I’m in the process of getting a new computer to do photo/video editing and photogrammetry.
Any recommendations on Windows (gaming, video editing, beefed up PC) vs Mac would be appreciated.
I’d also like to have an update on the WebODM M1 progress as we start the new year.
Thanks in advance.
1 Like
Welcome!
Same rules as ever apply:
Performance per dollar will almost always go to self-built PCs for new hardware. Things can get a bit less clear with used equipment (especially generation+ old server stuff).
Performance per watt is still quite strongly in the Apple M-Series camp. If you need to be mobile with your processing machine, I don’t see a better option for battery autonomy while working. Linux support is still quite limited and nascent, but advancing smoothly it seems.
As for the specific hardware, that’s up to a bit of interpretation. In general, all consumer/prosumer GPUs are massively inflated in pricing due (mostly) to being speculated upon for crypto. Factor that into your budget, but it seems pretty flat across the board so it seems it can’t really be avoided.
AMD vs NVIDIA? We’re using CUDA for acceleration in a few steps of our pipeline, so if that matters to you, it might be best to use NVIDIA.
AMD vs Intel? AMD still seems to be winning on aggregate (multi-thread) performance per watt and per dollar, though Intel still seems to have better single-thread performance in a lot of its lineup. Most of our pipeline is highly multi-threaded, so strong performance there helps. However, we do have some steps that are single-threaded and heavy, so keeping an eye on balancing single-thread performance with multi-thread performance can be helpful.
RAM? Realistically, as much as you can afford, and then add another DIMM or two. Having the headroom will make it much less mental load to plan your processing and collecting, and gives you room to grow for a while.
Think of it this way:
Weak CPU just takes longer.
Weak GPU just takes longer.
Too little RAM? High probability you can’t process the dataset or need to massively scale back quality.
If you don’t already use it, PCPartPicker is just a phenomenal resource for building consumer-grade hardware.
As for the M1, we’re still supporting it via Docker primarily for now. Native is something we’re looking at, but as far as I know progress towards that is still early.
5 Likes
WOW! Thanks for all the details.
1 Like
I have an older Intel based Mac. Am I better off using that instead?
Would be great if there was a native version of WebODM for Linux as there is for windows.
I don’t believe so, no. Both M1 and older Intel Macs are using the Docker version for the time being. Maybe if you wanted to run Windows on your Mac and then use the WebODM for Windows native client, but 🤷
Which Linux distro family? Debian, Arch, Slackware, Fedora, Suse, *BSDs? Which libc, GLIBC or MUSL? What package versions are available? What arch? 
Docker and SNAP versions really help abstract all that away. Otherwise, we have the native compile script for Ubuntu 20.04LTS.
Outside of that, each platform has highly different packages, versions, and other things that mean that realistically, it’d have to be tailored to each platform that isn’t basically an Ubuntu 20.04LTS derivative.
Trust me, I know. I’ve been trying to make it work native for Alpine Linux for a long while now (and Solus prior).
Do you have a particular distro?
I use Ubuntu a bit. But yes, the distributions would benefit to be more compatible.
1 Like
This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.