Random thought to throw in the mix: obviously I don’t know the specifics of how the devices work on a code level, but since a device like Flow Restructure
is, if I recall correctly, analyzing each point as well as its relation to each point around it, is it possible that, in an extreme edge case like this, the sheer number of pixels being processed is leading to some sort of internal identifier getting an overflow error that wasn’t accounted for?
My admittedly simple thought process
Build resolution of 50000x30000 fails. 50000x30000 is 1.5 billion total pixels. If each pixel is being analyzed itself and against each the 8 pixels surrounding it, that’s 9 total operations per pixel, for a total of 13.5 billion operations (assuming that results aren’t shared across pixels). If each calculation has an internal ID for the results, the number of IDs would exceed the maximum value of an unsigned 32-bit integer. The half resolution did not build correctly (which may be a different, though possibly related, issue altogether), but did not have the fatal memory allocation failure. 25000x15000 would result in 375 million pixels, 3.375 billion operations, which would allow each operation to have a unique ID while still falling within the range of an unsigned 32-bit integer.
As a late edit: When running tests on my own machine, I’m able to successfully build a project structured as Advanced Perlin
→ Flow Restructure
→ Height Output
at 25000x15000 resolution. At 50000x30000 resolution, I do not get the fatal memory allocation failure error, but I do experience the build completing successfully according to the dialog box, but the Flow Restructure
device not building correctly/completely. The status box is also a color I’ve never encountered before, a very pale, almost white, green.
For reference, here are my PC specs
CPU: Intel Core i7-10700K
RAM: 32GB (16GBx2)
GPU: Radeon RX 5700XT
Page Disk: ~50GB available (none used in either build)
It’s interesting to me that I didn’t get the fatal memory allocation error, I wonder what the difference is there. I don’t believe it’s a memory issue, since @HYLK has more RAM than I do, so I wonder if it’s possibly CPU related?
Build Results for 50000x30000