ZBrushCentral

Poor Performance Duel Xeon 56 cores, Windows 10

Zbrush 2019 1.2

System Info

Windows 10 Pro
Dell Precision R7910
Dual Processor Intel® Xeon® CPU E5-2697 v3 @ 2.60GHz, 2600 Mhz, 14 Core(s), 28 Logical Processor(s)
Nvidia Quadro P5000
128 GB 2400MHz DDR4 ECC

BIOS and Intel drivers fully updated. Even tried disabling WACOM drivers.

On models with higher than 40ish million point count and six subdivisions, sculpting starts to have lag and delay.
For example, dragging brush in a fast motion creates straight lines, not matching the pen path.

I tried this with both base models from the zbrush program itself and our own production assets.

The same test on one of our 2017 Mac Pros, yielded better results than this workstation, which is odd as the specs on this machine are much more powerful.

I tried disabling/enabling the following:

  • Multidraw
  • Quickanddirty
  • multithread io
  • lowering and raising the thread count
  • hyperthreading in bios

There were slight performance gains in reducing thread count to two.

CPU usage across all cores spike to near 100% during activity

Hello @natemythvfx

That’s not necessarily unusual. That’s a pretty dense mesh, well in to the range that might start causing performance or stability issues. While you can sculpt on meshes that dense with a slow steady stroke, it will probably be out of the range that many other program features can work comfortably with.

How much experience do you have working with meshes of that size? Is it your feeling they should be performing better based on previous experience?


[edit] :bulb:Also, check to see that Dynamic Subdivision has not accidentally been enabled.

Thanks for replying Spyndel.

Dynamic Subdivision is not turned on.

We do regularly work with models with that much geometry, 40million+. The iMac Pros we are running now have Kaby Lake i7 4core 4.2 GHZ processors run the same tests much better.

We are starting to suspect this may be a bottleneck issue.

Well, I can’t speak to what the differences in specs and architecture between the systems you’re comparing might account for. Obviously you’re well above spec, but you’re also working with fairly extreme polycounts. If you feel you’re not seeing the performance you should with that configuration, your best bet would be to contact Support. We’re not really equipped to address that kind of issue here.

No problem. I appreciate the help man.

Something you can try, if you haven’t already. Go to Preferences > Mem and make sure that Compact Mem is set to 4096 (max). Leave Doc Undo and Tool Undo at 4. Play around with the MaxPolyPerMesh slider until ZBrush tells you that it’s too much for your system. Then save your config (Ctrl+Shift+I).

Hi,
I’m curious about your projects which use models with this extremely high poly count.

For which output medium do you create this kind of models?

Thx in advance :wink:

Movie VFX

Thanks for the help everyone. We are chalking this up to a processor bottleneck.

Hi,
thank you for your answer.

I’m absolute not an expert in the Movie VFX area and maybe I’m totally wrong but I think single objects with 40 million polygons are really really really too much for any kind of animation.

  • May you could ‘brake’ the single object in different smaller parts
  • May you could use a low polygon object with a displacement map based on the high polygon object.

May could I see an example where this kind of high polygon objects are necessary?
Thank you in advance .

@HarryBee

I doubt he’s actually animating 40 million polygons, but rather using the extreme polycounts to capture fine surface detail for normal or displacement. This is fine is you have the hardware to do it, and the experience to get what you need from the mesh and get out. Most users should avoid casually carrying around a single subtool of that kind of density though, in favor of spreading out the polygon potential over multiple subtools and using features like HD Geometry.

1 Like