So, very little processor or GPU load- what's Fusion actually DOING?

[mavigogun]

I'm like 40 minutes into loading a design- it's got a lot going on, so not entirely unexpected.    Still, Fusion is only using a fraction of the CPU (~11-14%) and very, very little of the GPU (around 3%).   I've asked this before: what's going on?   It sure seems like Fusion is phoning-in it's effort.    Sometimes- say, when doing a Combine or Filleting a compound, irregular curve -Fusion will crank most everything up, the GPU will glow hot, and I at least know an effort is being made- but this.... this is just -ya, long- but also, dissatisfying- like a jumping in the Mach 5 and not being able to get it over a few hundred RPMs.

[mavigogun]

So, over an hour out now, here's what the Task Manager has to say:

[mavigogun]

Then it suddenly kicks into overdrive for a couple minutes-

Drops to around 30% CPU for 3-4 minutes more, and, hey-presto! it's loaded.

[mavigogun]

The question isn't a comment- I really want to understand what's happening with Fusion when a lengthy process is using only a portion of the available resources.   Just now, a Mesh Conversion has been taking forever- yet, again, CPU and memory aren't near tapped out.    What's going on?

[algor_neil]

How many cores does your CPU have making up that 100% ? How many cores is Fusion running on ?

[mavigogun]

My understanding is that the demands of CAD preclude the harnessing multi cores for most tasks, as many tasks are sequentially dependent, with the exception of rendering.    For Fusion, it's all about the clock speed, not the number of cores.    If a bottle neck, how to make it visible?   Must it be a bottle neck?    And if so, how does the process relate?    CAD models are amazing nets of mathematics- that is to say, if there really is a distinction, the product of a lot of hard number crunching and not just syntax.

[chrisplyler]

I don't know much about 'puters. Could it be that your RAM is overloaded, and necessary bits of the file are having to be shuffled in and out of the cache on the hard drive?

[mavigogun]

A good thought.    Take a look at the memory load in the screenshots- hardly taxed.    Could the bus architecture depend on multi core use to utilize all that memory?   I... doubt it.    Mayhaps some more knowledgeable than myself could recommend an analytic tool and what result to attend to?

[HughesTooling]

A couple of questions. Is the design native to Fusion or are you importing and is it a distributed design with external linked parts?

Mark

[mavigogun]

Native, self-contain design with no  external dependencies.    Converting a Mesh at the moment with same performance.

[HughesTooling]

Seems odd as you would not expect a recalculation just from opening. I made the attached file a while ago to demonstrate how Fusion is Not using multiple cores. If you open the file it should open in a couple of seconds but if you try Compute All it will take about 10 minutes to recalculate. If on your PC it takes 5 minutes for one component adding more will add 5 minutes for each. The time for one is pretty obvious, the fillets have to be done as a single thread as each new fillet modifies the body and in places surfaces are removed so you can't start calculating the next fillet as the edge might be removed by another fillet. I read that self contained components should take advantage of multi threading but this file seems to show that's not correct but that might be just the fillet command is not thread safe and can only be running as one instance.

Warning Only run a compute all if you've got 10 minutes to spare. If you run task manager you'll see only one core maxed out, it will occasionally change to another core but on a 8 thread CPU it'll show about 15% most of the time.

Mark

[TrippyLighting]

Performance problems as they pertain to CAD software - all CAD software, not just Fusion 360 - can be categorized into three main areas:

1. 3D viewport performance

2. Creation of Geometry

3. Rendering (if the software offers it)

3D Viewport performance is mostly dependent on what API the CAD software uses and how efficient it uses that API. The main graphics API's are OpenGL (Win 10, macOS and Linux) and DirectX (Windows only). Apple, along with their not so great OpenGL implementation has developed an API that supposedly is closer to the 3D hardware and is called Metal. I am not aware of any CAD or DCC software that uses the metal API.

On macOS Fusion 360 uses the OpenGL API and on Windows Fusion 360 and Inventor use the DirectX API.

Creation of Geometry in CAD software is handled by the Geometric Modeling Kernel. The two mostly used geometric modeling kernel in mainstream CAD are Parasolid and ACIS. Some CAD software such as Spaceclaim uses both.

Fusion 360 and Autodesk Inventor us the ASM (Autodesk Shape Manager). This is a fork of the early (V7 in November 2001) ACIS kernel. The official ACIS kernel is past V25.

Many of these kernes are quite old (15-20 years) and the reason they are still in use is that many years of costly development have gone into them. It helps to realize that when these were initially developed, multicore processors did not exist. Also, the math and algorithms involved in creating CAD accurate geometry are complex and do not always lend themselves to parallelization so they can actually run in parallel on multiple cores.

Thus it is often the case that Fusion 360 only uses a single core and hyper threading is very rarely used.

In general, however, Fusion 360 is pretty slow and compared to other CAD software uses a large amount on memory for some things.

Rendering is it's own area and modern  render engines are usually very well suited to run on multiple cores. Fusion 360 only uses the CPU, not the GPU (graphics card) and fully utilizes all available cores power when rendering locally.