rafalemiec wrote:
laszlonagy wrote:
Only a few rendering engines support GPUs. Most of them still utilize the CPU cores.
Octane Render is an engine that is available in ArchiCAD and it fully utilizes GPUs, which is much better since GPUs nowadays offer potentially much greater performance for rendering with these hundreds (or thousands) of cores:
http://en.wikipedia.org/wiki/Octane_Render
Thanks, that's what I've been looking for. I also know some people who moved to Unreal Engine for high quality, photorealistic rendering. I'm going to try exporting my models to 3ds max format, which is readable by the UE and render from that, hpefully faster and more effective.
Graphisoft's FAQ on the topic of hardware is severely outdated, where it claims GPUs lack processing power to render environments with high polycounts and detailed lightning.
It's not so much processing power as much as memory capacity that they lack.
High Polycount and detailed lighting environments entail a great deal of Memory usage and the problem with GPU rendering is that the whole model has to be loaded into Video Memory to process the render, although you'll find most GPU and Video cards have a Video Memory ceiling of about 4GB.
That may seem like a lot but for a medium sized model (with a relatively moderate number of Morph objects and SEO operations), ArchiCAD eats up that much and then some just to load your libraries before taking into account how much of the Video memory itself gets used up to allow you to work in the 3D Window.
And then if you have any high resolution textures or texture maps, HDRI maps, displacement maps and objects, grass,....forget about it.
GPU's are great because they have so many more hundreds of cores to utilize which allows them to give you almost instantaneous real-time unbiased results.
But the memory limit means that you will always be limited to small models or relatively low detail models.
Some render engine developers are trying to modify their renderers to allow them to utilize a hybrid of both CPU and GPU together for the real time (RT) renders, which would theoretically mean you no longer face this memory issue if you have enough RAM installed in your machine ( I believe VRay might be looking into it, and the Maxwell guys gave a demo of their Maxwell FIRE RT renderer for Beta of the next version using GPU processing and possibly also using CPU at the same time. The current engine only uses CPU).
A GPU renderer that works inside ArchiCAD could on the one hand be theoretically great since it would potentially be processing the renders in real time on a window to your side using one of your GPU cards while leaving your CPU cores free for use in ArchiCAD which uses them for everything from Plan views to Elevation/Sections.
But then the 3D Window in ArchiCAD does use the GPU so there might potentially be a conflict or drag on resources.
I'm not sure how Octane renderers (the only GPU renderer that allows you to work from withing ArchiCAD handles this load distribution currently, but theoretically it would be ideal.
Cinerender uses CPU which means your computer can easily assign one core to the render to allow it run in the background while the rest are used by ArchiCAD to allow you to continue working on your project while a render is being processed at the same time.
At least that's how I believe it works.
The only drawback is that it's slower than it would be if it were GPU, or ven better, a GPU+CPU hybrid.
