Maya 2018 update 1 instancer slow(er)

[Anonymous]

Hello,

I noticed that in Maya 2018 update 1 the instancer seems to be a lot slower in viewport 2.0 than in for example Maya 2017 update 3.

My test scene existed of a locator instanced on 100.000 mash points.

I tested the setup on 2 different hardware configurations and both gave the same results.

It would be great if someone could have a look at this.

Thanks!

Hi @Anonymous your title 'Maya 1018 update 1 instancer slow(er)' was edited for clarity by @sean.heasley

[sean.heasley]

Hi @Anonymous and welcome to the community!

Can you please explain what's happening in your scene and/or how Maya is slowing down?

I followed your steps setting up the MASH node with the instancers and wasn't noticing any slow down.

Also if you can, it'd be very helpful to see a video of your machine slowing down so I can get a better idea of what the cause is!

If you need a video capturing program, Screencast is a great option that we use here at Autodesk

Please let me know if anything changes or if you find steps to reproduce your issue!

[mspeer · Accepted solution]

Hi!

I can confirm this.

MASH with Instances in Maya 2017 (upd4) is about 5 times faster than in Maya 2018 (upd1).

Not sure why it' slower in Maya 2018, maybe a bug maybe a bug fix or a new feature support for instances with MASH, anyway

please report this problem (bug?) to Autodesk.
Help -> Speak Back -> Report a Problem

[Anonymous · Accepted solution]

Hi Sean,

Thanks!

If you increase the amount of instances to a number where your viewport starts getting slow you might be able to compare the two.

I reported the problem as mspeer suggested.

Cheers!

[sean.heasley]

Hi @Anonymous

Thanks and yup that was it! I didn't notice a slowdown at 100 but I did when I increased it past that .

Thanks as well for submitting a report! I'm double checking as well to make sure this issue is logged with our dev team so a fix can be in the works for a future update!

[Anonymous]

Hey,

I noticed that this problem still persists in Maya 2018 Update 2. Is there any word on if this will be fixed in any update for maya 2018?

I made a test scene where I just use MASH to instance a (hidden) empty group with a grid distribute node. In maya 2017 update 4 it can comfortably instance 100.000 without slowing down the viewport. In maya 2018 update 1 and 2 it already starts struggling around 5.000 instances. We want to use MASH to scatter vegetation and rocks. Even if we just use some procedurals that have no display in the viewport it will still slow down maya (hence the test with the empty transform node)

I also tested using the instancer with nParticles (and again an empty transform) And it gave the same results: very slow performance in maya 2018 and much better performance in 2017. (200.000 particles: ~0.1 fps in maya 2018, 30 fps in maya 2017)

[sean.heasley]

Hi @Anonymous

I just checked and this is logged with our dev team but sadly did not make the cut for Update 2. It is still being worked on by the dev team but I can't say at this time what update/version of Maya will have a fix for this.

[Anonymous]

MASH recalculating every frame animation and if use Proxy as instances Render load Proxy every frame. 

for example scene frame render 40 sec than rerender this frame and get 5 sec render. render every next frame is 40 sec per frame. 

Need to add freeze function to stop recalculation for scatter

[smbell1979]

I would like to add that turning on the bounding boxes or single bounding box setting in the instancer node does nothing to speed up the viewport, it's as if the full calculations are happening regardless of the setting.  You can test this out by using MASH to instance a subdivided sphere a couple hundred times, and turn on bounding boxes. This should make the viewport lightning fast, but does nothing.

[NandoStille]

---

@smbell1979wrote:

I would like to add that turning on the bounding boxes or single bounding box setting in the instancer node does nothing to speed up the viewport, it's as if the full calculations are happening regardless of the setting.  You can test this out by using MASH to instance a subdivided sphere a couple hundred times, and turn on bounding boxes. This should make the viewport lightning fast, but does nothing.

---

True and utterly annoying. I reported this 2 days ago (BSPR-30784). Found this thread just now.

[mspeer]

Hi!

@smbell1979 and @NandoStille.

For such a simple object like a (default) sphere you will of course not see an improvement when switching to Bounding Boxes or Bounding Box.

(It doesn't matter if the object has 5000 faces or just 6)

Replace the sphere with an object that has more than 500 000 polygons and you will see a big difference.

[smbell1979]

That's exactly my point, very heavy geometry with bounding box turned on
makes ZERO difference in viewport performance. Absolutely none. It's as if
it's not even on at all.

[mspeer]

Hi!

In Maya 2017 it makes a big difference,

in Maya 2018 the difference is not as big because of the reported problem with Instances in Maya, but it's noticeable.

Create a MASH Network with 200 Instances of a Torus (subdivision 300) then animate a parameter (all visible in Viewport), what

FPS do you get (Geoemetry vs Bounding Boxes, vs Bounding Box)

[NandoStille]

This was - kindof - addressed in Maya 2018 Update 3.

The new "Display Percentage" attribute on the Instancer speeds up the display performance.

But I need to keep on nagging, because the way how the number of displayed instances is actually reduced, leaves room for improvement:

a sphere instanced on a particle grid with displayPercentage=10%

[laurence.cymet]

Hey Nando, 

The instancer percentage display work prioritizes more random distributions - like scattering on a mesh. Here's 5% of 3000 spheres for example:

In your more perfectly ordered case, in Maya 2018.3 there is a simple fix to get the display percentage feature to give you more random results. You can add an offset node, and apply a random point order like so. Hope that helps. 

We're looking into the performance reports folks - thanks for the heads up. 

Cheers,

Laurence 

[NandoStille]

Well of course it becomes random, when the particleIds are randomly distributed, position-wise. Adjusting the way the particleIds are distributed to get a random kind of display percentage is like putting the cart before the horse, don't you think?

[mspeer]

Hi!

For 2 reasons i find the current implementation good.

1. Performance

2. If you have a straight / pattern like distribution, it's better for me to see a small but complete part of the pattern instead of a random distribution.

This kind of Percentage reflects better (for me) the end result.

[laurence.cymet]

We aren't adjusting the point orders at all - which is why the results are linear on a grid, but random on a mesh scatter. It would be great to be smarter about the underlying distribution and do the right thing to always give you a random look, and we'll note this for the future. For now though we opted for keeping things clean and predictable and focusing on the more common use, and then providing tools to re-order points to get whatever effect you want - as there are so many uses out there for MASH. 

[NandoStille]

---

@laurence.cymetwrote:

We aren't adjusting the point orders at all - which is why the results are linear on a grid, but random on a mesh scatter. It would be great to be smarter about the underlying distribution and do the right thing to always give you a random look, and we'll note this for the future. For now though we opted for keeping things clean and predictable and focusing on the more common use, and then providing tools to re-order points to get whatever effect you want - as there are so many uses out there for MASH. 

---

There are no common uses.

The point order aligns to the function of the particle system, not the visibilities of the instancer.

There are setups where a sorted point order is intentional and serves a purpose.

@mspeerI doubt that controlling the instance visibility by a condition like "if percentage < random(id)" rather then "if percentage > id / count" would make a major difference in performance.

(MASH? We are talking about particles and the instancer. This doesn't necessarily have to do with MASH.)

[mspeer]

Hi!

@NandoStille

This thread was about MASH points ("a locator instanced on 100.000 mash points"), but you are right that it doesn't matter which tool uses the Instancer.

Every calculation needs time, if it slows down something depends on the code and how often it's evaluated.

This new feature helps nothing with the problem of this thread.

Still some data for 100.000 particles need to be calculated for at least 24 times per second for fluid playback, the visibility of instances has no impact on this. Some short tests showed a difference between Maya 2017 / Maya 2018 of 2/1, which looks like performance has been increased, but this new Percentage feature helps only when using high poly objects as instances and a low instance count (or no animation).