Community
Fusion Design, Validate & Document
Stuck on a workflow? Have a tricky question about a Fusion (formerly Fusion 360) feature? Share your project, tips and tricks, ask questions, and get advice from the community.
cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 

Fusion only using 1 core?

58 REPLIES 58
Reply
Message 1 of 59
MTregz
16364 Views, 58 Replies

Fusion only using 1 core?

Hi there,

There seems to have been substantial performance hit over the last update or 2.

When re-calculating a model, joining a component to another or anything really that require calculations the CPU seems to only be using 1 core at 100% while the other cores don't do anything.

Using Windows 10, as of this post fusion is completely upto date. I have reinstalled the program a couple of times but am still encountering this issue.

System is quite new:

CPU: Ryzen 2700

GPU: Quadro P620

 

Anything that could help would be greatly appreciated.

As far as we can tell this is a Windows 10 specific issue. Mac seems to be utilizing all 4 cores, same with our windows 7 machine.

58 REPLIES 58
Message 21 of 59

My request for a detailed article would include using common terms used in the industry and an in-depth technical article that takes a deep look under the hood.

 

The very basic conclusion Ben is requesting are already pretty firmly established.

 


EESignature

Message 22 of 59

@brianrepp: This might sound rough, but that's sound like a message directly from the product manager. After some years of software development I have to say that you must have defined if a specific function should scale over threads (CPUs, cores, hyper threads) or not and at one point during the tests you must have specified performance test and the test hardware. That's pretty basic development stuff... Should be just a question of collecting data. 

Message 23 of 59
Anonymous
in reply to: Beyondforce

Thanks! My system work on one core because of the power setting.

Message 24 of 59
wcndave
in reply to: TrippyLighting

Bit of an old post, but what came up when I searched.

 

I have what I think is a typical example for me.

 

I have a fairly simple solid, which I wanted to cut out a honeycomb shape into.

I created one hex, and then used patter tool to create a grid of about 400.

That was all fine, however when I tried to move the grid... Computer stuck on one core at 100% for about 10 minutes now.

 

Reason I searched, because this seemed very odd to me, is that I have 24core CPU, 64GB RAM, NVMe drive that can do 5GB per second.

 

For a consumer PC it doesn't get much better, yet moving a set of sketches stops Fusion for going on 12 minutes now...

 

Is there any development in roadmap for multi-core support?

Message 25 of 59
TrippyLighting
in reply to: wcndave

The math and algorithms for the creation of geometry are part of the geometric modeling kernel. Many of those do not lend themselves to parallelization.

Unfortunately, math isn't as simple as that!

 

However, "moving sketches" sounds like a completely wrong workflow to me!

 

Maybe share your model and let's see what can be done to improve the situation. 400 patterned objects should definitely not take 10 minutes to update!


EESignature

Message 26 of 59

Splitting the task into many small task isn't an easy job for this kind of work. So if it comes to Fusion 360 (and lot of other CAD systems) higher clock speed (and higher instructions per second) are still king. Depending on the size of your model the RAM size shouldn't matter to much and the ssd only comes into play if you load things or on startup.

Moving sketches is mostly a bad idea in Fusion 360. Sketches do use much more compute power than 3d geometry. Is you pattern sketch or feature based? If it's feature based did you pattern bodies, features or faces or components? All matters if it comes to performance. 

Message 27 of 59

Ich weiss nicht...  Not sure I know what a feature pattern etc is...

 

I had a 3d modelled object, which is going to use a lot of print time.

So I created a honeycomb by making a polygon, then doing rectangular pattern.

However, it wasn't in the right position, some hexes cut through the edge, and the other side they were too far away.

I didn't want to move the model, because (for reasons I won't go into), not all the model was based on sketches, so if I move the model, some are offset from sketch, if I move the sketch for the model, some doesn't move....

 

In any case, it turns out that due to wall perimeters, a honeycombed model takes far longer, and much more material than a "solid" with infill.

 

However, I have been doing other things that were very slow.  Usually when there were many tens of thousands of faces from an imported mesh converted to model, before cleanup.

 

It's interesting that this work can be parallelised so easily, I work on things from video editing, to data science, to complex code, and it all uses multi-core.

 

One lives and learns.  Well.  Lives.  Mostly.

 

Thanks people!

Message 28 of 59
TrippyLighting
in reply to: wcndave


@wcndave wrote:

...  Usually when there were many tens of thousands of faces from an imported mesh converted to model, before cleanup...

 


A triangulated mesh does not require much data and is very simple. A mesh has a finite resolution.

However, in CAD software geometry is represented as either analytic geometry (simple prismatic shapes) or NURBS surfaces, or in the case of solid modelers such as Fusion 360 BRep containing both analytic geometry and NURBS data.

Those are mathematically precise descriptions of geometry and are free of resolution. That requires more data and is computationally much more involved. 

 

When you convert a mesh directly to a BRep, each facet of that mesh is converted into such a trimmed NURBS surface, with predictable penalties in performance!

Converting high polygon count meshes into Brep is generally not a good idea in any CAD software..


EESignature

Message 29 of 59
wcndave
in reply to: TrippyLighting

Right, I read that, however it's the only way to get STLs in for modification.  I try to clean them up as quickly as I can.

 

Just did the TESREG test, and had 5th fastest rendering time, with that engine, at 1:29, so it seems that's performing as it should, however noticed the render used all 24 cores 😉

 

Also 3rd fastest FPS, and fastest opening, so I guess all is well - I will just try to avoid too many complex elements being modified at once.

 

Thanks all!

Message 30 of 59
g-andresen
in reply to: wcndave

Hi,

the screencast shows in real time how 400 polygons are combined to a honeycomb plate and then cut for a frame.
With the file you have the possibility to follow the creation in the timeline and to reproduce the processes with your system.

polygon Combine & Cutpolygon Combine & Cut

 

günther

Message 31 of 59
TrippyLighting
in reply to: wcndave


@wcndave wrote:

Right, I read that, however it's the only way to get STLs in for modification.  I try to clean them up as quickly as I can.

 

Just did the TESREG test, and had 5th fastest rendering time, with that engine, at 1:29, so it seems that's performing as it should, however noticed the render used all 24 cores 😉

 

Also 3rd fastest FPS, and fastest opening, so I guess all is well - I will just try to avoid too many complex elements being modified at once.

 

Thanks all!


That's all pretty irrelevant!

The algorithms for physically based render engines are not part of any geometric modeling kernel.

Render engines work exclusively with tessellated geometry. Triangulated meshes in essence!

 

The question is whether you actually need a CAD software to modify .stl files. Many things can be done in Blender and Meshmixer.


EESignature

Message 32 of 59

I still don't understand the use of the TESREG test... 

Message 33 of 59
wcndave
in reply to: TrippyLighting

I don't really want to learn blender and meshmixer too, have enough with all the other software I need to use!

 

I also do want to learn it, and do design from scratch, it's just sometimes, I've noticed a file with many faces will be slow until I clean it up.

 

Thanks again!

Message 34 of 59

If you want to speed up your FFF you can

  • decrease you'r infill
  • try different infills, because some are easier to print than others
  • if you're worried about stability increase the number of perimeters slightly. Stability of prints is based on wall thickness and less about infill
  • if you're worried about your top surface increase number of top layers or increase infill below the top layer

Another approach is to switch to a wider nozzle and reduce the number of perimeters.

Message 35 of 59
g-andresen
in reply to: wcndave

Hi,

I've been wondering for some time what this is all about.

There is neither a file nor a simulated construct that explains the facts.

Even the simplest image representation is not provided.

 

günther

 

Message 36 of 59

@lichtzeichenanlage 

Sorry, I didn't mean slow to print, I meant slowing fusion down.

 

@g-andresen 

Sorry, do you mean you're wondering what I am talking about? and that I didn't provide a file etc.  Or that you wonder what the whole performance on one core vs multi-core is, and that no one has provided documentation on it?

 

Assuming you mean the latter, I can explain it in a bit more detail.

 

As someone who does 3d printing, I'd been using sketchup, as I was very familiar with it for woodworking.  However it has limitations and a lot of people recommended F360, and after an initial learning curve, I discovered some of the power of it, and am keen to learn more.

When I have an STL file, eg a router baseplate someone made, that needs modification, I may start from scratch, or, if it's a complex model, want to modify the existing model.

mesh router base plate.png

So I found online how to do this, turn of history, convert to BRep, then clean up.

And I get the warning about number of facets, and performance.  In this case 39,252 facets.

warning.png 

But if I want to use this model, I have to do this, and I was initially thinking it shouldn't be a problem, because I have a very fast computer... fast for everything I have done so far, 12/24 cores, 64GB RAM, NVMe Drive. (I do video editing, photoshop, coding, virtual machines, programming, sketchup, 100 things open at once etc).

So I continue with the conversion.  It took a few seconds, but then when orbiting furiously around the object, I can see it's using all my CPU cores, and quite a lot of GPU too. (nothing else is really running)

high facet navigation.png

But doing things that involve changing the model, could be quite slow.

So I reduced the number of faces, by just deleting one in the middle, seems to "auto clean", and also remembered to delete the original mesh, and things got faster again.

cleaned up.png

 

Then putting a honeycomb pattern sketch ready to cut extrude and moving it made everything hang for 10 minutes.

I saw only one CPU core was working.

I read this and other threads / posts, and have come to realise that some of the 3d modelling operations do not easily lend themselves to multi-core processing.

I did that RESREG thing as it was listed, and I guess the purpose is to check if what you experience, when using the same model as others, is as expected.  My numbers were very good, so both Fusion, and my PC are operating together "as expected".

I have a high CPU count, so rendering which is multi-core is super fast, whereas my CPU speed is only 4.2GHz, so some operations will be only "ok".

 

The video by Günther looks very interesting, and have watched it once, will need to watch again to understand exactly how that works.

 

In my case I wanted to create a honeycomb in an existing solid, to use less filament for printing.  However turns out it takes more due to walls vs infill.

 

I can't create cut-outs, as I need a contiguous flat/smooth surface.

 

It's been very useful to understand some of the performance factors/limitations involved, which will help me going forward.  Still not 100% sure why my sketch took so long to move, however if it happens again, I will record/repeat the steps.

 

Many thanks to everyone for input!

 

 

 

Message 37 of 59
joeEVLJ4
in reply to: MTregz

This issue has caused Fusion 360 to be almost unusable to me.   

I can't remember a time I didn't swear while using this program for so many other reasons.

If there isn't a solution to this issue I'll not be renewing my license. 

--Joe Urban
System: CPU: Intel(R) Core(TM) i9-7920X CPU @ 2.90GHz, 2904 Mhz, 12 Core(s), 24 Logical Processor(s).
MB: X299 XPOWER GAMING AC (MS-7A91)
GPU: GeForce GTX 1080
Installed Physical Memory (RAM) 16.0 GB
Total Virtual Memory 32.7 GB
Water cooled
Message 38 of 59

Can you upload the design that makes Fusion 360 slow? Sometimes it's about
the way things are used. Like many sketch lines in one sketch or patterning
in sketches or patterning bodies vs. faces.
Message 39 of 59

That is probably the issue.  But, for this quantity of lines on a sketch to overwhelm any adequate computer is an unacceptable performance for a CAD system.  Below is a screenshot of the sketch I was merely trying to move.  When all of these items are selected the system locks up - even before the move.  The 'browser' menu will also go black when I execute the selection.  I will also need to mirror these in the next step, so there will be 2X the elements.

joeEVLJ4_0-1612453160587.png

I did try to insert a reference file with a smaller set of this pattern, but that took forever to generate and also crashed the system.   

I have 25 years of experience working with AutoCAD, and executing these sort of operations on weaker computers was trivial.  This application feels like 3 steps backwards.  It's like a high school science project - not read for commercialization.  (Geez, even the 'preview' function for posting to this website doesn't work.)

 

--Joe Urban
System: CPU: Intel(R) Core(TM) i9-7920X CPU @ 2.90GHz, 2904 Mhz, 12 Core(s), 24 Logical Processor(s).
MB: X299 XPOWER GAMING AC (MS-7A91)
GPU: GeForce GTX 1080
Installed Physical Memory (RAM) 16.0 GB
Total Virtual Memory 32.7 GB
Water cooled
Message 40 of 59
TrippyLighting
in reply to: joeEVLJ4


@joeEVLJ4 wrote:

 

I have 25 years of experience working with AutoCAD, and executing these sort of operations on weaker computers was trivial. 


Fusion 360 or any other modern parametric CAD system isn't AutoCAD. It's time to update your skills to work with the sketch engines of modern parametric CAD systems. I've worked with CAD since college so for over 30 years and ditched AutoCAD for Solid Works in 1998. Patterning in a sketch has been considered a bad practice since I started working with 3D CAD systems,

 

While some CAD systems do this better than FUiosn 360, it still isn't a recommended practice.
You either learn this quickly or stick with AutoCAD, because otherwise, you'll just get more frustrated.


EESignature

Can't find what you're looking for? Ask the community or share your knowledge.

Post to forums  

Autodesk Design & Make Report