning if the limit has
> been reached and none of the tiles can be unloaded).
>
>     Does it get slow (maybe with a message) or does it crash?
>
> Any slowness should be caused by displaying too many triangles/points in
> the 3D view (or possibly there could be something in QGIS code that runs
> on CPU that is slow and affecting 3D performance for example when the 3D
> scene is getting updated as the camera moves, scheduling loads of tiles).
>
> To be honest I don't really know what happens if there's not enough GPU
> memory - in theory there could be crashes, or the scene would be just
> missing some data - or maybe something else, like freeze. It is worth
> trying - it may depend on the operating system and the graphical drivers
> as well.
>
>     Which renderer does it use on which platforms (I guess always
>     OpenGL?). Which OpenGL version is needed?
>
> Yes it is always using OpenGL. Our 3D window requests OpenGL 4.3 core
> profile. Shaders of materials that we use only need OpenGL 3.2 or 3.3.
>
> I am not 100% sure how things are on Windows and macOS. Apparently,
> Windows by default only provides very old OpenGL 1.1, and users need to
> install dedicated graphics drivers to get more recent OpenGL. But Qt
> also comes with ANGLE library (that translates OpenGL calls to DirectX)
> which can be used as a fallback option, and there's also software
> rasterizer as a last resort. There's even QT_OPENGL env variable to pick
> an option explicitly [1]. I think QGIS 3D will not work with ANGLE
> though, because it uses desktop OpenGL, not OpenGL ES which ANGLE is
> meant to support.
>
> All this is related to Qt5 which has been very OpenGL-centric. In Qt6
> things will be different again, because Qt departed from being
> OpenGL-centric and embraced Direct3D, Metal and Vulkan, and introduced
> their own abstraction on top of those APIs called RHI. (They also
> stopped using ANGLE.) So in Qt6, Qt3D has two options - either use the
> new RHI renderer (default) or use the existing OpenGL renderer. We use
> the latter, because we would need to migrate our materials/shaders to be
> RHI compatible, and also stop using some features that RHI does not
> support, like geometry shaders. Given that QGIS does not support Qt6
> officially yet, this is not something to worry about too much for now,
> only something to be aware of for the future.
>
>     Which extensions are needed?
>
> I am not aware of any OpenGL extensions that we would need in QGIS 3D (I
> may be wrong though).
>
>     Could missing extensions get logged prominently to the user to avoid
>     the "my 3D canvas is white/blank" issue (I guess that is the reason)
>     that often comes up?
>
> Blank canvas may have multiple sources of problems - it could be that
> OpenGL failed to initialize for some reason, or that the graphics driver
> did not like something in our shaders or rendering pipeline or
> geometries, or it could be even as simple as us misplacing camera in the
> initial zoom and all the actual content is actually behind the camera.
>
> For any of the issues with white/blank canvas we would need to
> investigate them thoroughly - what kind of data and 3d symbology was
> used, what is the OS, graphics drivers etc - I am sure we can get to the
> bottom of that with enough effort. I would like to go through the
> existing tickets about 3D issues and try to address those that can be
> replicated.
>
>     Which features depend on which extensions and what happens if they
>     are not available?
>
> As I said, I am not aware of any extensions that we could be using, but
> I could be wrong...
>
> Hopefully this helps a bit...
>
> Regards
> Martin
>
> [1] https://doc.qt.io/qt-5/windows-requirements.html
> <https://doc.qt.io/qt-5/windows-requirements.html>
>
> _______________________________________________
> QGIS-Developer mailing list
> QGIS-Developer at lists.osgeo.org
> List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
> Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer