Feature Request: Support for slicing using GPU via OpenCLI or Cuda
Posted: Tue Jan 29, 2019 10:52 am
The hardware acceleration is a nice feature to display the rendered model but it would be even better to be able to speed up the slicing process for complex models. I believe that this feature would really make this product stand out even more than the other slicers. I do not begin to understand how the slicing takes place. In my small little world. I see the slicing being broken out into a multi-threaded GPU scheme.
In my mind, to make this work and be successful, I would break the model into multiple Z height points and pick a zero point for the layer to connect.
Ie Calculate the number of layers based on the height of the model and layer height.
Break it into sub layers at maybe 1mm (not sure what the magic sauce is here without trying it)
Just an example..
1mm interval sub split. so if the model is 150 mm talk it would have 150 sub splits. (including rafts and such)
Super Basic example to send to the gpu:
Z=1mm endpoint for that layer is X=0 Y=0 Z=1 - Calculate this backwards to 0,0,0
Z=2mm endpoint for that layer is X=0 Y=0 Z=2 - Calculate this backwards to 0,0,1
Z=2mm endpoint for that layer is X=0 Y=0 Z=3 - Calculate this backwards to 0,0,2
and so on....
Send all 150 jobs to the gpu's, at each interval render down to the previous layers.
By doing it in reverse you will always be in the right point if the layer is mapped contiguously.
Then as the calculations finish you just assemble the temp files.
I would think on a midrange computer with a decent'ish GPU, that it would cut the rendering times down to seconds for large complex models.
I searched for this in the forums but I didn't see this request previously for this so I am sorry if this is a duplicate.
-Casey
In my mind, to make this work and be successful, I would break the model into multiple Z height points and pick a zero point for the layer to connect.
Ie Calculate the number of layers based on the height of the model and layer height.
Break it into sub layers at maybe 1mm (not sure what the magic sauce is here without trying it)
Just an example..
1mm interval sub split. so if the model is 150 mm talk it would have 150 sub splits. (including rafts and such)
Super Basic example to send to the gpu:
Z=1mm endpoint for that layer is X=0 Y=0 Z=1 - Calculate this backwards to 0,0,0
Z=2mm endpoint for that layer is X=0 Y=0 Z=2 - Calculate this backwards to 0,0,1
Z=2mm endpoint for that layer is X=0 Y=0 Z=3 - Calculate this backwards to 0,0,2
and so on....
Send all 150 jobs to the gpu's, at each interval render down to the previous layers.
By doing it in reverse you will always be in the right point if the layer is mapped contiguously.
Then as the calculations finish you just assemble the temp files.
I would think on a midrange computer with a decent'ish GPU, that it would cut the rendering times down to seconds for large complex models.
I searched for this in the forums but I didn't see this request previously for this so I am sorry if this is a duplicate.
-Casey