performance degrading with time.

[houou]

Hello, 

Since some time ago.. I started no notice that Tyflow sim performance would degrade with time once the scene is open for some time.
This is hard to demonstrate, I know, and I should do some mor tests, but this is a recurring thing I need to do lately:

Somewhat 1 frame takes 1 second to simulate, after several iterations adjustments... 1 frame takes 10 seconds... starting to feel slow, I reboot max, back to 1 second per frame on the exact same scene It was taking 10 seconds.
The Ram seems to not be the issue either, as it remains half full when it goes slow.

This is not just one scene, and usually happens everytime I'm working with Tyflow (pro) recently.

3DsMax 2023.3   
Threadripper 3990x
64GB ram.

Has anyone encountered a similar problem?
I'll try to record next time this happens...

[tyFlow]

Haven't heard of that happening, no.

[houou]

Haven't heard of that happening, no.

I got it on video is not as exagerated as I mentioned before, but is like x2~4 times slower.

First time 0 to 40 frames. (10~11 seconds)
Second time on a new file 0 to 40 frames ( 3 or 4 seconds )

It can get worst than this.

[tyFlow]

If you right click in the editor, there's a profiler you can open to measure the time of all tyFlow functions. Open it up during your tests and see where it says the slowdown is.

[houou]

Oh! didn't know that existed, that is cool.

I did a fast test but is inconclusive I get lots of mixed performances now... sometimes fast some times slower then fast again.
MTF: &cacheEventParticles  seems the one that changes the most as for now...

I will wait until I get a clear slow down and try again tomorrow.

---

hummm clicking away from max.. coming back and resimulating and boom it choked suddently.

I tried on a new file side by side the same exact range of simulation ( ups now I checked and I'm missing one frame, but... anyways... the simulation is very stable so is not that frame.), and was 25seconds vs 6seconds.
and yes the slowest with difference MTf: &cacheEventParticles.

[tyFlow]

I just remembered....

There's a known issue in the current build (fixed in the next build) that causes slowdowns when playing a simulation for the first time during caching. This only happens when playing a simulation frame-by-frame (ie, hitting play and waiting for it to cache)...it doesn't happen when simulating a chunk of frames (ex: snapping the timeslider to the end frame and waiting for the simulation to finish). This slowdown would indeed be happening under the caching event noted in the profiler....however, it doesn't seem to be what you were running into in your video...but it's possibly related.

Next build will be out soon....once it's out it would be a good time to test. I can't recall when this regression was introduced...you could experiment reverting back to the last beta build and see if you see the same performance degradation. If you don't then this bug is likely the cause. Grab the latest beta here: https://docs.tyflow.com/faq/beta

[tyFlow]

Actually...for low numbers of particles, the regression may affect performance even when not playing back flows frame-by-frame...so performance should be improved overall in the next build.

[houou]

Humm interesting, I'll test this ASAP. thank you.

I'm actually using some Tycache to give birth some particles in this flow. ( more than 100k ) but just 1 frame at start to be used as targets.
Particle count now is arround 10k at the momment and they go to find this targets.

I'll try to reproduce in a more controllable way in another file and see... If I can narrow down a bit what is going on. Thanks to the profiler is much easier to track down.

[houou]

Not yet too much testing, but seems this got better on the latest build.

BTW super cool build! congratulations and thank you!