Which Fast Graphics Card is best for SketchUp?

[denisroy]

@sketch3d_de

probably because SU is - as the vast majority of 3D modelers for modeling operations - single-threaded and therefore uses 1 kernel only, i.e. a quad-core CPU with 25% CPU utilization means one core running at 100%…

… SU itself relies mainly on the CPU and doesn’t need GeForce monster cards, they make sense for e.g. rendering plugs only.

Hi @sketch3d_de,

Unfortunately, I have to disagree with your statement, at least a far as my own setups and experiences are concerned and from which I draw most of my conclusions. I understand it might not reflect all possible uses cases (see Disclaimer at the bottom) and know that I do not wish this post to be anything else than informative!

I have run the following test…

Tasks performed

( Settings: Bold = Enabled setting, Italic = Disabled)

• Open Sketchup Pro 2016
• Open @TDahl 's Viking '75 Mars Lander Model from the 3DW (a beautiful and complex 40MB model)
• Rotate, pan and zoom in/out
  • “View → Edge Style → [Edges, Back edges, Profile, Depth cues, Extension]”
  • “View → Shadows”
• Rotate, pan and zoom in/out
  • "View → Edge Style → [Edges, Back edges, Profile, Depth cues, Extension]
  • “View → Shadows”
• Rotate, pan and zoom in/out
  • “View → Edge Style → [Edges, Back edges, Profile, Depth cues, Extension]”
  • “View → Shadows”
• 1 min rendering with Visualizer v1.3
• Push/pull random faces within components
  • "View → Edge Style → [Edges, Back edges, Profile, Depth cues, Extension]
  • “View → Shadows”
• Close the model

Hardware used

• Asus G750JS Laptop (the “1500$ laptop” I’m referring to in my post)

  • Intel Core i7 4700HQ Processor
  • 16GB DDR3L RAM
  • Intel 4600 On-board GPU
  • nVidia GTX870M 3GB GDDR5 VRAM GPU (nVidia Optimus lets you decide which GPU to use on a per-app basis)
  • Samsung Pro 256GB SSD (should have very little influence other than fast program/model load times)
  • Windows 10 Enterprise 2015 LTSB (a heavily trimmed down, strict-minimum branch of Enterprise edition)

• A minimal process list

  • I close all process that was not a Microsoft Windows process, a necessary driver or the Open Hardware Monitor used to record ressource usage

Notes

• Static SketchUp Settings for all tests:
  • Style used: “Default → Architectural Design Style”
  • “View → Face Style → Shaded with Textures” even if the model uses 100% opaque hex colors and no bitmap materials
  • “View → Hidden Geometry”
• I had the hardware monitor capture at a rate of 1 poll per second
• I waited +/- 30sec between each step to give time to both CPU/GPU to fall back to normal state
• I have run the test twice. First with the nVidna GPU and then on the Intel GPU.

Result plot for nVidia GTX870M GPU

Result plot for Intel 4600 GPU

Conclusions

• Using the nVidia GPU

  • We see SketchUp using mainly one CPU core (Core #2), as expected.
  • (1) and (2): The spike versus the spread in (2) comparison is probably due to poll rate frequency and settings being changed as no noticeable difference was seen during these two steps.
  • (3): With back edges turned on, rotate/pan/zoom resulted in some GPU usage and slightly higher CPU usage but never took the single CPU core to or above the 50% mark.
  • (4): With back edges turned off, rotate/pan/zoom resulted in higher GPU usage and average CPU usage.
  • (5): With all edges and shadows turned off, GPU and CPU usage are both average.
  • (7): Push/Pull never took the single CPU core to or above the 30% mark, and made little use of the GPU.
  • nVidia GPU seems useful as it seems to take a good load in viewport operations, even with shadows turned off.
  • The noticeable difference in higher CPU usage for the nVidia test is most likely due to the superior visual quality experienced during this test.

• Using the Intel GPU

  • We see SketchUp using mainly one CPU core (Core #3) but with some activity from the other cores as well, most probably representing the OS management of the Intel GPU load that ends up spreaded across multiple cores.
  • (3): With back edges turned on, rotate/pan/zoom never took the any CPU core to or above the 25% mark.
  • (4): With back edges turned off, rotate/pan/zoom never took the any CPU core to or above the 35% mark.
  • (5): With all edges and shadows turned off, rotate/pan/zoom spiked one CPU core slightly above the 35% mark.
  • (7): Push/Pull spiked one CPU core to the 40% mark.
  • Although I saw a lot of degradation in the graphic quality during all viewport tests (mediocre performances, no AA, shadows being automatically turned off during camera movements), CPU core utilization sat between 20% and 35%, going up to 40% only once during (7) which tells me that SketchUp does not make such an extensive use of the CPU during viewport operations even when a low-end on-board GPU is used.

• (6): Although some renderers do use GPU, Visualizer v1.3 made use of all 4 CPU cores during both tests but made no use of the nVidia GPU at all to render the still.

DISCLAIMER: No polygons were harmed during this test. This “backyard analysis” is not meant to be treated as “utter scientific, undeniable proof” of anything. I just wanted to show the data from which I draw my claims and opinions. If results were to be dissected and/or destroyed by SketchUp Sages, Trimble devs or any other higher-power entity, I will humbly crawl back in my cave and resume my SketchUp self-education.