Currently building a new machine or considering Google Cloud Compute configurations with a GPU.
How does the GPU contribute to processing speed and efficiency in relation to just adding more CPU cores (16, 32, 64 cores) ?
My current PC has an NVIDIA card, yet when I'm processing in Manifold 9, Windows Task Manager only shows the windows process "Desktop Window Manager" as using the GPU, never Manifold:
In this case, the 19% GPU usage is attributed to the process "Desktop Window Manager".
For now, GPGPU makes a large contribution, in a restricted range of cases.
As I understand it, GPGPU is enabled for all SQL Tile* functions which do not depend on neighbouring pixels. Whereas functions which operate over a window (functions using a border) are currently CPU-only.
For those SQL functions which are already GPGPU enabled, they are enabled everywhere in SQL.
In both respects, this is very different from the situation in Manifold 8.
And in the further respect that Manifold 8 did not use multiple CPU cores, while Manifold 9 does, whenever that makes sense and with minimal effort.
Bang-for-buck for now is to invest in many CPU cores, supplemented by a modest recent CUDA GPU.
Thanks - concise and really helpful.