The rotating test is nice and simple for a quick impression of system performance (“does my current system and configuration run sketchup okay, or not?”). But running it within SKP makes it very prone to differences in user settings, including screen resolution, extension loadout, versions, etc. We’re not exactly sure what it tests (cpu, gpu…OS version?). It gave me some wildly different results on similar-spec machines, even with the same settings applied.
We do know SketchUp runs well in most situations (medium sized models) on almost every CPU; there’s no real lag. And we’re pretty sure that powerful GPUs dont make too much difference, above a mid-range spec. But large/complex models do seem to “hit the wall” and this is where people are chasing CPU upgrades to improve productivity… so the tests should be designed to push the upper limits.
Ideally I would like to see an standalone & automated performance benchmark where a series of operations are performed (like the Blender rendering test). It would also be possible to run the same test in a sketchup app that is set to all default settings, and then to run the same test model and processes within our program, including all its setup customistations. (this would help us to find out if extensions or other settings are impacting performance).
Ttest processes could include:
Importing and placing a large component. (something with a lot of sub-groups, tags, materials, text).
Making 100 copies of that component then exploding them all.
Rotating the model (similar to your test), and testing with a few different modes (textured, xray with profile)
Creating a surface from contours, simplifying contours.
Draping or stamping a component onto the surface.
Putting a Section Plane through everything and making a Section Cut.
Merging all Tags into Untagged.
Updating a large-ish file and then Rendering it in Layout (one viewport, then several at the same time).
Any outputs from a benchmark would ideally give results in each test, which shows us what part of the system is being tested (CPU, GPU, HDD, etc). Rotating a model with shadows would involve the GPU whereas making contours would be purely CPU, right? (NB I dont really understand how SKP works under the hood.)
I imagine that creating such a benchmark would take more than an afternoon of a programmer’s time, so a short-cut approach would be to clarify what types of operations SKP uses most for heavy-duty work, and then recommend the closest possible industry benchmark. In the past Trimble recommended 3dMark points as the baseline for SKP performance (I think that emphasised GPU compatibility at the low end?), and CPUMark has also been promoted a bit. Maybe this just needs to be clarified for 2020/2021 (3d mark…but which test?..what about Cinebench r23?). EG If we knew that 3dMark TimeSpy (a CPU-intensive, but not computationally difficult test) was a good indicator of SKP&LO performance, then we would have a wealth of public data to guide our upgrade paths.
Some of the M1 results do reveal that some benchmarks are horse doo-doo in relation to real-world applications… so we can’t rely entirely on Geekbench or Passmark (or Prime95)…but maybe some components of these are perfect for SKP? .
I’d be interested to see how Layout performs. Particularly zooming in and out (which has always been laggy) and refreshing updated model references with multiple viewports per page (like internal elevations sheets).
I’ve just ordered a 16gb M1 Mac mini as a transition machine. I’ll let you know the experience. Unfortunately the lead times on the 16gb version are currently at least 3 weeks (here in the UK)
This isn’t too conclusive, but here is my MacBook Pro on the left, one of the highest end models with 32 GB of RAM. On the right is the A12Z version of the Mac mini (that’s the same chip as in the iPad Pro) with 16 GB of RAM. The clock speeds are both 2.4 GHz.
Subjectively the A12Z Mac is slightly smoother, even with half the RAM and using an iPad processor. The production machines have a processor that is improved over the A12Z, and runs at 3.2 GHz. It should be smoother.
Here’s a test using the Mac Mini at at its minimalist with only 8gb of ram. This processor action took place with many other programs open while Quicktime recorded the screen video.
You can see in the Outliner how many items and the complexity of those items SU was updating when a material changed was called for.
On the mac mini these group updates took just under 5 seconds. On a 4 year old decently fast Windows machine this same change to just over 6 seconds. I get this is a relatively small drawing but to do all of this on this minimal of a machine ($750.00) with so many other programs\processes going on at the same time is amazing.
I’m sorry but until we see multiple real-world benchmarks showing this is reliably true with hard numbers, that is marketing BS.
During their big video presentation on November 10th, Apple claimed it is the “World’s fastest CPU core”. But later the same day, they said it is the “world’s fastest CPU core in low-power silicon”. The difference is pretty important! It means that it’s the fastest CPU core per watt, but it doesn’t mean more power-hungry CPU cores are not faster.
This is still very impressive, and critically important for laptops and other low-power computers. But it also means that big fat desktop-class CPUs will probably continue to have better single-core performance.
The real world benchmarks have been posted countless times the last two weeks. It’s not marketing only and certainly NOT BS. These aren’t theoretical anymore. They’re out there.
Please tell us exactly which processor beats the M1 in single core performance. Doesn’t matter how much energy it uses. Go ahead and post the model and it’s benchmark. Thanks!
This is hard to test without a more demanding file. I tried a fairly demanding file (31 million edges, 13 million faces in the SketchUp model, that appears in three viewports on the page), I set it not to auto render, and to manually render in Hybrid. I then made changes in the model, saved, right-click updated model reference, and then manually rendered all three viewports. For my MacBook Pro and the Mac mini it took about two seconds to update the model reference, and 5-6 seconds to render the three Hybrid viewports. The mini was only slightly quicker.
I am very tempted to go this route too, the M1 is looking pretty exciting. I’m also tempted to wait and pick up a used last generation non M1 MBP on the cheap when everyone starts upgrading. Gonna depend on how good the performance really is. eGPU support would be great too, I’m hoping for a boost to Twinmotion performance.
I guess most rendering software will go RTX. An older mac with eGPU would be a sound route on that regard. I’m not into mac’s for working but I think the iPad graphic design and 3d tools are very cool. I was thinking on a surface pro as a working tablet, but it’s a little underpowered. I would love a tablet that could handle SketchUp pro. I will keep my eyes open for what happens here in the next couple of months.
On my windows laptop with 16GB RAM, i7-6700HQ 2.60GHz, Nvidia GeForce GTX 960M and Windows 10: 72 Frames displayed in 2.8092 sec, Avg frame: 0.0385 sec, 25.9859 frames/sec
On my brand new Macbook Pro M1 with 8GB: 72 frames displayed in 2.4353 sec, Avg frame 0.0334 sec, 29.9756 frames/sec.
Both tests done with trail version of sketchup 2021.
So, the M1 is marginal faster - but i bet: with native ARM-Code for sketchup the performance will be better.
