Performance
Faster slicing, Sidebar (Re)Slicing that reruns only the steps a change affects, and a Preview that remains smooth to navigate on huge prints.
Two things got attention. The first is how long a slice takes, and especially how long a reslice takes now that Sidebar (Re)Slicing runs one for every change made in the Preview. The second is the Preview on very large prints, where frame times of half a second made panning feel broken and loading froze the window for seconds.
Slicing
Slicing on Windows is 10 to 17 percent faster. Memory allocation now goes through tbbmalloc instead of the Windows heap, which matters because the pipeline runs layer-parallel across every core you have and the Windows heap does not scale well under that load. If Windows blocks tbbmalloc, for example with Arbitrary Code Guard turned on in Exploit Protection, preFlight falls back to the Windows heap and Help > System Info shows which allocator is in use. This does not apply on Windows ARM64.
A settings change only reruns the steps it affects. Changing a seam setting or turning ironing on no longer slices the mesh again, so many reslices finish in a fraction of the time a full slice takes, which is what Sidebar (Re)Slicing depends on.
A few specific cases got their own fixes:
| Case | Before | After |
|---|---|---|
| Projects using Interlocking Perimeters or an over-bridge speed | baseline | up to 2x faster |
| Ironing on narrow top surfaces (one test case) | 30 s | under 5 s |
| Large Baobab supports | baseline | much faster |
| Travel planning with Avoid crossing curled overhangs | baseline | much faster |
The Preview on Large Prints
The test print for this work has 31 million segments. Before, the Preview drew every segment that the sealed-bead pass could not prove hidden, and that pass decided once at load for the whole print. Cut the print with the layer slider and the cavities it crossed opened up, so the Preview drew nearly everything below the cut. Hide one feature type and the pass was switched off entirely. Either way, frame times went from tens of milliseconds to hundreds.
The Preview now decides visibility for the list actually on screen, from the camera actually in use. It draws only the toolpaths the camera can see, at any layer range and with any feature types hidden.
| 31 million segment print | Before | After |
|---|---|---|
| Frame time while panning, zooming and orbiting | 170 to 670 ms | 27 to 98 ms |
| One step of the layer slider | over 1 s | about 50 ms |
| Longest freeze while the Preview loads | 15 s | under 1 s |
Rendering
Full lighting with 2x supersampling now renders more than three times faster. The color passes still render at the full supersampled size. The ambient occlusion pass and the depth and normal buffers it reads, all soft and low-frequency, now run once at the output size instead of being supersampled too. On an Apple M4 driving a 4K display, a frame went from 99 ms to 29.