FreeCAD speed tests

[PrzemoF]

It's just an idea for now, but I think we could try to gather database of execution time of "FreeCAD -t 0" + some hardware info and see if some gear is better than another. Thoughts?

[kkremitzki]

This would also be nice in terms of catching speed regressions.

[Kunda1]

+1
how do we go about do this ?

[freedman]

How about create a sticky (or use a current one) and place a few files in it, this way folks can download the same file, obviously with creater permissions. I guess then start a speed test thread and see how it goes.

I always use the Aeromotor for my at home testing, it is 8 megs. That takes around 14 seconds to load which seems like a resonable time to test different machines.

[PrzemoF]

Loading a large model seems a good idea, but I'm not sure how to solve caching problem - running the same thing again might be much faster.

My initial idea was to run full test suite - it gives execution time as well. We could do both.
No the question what we want to know about the system? I think it's enough to get CPU type, graphic card type & driver and amount of memory.

Add data stored in a git repo - we'll think about visualisation later.

[Kunda1]

This may be too ambitious but perhaps we could code a script that automates the tests we want to get data for? That way we know precisely how long it takes instead of relying on the user to

• The script would be either bundled with the test files (or pulled from the official FC repository online)
• The script then runs some operations similar to the Test WB.
• Asks if the user would be OK with reporting the results via the script to a remote server
• The remote server has a DB that captures the results.
• The data is then published to be compared.

None of the data will have identifying/personal info so privacy will be honored.

[PrzemoF]

Sounds very good. It could be part of current Test framework, so no need for external script. Hardware info can be obtained in python [1] - that should solve some problems with scripting on 3 supported systems.

The remote server and db would be great, but it might be hard to do. I mean it's easy to code, but you have to have server and someone willing to make sure it works and it's safe. I'd say for the first step we could have a topic here with attached result files. Once we know the info is worth something we could try to implement the remote server/database.

[1] https://www.thepythoncode.com/article/g ... n#CPU_info

[freedman]

To test the graphic card there could be a macro that rotates something (like the aeromotor) at 1 degree increments and then time a full 360 degree rotation. I would think a simple stop watch would do the job but then I time stuff a lot. I'm not a fan of web based stuff when it can be done locally.

[jmaustpc]

No the question what we want to know about the system? I think it's enough to get CPU type, graphic card type & driver and amount of memory.

I would also suggest operating system, ... prove that Linux rules ... hopefully!

The last time we did any speed testing we just got a very complex model then marked all for recompute and ran recompute.

[sgrogan]

see here: https://forum.freecadweb.org/viewtopic.php?f=10&t=48464