Trouble debugging network issue (In TOPs take 4ms? Scene changer and Optical Flow issues)

humbletang · March 22, 2023, 1:03am

I can’t tell whether these issues are all related, so I’m just creating one issue initially. If it turns out that there’s several things going on then I’d be happy to make a few separate threads to make it easier to assist in an organized fashion.

I have a show coming up (sat) and I’m running into a few issues

Although I’m using the scene changer it appears that effects other than the one currently being shown are using up resources (cpu,gpu time)
stranger still the main offender appears to be the IN top from the optical flow in the palette. Which seems like it shouldn’t really be using much cpu/gpu time at all.
there also seems to be a problem elsewhere in the network where a feedback TOP texture buffer seems to be auto clearing itself when it shouldn’t but only when it seems like the system is under heavy load (videos here 2023-03-21 17-51-02.mkv - Google Drive, 2023-03-21 17-50-24.mkv - Google Drive, 2023-03-21 17-51-49.mkv - Google Drive)

What I’m concerned about is that there’s something going wrong with the scene changer making it so that too many resources from undisplayed scenes are getting used. Each regular issue from the optical flow isn’t significant on it’s own, but with 5 scenes together it seems to cause enough strain on the system that the feedback buffer is getting cleared. That’s the way I see these things relating to each other, but I’m not super sure whether this is a mac os touch designer bug or a problem with something else.

I’m happy to share the network with the developer team if that will help. I’ll also attach a performance monitor dialog output here.

gist.github.com

https://gist.github.com/DevinBayly/9cf8880b74dd9d387a25e356c654a9a1

gistfile1.txt

                   frame  Start Frame
    0.060 ms   109 cook  Forcing nodes to cook at start of frame
    0.006 ms       misc   Check Execute DAT /project1/base1/s5/s4/egg/higher_res/electric_curl_single_growing_center_blur1/probe/inspekt/executeEveryFrame (DAT:execute)
    0.002 ms  10506 cook     /local/time/clock (CHOP:null)
    0.010 ms  10506 cook      /local/time/scrubber (CHOP:switch)
    0.004 ms  10506 cook       /local/time/independent (CHOP:switch)
    0.009 ms  10505 cook        /local/time/cframe (CHOP:speed)
    0.042 ms       script    /project1/base1/s5/s4/egg/higher_res/electric_curl_single_growing_center_blur1/probe/inspekt/executeEveryFrame /project1/base1/s5/s4/egg/higher_res/electric_curl_single_growing_center_blur1/probe/inspekt/executeEveryFrame (DAT:execute)
    0.006 ms       cook     Propagating changes to other nodes /project1/base1/s5/s4/egg/higher_res/electric_curl_single_growing_center_blur1/probe/inspekt/counter (CHOP:constant)
    0.007 ms       misc   Network Message Check /project1/skeleton_data/tcpip1 (DAT:tcpip)

This file has been truncated. show original

snaut · March 22, 2023, 9:41pm

Hey @humbletang,

when you get strange long cooks in places you wouldn’t suspect and they tend to jump around when bypassing the operator, it often means that your gpu is under heavy load.
I’d be interested to take a look at the network to see why you get extra cooking (you can pm me). There are a few operators that could cause this due to their functionality. For example Cache TOPs and Cache Select TOPs can cause constant cooking.

Cheers
Markus

humbletang · March 26, 2023, 3:18am

Yea Let me get you a copy of the network. should I send to derivative support?

snaut · March 27, 2023, 2:57pm

Hey @humbletang,

you can also just direct message me here on the forum.

Cheers
Markus