Standby! On it…
Happening now!
Whoa! This morning was using between 50% and 250% of the processors…now its hanging out between 500%-1500%. About 5-6 times faster for me than this mornings’ version.
This is a 3.5k clip, 119 frames, set to 1/2.
Previously the application was reporting about 13s/frame, now its reporting 2s/frame. it is longer than that, but, sooo much better!
I’ve got 160GB RAM, forgot to mention.
Similar here on the MacBook Pro. 64GB here. I was getting 15% on the processor, now it’s averaging 50%. Render time is doubled over what I tried with the previous version, and RAM usage seems on par with what you said in your OP.
Amazing!!
iti s 2sec per frame, not 2 frames per sec, so it shorter. If you open the file the cpu number is set at line 124. Could you try to just comment this line and use full stack if your memory allows it? It going to be threads that should make it even faster
Same here Andy, could you try to comment out line 124 and check again please
the one that says: cpus = int(cpus/2) - 2 so it will blow the box fully
In the create_slomo.py file?
yes, line 124, cpus = int(cpus/2) - 2
yes but watch your memory my box got just all hung on 1/4 speed. The guys at arXiv2020-RIFE wrote it the way it does 1/2 and then calls itself recursively wich makes it faste but bloats memory exponentially. We will need to amend it to return intermediate resut and fire threads again for quater frames, etc
Whoa! Watching the CPU meter is like throwing it into hyperdrive!
yes but memory could be an issue at the moment that was the reason I’ve limited it to something more safe. But you have the idea on where to set the sweet spot in the meantime
Understood. I’ve only been doing 1/2 speed and this maxed out at about 50% of RAM. Will try 1/4 speed next.
Yeah, my Mac just shit the bed trying to do 1/4 with the throttling commented out.
it is possible to optimize memory usage here in exchange oif a very slight speed penalty will try later
hahah…On 1/4 speed I’m now swapping over 90Gigs of RAM aaaandndnddd I’m out of system memory. Yeah, we might need to go back to where you had it.
The way here is to probably make it adaptive by throwing a humble number of threads first and watching memory consumption and adapting on the fly