Messages

1

General / Re: Agisoft Metashape 2.2.0 pre-release

« on: January 13, 2025, 10:47:57 AM »

If you run out of 32GB memory, then it is obvious the processing will take very long time. Your laptop is only powerfull as your cooling solution is and your CPU is probably throttling. What are you current CPU frequencies and what phase is Metashape calculating now?

2

General / Re: Alignment very slow on large project

« on: December 19, 2024, 11:31:47 PM »

Then the task which takes days has exponential dependence or at least non linear. You can try manually split the area on 2 or 4 regions and then join them in the end.
It seems to be the same behaviour like when copying many files in the same folder. When number of files are in thousands, everything is fast, but when it comes to houndreds of thousands of files, the copy process can almost stop, just several KB/s on any fast disk.

3

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 19, 2024, 11:22:30 PM »

I'm not sure that they might affect our run times that much, as it's not as if they're mining crypto in the background...

In my case those ~5% is caused by number of running  processes/threads/kernel threads:
In windows ~141 processes / ~1720 threads / 50500 OS handles according task manager.
In linux     ~62 tasks / 149 threads / 259 kernel threads according htop task manager.
Operating system needs to constantly loop through all threads, which are just microseconds, but in total sum it steals ~5% of performance on my machine.

4

General / Re: Alignment very slow on large project

« on: December 19, 2024, 11:06:01 PM »

And how does your project look like? It is drone photogrammetry of large area, or ground photogrammetry of some building, or...?

5

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 18, 2024, 11:52:41 AM »

Locking stuff down only tests Metashape compiled for Windows vs. Metashape compiled for Linux.  Not so much Windows vs. Linux.

Use the same frequencies is mandatory when it comes to benchmarking software quality like OS kernel algorithms, which is the most important in case linux vs. win and according to my findings and measurements it is under ~5%. The rest % differences which I noticed are not related to linux kernel and are caused simply by running many background processes/services not needed during Metashape processing and SSD behaviours in case of free/full disk space I have(therefore I precached all JPEG files).

I could grab heavy Linux distro and compare it with very light Win distro and the results could be suddenly in favor of Win.

In case of running many local workers, there could be advantage in linux when using XFS file system, but only in subtasks where data are written to disk, which is only a tiny % time fraction of the the whole processing. The rest of the time data are read from and write to RAM, where file system has no affect.

6

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 17, 2024, 12:27:17 PM »

I can also test XFS, for me it means only shrink system partition and create another partition, so easy task. I will let you know.

For debloating win, there are more open source utils, I used Sophia script https://github.com/farag2/Sophia-Script-for-Windows  in combination with "O&O ShutUp10.exe" utility.
Some video tutorials how to debloat you can find on https://www.youtube.com/@ChrisTitusTech  and https://www.youtube.com/@CyberCPU

If you want to create custom windows installation ISO with already removed services, preconfigured options, ... (everything is done from origianl windows ISO), then it is best to use Win Toolkit utility. You will also find tutorials on those YT channels.

For you it would be maybe option to install Linux as a secondary system side by side with win(if it is still somehow important for you). In that case win can time to time after updates disable quietly  multi boot option, which can be fixed.

EDIT: I tried XFS and the times were the same, except gen. point cloud, which was for some reason 2% slower.

7

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 16, 2024, 11:55:04 PM »

LTCS comes with less features. I chose it, because those features were not important for me.
In terms of performance I never compared them directly side by side. The important is disable services you don't need(updates, cortana, MS store, defender, superfetch, telemetry, ...). Then you will be closer to linux performance, but it's level is unreachable, because number of active services/processes/threads is way lower than on win... except special unofficial versions like "Phoenix LiteOS 10 LTSC (19044.1381) FBConan" 1.3GB ISO, which are designed to be minimal.
So do not expect significant diferences between untouched LTSC and full PRO version in terms of performance.
I am still on win10 LTSC. Win 11 LTSC is probably a little worse .

I am not expert on linux file systems, but what I searched, XFS is a bit better optimized for large file transfer and parallel read/write scenarios. I don't think Metashape could benefit much from XFS. Metashape is reading and writing in batches, everything happens mostly in memory.

Rocky linux vs other distros, it depends only on used kernel version and also on running services. I chose Arch linux, because it allows me to install everything from scratch, so I can end up with very minimal custom version.
Try to avoid large distributions and chose some lite editions.

8

General / Re: Alignment very slow on large project

« on: December 16, 2024, 05:38:18 PM »

Can you at least decrease keypoint limit to 40k or even 20-30k? Would it be sufficient for your project? Also tiepoints limit 6000...for my projects 2000-4000 was fully enough.

9

General / Re: Alignment very slow on large project

« on: December 16, 2024, 05:28:54 PM »

In network scenario it seem to be a lot of work to join small partly results to one chunk, especially on large project.
Try to spawn more local workers on each node, it should speed up single threaded tasks.

10

General / Re: Alignment very slow on large project

« on: December 16, 2024, 04:42:13 PM »

Do you see at least GPU utilization or some network traffic between nodes? Selecting pairs and matching points are almost pure GPU tasks.

11

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 16, 2024, 04:26:10 PM »

So here are my results:
OS:
Win10 LTSC(21H2, build 19044.1348), updates disabled, nvidia driver 566.14 DCH, NTFS file system
Arch linux, latest kernel 6.12.4, XFCE4 desktop env., proprietary nvidia driver 565.77-5, EXT4 file system
Project: 1472 18Mpix JPEG
Metashape: Standard 2.1.4 build 19370

I locked frequencies for CPU(Intel 11700F) to 4.4GHz and for GPU(RTX 2060 super) to 1700MHz.
I also precached JPEG files using command "copy /b *.jpg nul" on Win and "cat *.jpg >/dev/null" on Linux ...otherwise there were huge time differences in detecting points phase because JPEG files on win are stored on 3/4 full SSD drive(1 partition) and on linux they are stored on other SSD(on almost empty EXT4 partition).

win | linux times:
257s   | 290s   ...detecting points (without JPEG precaching, times were 762s | 300s)
363s   | 361s   ...selecting pairs
941s   | 936s   ...matching points
609s   | 551s   ...estimating camera locations
1039s | 928s   ...gen. depth maps
2000s | 1827s ...gen. point cloud
...when I have free time, I will compare mesh and texture phase.

I am not sure why detect. points on linux is slower(I checked 3 times, to be sure).
Pure GPU tasks times are almost the same. When it comes to CPU side, there is 9-10% speed up in favor of linux... 5% are caused by fact I am running apps on win(like MSI afterburner, HWiNFO, desktop widgets, cusotom firewall, throttle stop, FAN control, email checker), the rest 5% can be caused by OS kernel.

12

General / Re: Metashape 2 Benchmark - Windows 30-40% slower than Linux?!

« on: December 14, 2024, 11:49:04 PM »

I installed Arch linux and tomorrow I will try to bench on one of my project.

13

General / Re: GeForce RTX 4060 Ti 16 Go any good for Metashape Pro

« on: November 26, 2024, 01:15:06 PM »

Yeah, that card has good design, big 3 fan cooler, so no problem at all. https://www.techpowerup.com/review/asus-geforce-rtx-4070-ti-super-tuf/3.html
Your CPU is missing HyperThreading, which is giving +30% performance, so if you spawn several local workers to fully saturate GPU without seeing drops in utilization, then your CPU could be not fully able to keep GPU depth maps processing tempo and 4060TI would be better choice now. For the future definitely 4070ti super.

14

General / Re: GeForce RTX 4060 Ti 16 Go any good for Metashape Pro

« on: November 24, 2024, 11:49:26 AM »

The overall GPU performance in Metashape depends also on used CPU, how much worker threads are used in local network processing, how big is the dataset and how many Mpix photos are processed.
More performant CPU can better feed the GPU when it needs that.

RTX 4070 Ti has 12GB VRAM, so it would be better 4070 TI super variant with 16GB or stay with 4060 TI 16GB, which will be nice boost for processing in comparisson old 1060 super. 16GB VRAM is usefull for texture building task.

What is your current CPU?

15

General / Re: GeForce RTX 4060 Ti 16 Go any good for Metashape Pro

« on: November 23, 2024, 10:24:03 AM »

Hi, RTX 4060ti 16GB is good option or better Radeon RX 7800 XT in that price tag.