Cura 3.6 still slow since 3.5 Beta

[phantom 17]

Hi,

 

Since 3.5 beta i adressed the issue that it takes much longer to slice an object then it did previously before as in 3.4.1,

It doesnt seem to be resolved in 3.6 at all!

 

Same settings on 3.4.1 as on 3.6 i picked a model and timed both how long it takes to slice it.

3.4.1 clear winner at 1.57 minutes

3.6 ridiculously slow at 3.59 minutes 🐢🐢

 

discussed here and reply from Ghostkeeper as 3.5 beta was the first to be this slow:

 

• Posted September 25 in Introducing Ultimaker Cura 3.5 | Beta

  On 9/23/2018 at 11:54 PM, phantom said:

Hi Samrtavionics,

 

Saved it as you said how to save it, let me know if this is sufficient.

destroyerfinal 3-4-1.curaproject.3mf

destroyerfinal 3-5.curaproject.3mf

I did a performance test on that 3.5 file using Callgrind and got this result: http://dulek.net/work/3.5-beta-callgrind.out.3886

 

This seems to point to this change: https://github.com/Ultimaker/CuraEngine/pull/791

We'll continue discussion there.

 

 

Any thoughts on this?

 

Greetz Phantom

 

[AAGC 3]

We have 3 iMac and cura 3.6 is working, but very slow.

[burtoogle 516]

Hello @phantom, please attach the 3.6 project that is slow to slice. Thanks.

[phantom 17]

Hi @smartavionics,

 

Any model can be used, its not dependent on the model. 

The larger the model the more obvious the difference in slicing gets, it's the same as what I posted in the 3.5 beta topic.

When slicing anything, the filament usage as well as printing time are very close so it is not the settings (difference of around 15 minutes on a 19+hour print).

It basically already starts when opening any model, there is already an noticeable delay in projecting it on de buildplate, and before it starts slicing compared to previous versions before 3.6 and 3.5

[burtoogle 516]

OK, well unless you can give me a project file, I can't help. Sorry.

[phantom 17]

Hi @smartavionics,

 

I also got a response on the 3.5 beta topic and it seems to be an update in version 4 according to @ghostkeeper

 

If this is non related, here the 2 files i made:

3.4.1 1.23 minutes to slice 

3.6 took 2 minutes 

 

Thanks for the help !

3_6 version testfile.3mf

3_4_1 version testfile.3mf

[burtoogle 516]

Hi @phantom, thanks for the files. I found that setting Top Surface Skin Layers to 0 made a big difference. The slice time for my Cura (roughly based on the master branch) dropped from around 260 to 190 seconds. Turning off combing completely didn't make a huge difference. Setting Max Resolution to 0.05 (from 0.01) dropped the time again to 143. Hope this helps.

[burtoogle 516]

As the print takes 15+ hours to print, the slice time is pretty inconsequential. I know that everyone wants the slicer to be as quick as possible but, obviously, there's a trade off here whereby increasing the quality of the sliced gcode almost always increases the time to slice. Also obviously, the implementation could be buggy or just not that well written in which case it can be improved once the devs are aware of the issue. Producing good quality gcode in a time efficient fashion is a challenge!

[phantom 17]

Hi @smartavionics,

 

Thanks for checking! 

The top surface layer is equal in both version right, as is maximum resolution ? Changing these settings would make a difference as would changing a layer height to a higher value, it's obvious that the slicing time would drop considerably, but that is a choice that you know will take more time to process(and print)

So the question is what changed that the slicing times are this much higher then before 3.5.

 

Slicing on my computer is very consistent, i have a gaming laptop with i7 desktop cpu and high end gpu and all on ssd drive, so I am not used to long slicing times and staring at the progress bar.

I can do it several times on files bigger then this, and it will be consistent with a stopwatch to within 2-3 seconds.

 

 

So if this print is 15 hours, you can imagine what the difference in slicing time is between 3.4 and 3.6 when I slice a model that prints more then a day, I'm talking 3-4 times longer, and in minutes that's a huge increase, so there must be either a bug or a setting so different that wasn't there before.

 

Let's hope we can get the speed as it was previously (same settings offcours) and hopefully @ghostkeeper has a fix in 4.0

 

Guys I appreciate the help and looking in too it!

 

Greetz Phantom

 

 

 

[burtoogle 516]

@phantom, @ghostkeeper, I have investigated why using one or more top surface skin layers causes a big slowdown. The culprit is my contribution (https://github.com/Ultimaker/CuraEngine/pull/631).

 

What that PR does is fix the problem described in https://github.com/Ultimaker/Cura/issues/2656.

 

I can't see right now how the implementation can be made quicker so we either have to live with the slowdown when that feature is enabled or we remove the fix and allow prints to fail again.

[burtoogle 516]

Looking again at that earlier PR, it is possible to save some time and so now it takes around 220 seconds compared to 260 so I shall submit a new PR for the changed code.

[burtoogle 516]

But then I tested the same model with different settings (layer height, line width, etc.) and it actually took slightly longer slicing with the newer code so maybe it's better to stick with what we have at the moment?

[phantom 17]

hope your not asking me :-P