Just installed Pix4Dmatic and am processing a map with 405 images on my current laptop with a 13th Gen Intel(R) Core™ i7-1365U with Iris XE graphics. I shared the system requirements with my IT and he said he has some new laptops coming in with Intel® Core™ Ultra 7 165U vPro® Processor with Intel Arc graphics. He claims this is the most powerful I can get without adding a ton of cost. Is an Nvidia RTX laptop going to make a ton of difference or is it minor? I ran the same photos through WebODM on my M2 Macbook Air and it created the ortho and las in less than and hour. I’m on hour 3 with Pix and it’s only at 20%.
1 Like
While I’m far from an expert, I use the following hardware:
CPU: 13th Gen Intel(R) Core™ i9-13900HX, cpus=1, threads=32
RAM: 63.75 GB
GPU: NVIDIA Corporation NVIDIA GeForce RTX 4070 Laptop
The amount of processing time varies greatly depending on the processing settings used. I’ll attach an example of a dataset I processed using 419 photos, close to what you’re referencing. Bare in mind, I processed using a 1/1 Image scale for my Dense Point Cloud which does significantly increase processing time.
After you fully process the photos, create a fresh Quality report. On the last page of the Quality report there should be a list of the processing settings along with the amount of time taken to complete each step. Can you share a screenshot of that? It should look something like the example I’ve attached.
Thanks and have a great day!
Calibration took a LOT longer. I just used the default 1/1 for calibration and dense point cloud. When I demoed Matic, I was flying with a mini2, so 12MP camera vs the 20MP I have now with the M3E. What does reducing the scale on Cal or DPC to 1/2 do to the quality? Their documentation isn’t very clear, to me at least.
So, comparing our reports, we’re pretty identical on most settings except DPC scale I’m at 1/1 and you’re at 1/2, and I’ve got SkyFilter turned on. The biggest difference is the DPC scale. I noticed a significant processing time increase when I swapped to 1/1. However, my DPC processing time is roughly half of yours. So, I’d say having the dedicated graphics makes a significant difference. Bare in mind, your RAM is also a key player here.
I haven’t tried messing with the scale in calibration, so I can’t speak on it at this point. When I first started, I was running the same scale as you on the DPC, but I was running into issues when I imported the data from PIX4DMatic to PIX4DSurvey, my 3D models were translucent and very hard to read, measure, and work with. It didn’t happen all the time, but often enough it was causing issues and delays. When I swapped from 1/2 scale to 1/1 on my DPC, I haven’t run into that issue since. I’ll share an example of what I was seeing incase it helps you. All of these images are off of a dataset processed with DPC at 1/2 scale. You’ll notice if I tried to look at the whole site like an ortho it’s barely visible, especially when looking straight down. Once closer it looks fine. If I just need to do some quick line checks, volume measurements, etc, 1/2 is fine, but if I need a detailed look at the site, I have to go 1/1.
This is helpful. I spoke with IT and they aren’t in favor of getting a workstation type laptop with NVIDIA gpu, but they seem open to an RTX 3060 or simular eGPU. I’m looking at options.
1 Like
Is it due to a budget restraint or platform issue?
For reasons I am not privy to, we only get ThinkPad T14’s. They said they tried the P (workstation) series in the past and had too many problems. But it’s a mix of that and budgetary ultimately. I’m trying to convince them not to waste the money spent on the drones and software by going cheap with the computer. I’ll get them there. I think I’ll end up going the eGPU route which is fine as I don’t need all the processing power when I’m out and about.
Hello,
This support page might be interesting for you: the processing steps that have High GPU usage will highly benefit from a NVIDIA GPU card. For the order of magnitude, we saw that on the same hardware the calibration is ~4x faster and densification is ~2x faster with Nvidia GPU.
For the orthomosaic generation, this page demonstrates a significant (4-5x) time reduction when a GPU is used, both on Windows and macOS. We also saw that thanks to new Apple silicon CPUs, MacBooks show great results on the processing steps that require High CPU usage.
It is general observations, but hopefully they give an idea.
Kind regards,
Alexey
2 Likes
