Ok, I’ve outlined in my previous post that the Surface Studio Laptop is a lovely device, and a potential candidate for replacing my aging desktop. So the question is, how well does it cope with the kind of workloads I have?
To answer that question, I’ve done some timing of photogrammetry using Agisoft Metashape, and rendering using Blender. I’ll compare the top-spec Surface Laptop Studio against an M1 MacBook Pro, and my old desktop, specs are below. I’m using Metashape rather than, say, Reality Capture, because the mac I’ll be comparing against can’t run Reality Capture (as apple devices don’t use Nvidia).
|Surface Laptop Studio||14″ MacBook Pro||Desktop|
|Processor||11th gen i7-11370H @ 3.3Ghz||M1 Pro, 8-core CPU||4th gen i7-4790K @ 4Ghz|
|GPU||Nvidia A2000 (4GB) (+intel Iris XE)||M1 Pro 16-core GPU||Nvidia 2060 Super (8GB)|
I’m well aware this is an apples to oranges comparison, in testing laptops against a desktop.
Comparison – Agisoft Metashape
For this test, I used my standard photoset of the Styracosaurus model, it consists of 53 images. I used this for a brief comparison when I reviewed the MacBook, but this time I’ve been far more rigorous. Each device ran a batch process to align photos (high), create a mesh from the depth maps (high quality, medium face count), then texture at 8192px. Each computer was given the exact same parameters, and each had the main GPU enabled in the Metashape preferences, and the integrated GPU disabled (as recommended by Agisoft). Times for each stage, and the total were reported in the dialogue box in Metashape. All trials throughout this post were run three times on each machine then averaged.
Much of a muchness in the results – the Desktop is winning, that 2060 Super is simply way more powerful than what can fit in a laptop, but the Surface and the Mac hold their own (the Mac more so). It’s interesting that some stages ran consistently faster on some computers than others. Alignment flew on the mac, for instance, while the SLS had the edge in texturing for some reason.
I also tested on Battery life, and there’s two things of note here: Firstly, the SLS performance very obviously drops when you unplug it, while the Mac maintains that performance. This wasn’t a long-term test, so it’s not clear if one or the other would throttle more with more time. After running the test three times though, it’d been 15 minutes of pretty heavy duty processing. The second thing to note is that the SLS went from 98% to 82%, which is 16% battery life in about 15 minutes. Not good, though if I were to make excuses for the surface, it was only setup today, and it is the case that battery usage tends to settle after a day or so (when indexing is finished and so forth). But that is an excuse. The Mac barely moved, dropping from 85% to 80% over the three tests: significantly less drain.
Comparison – Blender (simple scene)
The first of my two Blender scenes was simple in terms of geometry, but a little heavy in materials – there’s a mirrored plane, a cube with subsurface scattering, and Suzanne (the monkey face) with transparency. A single image was rendered at 1920×1080, with 128 samples and denoising with OpenImageDenoiser turned on. This is what the rendered scene looks like:
In all cases, Rendering was done with Cycles, and GPUs were used. It’s worth noting that it’s only recently that Blender has added Metal support for Mac M1 GPU acceleration, and future releases will probably have more optimization. As with the previous test, renders were done 3 times, and then timings were averaged.
When plugged in, the SLS and Macbook are pretty comparable. When you unplug them though, the Mac retains its performance, while the SLS drops a lot. The Desktop was clearly faster here however the laptops were used. I didn’t record battery usage for this one, as it was only a few seconds for each render.
Comparison – Blender (heavy scene)
Finally, I wanted to test how well all the devices could handle a really big scene. I’ve been putting together a big model of a dinosaur trackway using Reality Capture. The model is constructed from a couple of thousand images, and the final mesh contains 39,000,000 triangles – a number determined by the available RAM in the work machine I used.
This was interesting for me – the Desktop GPU has 8GB, like the work computer used to build the model, but the computer only has 16GB. Meanwhile, the SLS has 32GB Ram, but only 4GB VRAM on the GPU. The Mac uses unified memory, so while it only has 16GB, a large portion of that is available to the GPU.
There was no special lighting in the scene, this was just rendering a large mesh. Again, this was done at 1920×1080, but only for a max of 10 samples, and with no denoising. This is the resulting render:
The results were… embarrassing… for the windows machines. It wasn’t even close.
The Desktop could not render the model – it simply ran out of memory trying to render it. It could load it into blender, for what that’s worth (though that took a very long time). But of the two laptops, it wasn’t even close – the smaller MacBook Pro was heading for nearly half the time the Surface Laptop Studio needed. That’s especially brutal given the MBP only has 16GB RAM – the same as the Desktop.
Let’s be honest, the MacBook Pro really shone in these tests – credit where it’s due. But at the same time, I think it’s interesting that two modern computers, (admittedly laptops), aren’t significantly beating a 7-year-old Desktop (albeit with a 2-year-old mid-range GPU). If we’d looked at changes in performance between a 1998 desktop and a 2005 laptop the jump in computing power would have been far, far, more significant.
I don’t enjoy using MacOS, and I do like touch and pen capabilities, but if either of those things changed, it would be obvious that an M1 mac is the way to go for a personal computer that can do what i need it to. I’d really like to see a SLS with a 12th Gen chip and see if the performance difference (and battery drain) is still so egregious.
As I said at the end of my initial thoughts post, I’m not sure if the SLS is the right computer for me, and these results really do push me towards sending the SLS back and waiting to see what else becomes available. Unfortunately, during the afternoon while collecting this data, my Desktop crashed twice and stopped reading USB drives temporarily, reinforcing that I do need a new computer, even if just a stopgap until something really revolutionary becomes available.
UPDATE: Gaming performance
I got around to this after writing the above. I tested Hitman 3 via the Dartmor Benchmark. Identical settings between both Desktop and SLS:
The results were not even remotely close, with the Desktop absolutely destroying the SLS in performance:
I don’t have much time for gamin these days, and will have even less in the future, but it’s still a decent part of what I use my home computer for. I’m not sure if this is partly down to the A2000, rather than 3050TI in my SLS model, but I can’t imagine that the 3050TI improves performance much.
I assumed the desktop would still outperform the SLS, but I thought it would be closer. Really disappointing.
Leave a Reply