GPU full 4k and Speed Warp RTX 2070 Super

So figured out some things with 4k timelines and getting failed export/GPU full errors on export.
I usually export in .mp4 h.265 using the Nvidia encoder.
Usually gives an export that's a bit faster and cleaner than the Native encoder.

BUT -
It kept failing trying to export 4k clips with Optical Flow Speed Warp.
(4k 29.97fps from multiple sources and from 10-100 Mbps)
Tried lot's of things. (removing grading nodes 1 by 1, removing video transitions, transcoding etc)
Only thing that worked was switching to h.264 and using the Native encoder.
And even with a RTX 2070 Super, it exports at .... 1 FPS
But - it works.
(1080p export is like 5 FPS - I sometimes have exported 1080p, then re-import and supersample 2x)


Anyone have 4k speed warp export results with an RTX 3080?
Been wanting one.... guess it's good they've been sold out for so long as I would for sure prefer the 3080ti with 12GB VRAM rumored to be coming out in a few months/

I have a RTX 2080TI and it can't even show one frame when I take a 60fps 4K clip trying to speed warp it to 25 fps. Trying to play or render it out just gives me a slowdown of the entire computer and then a GPU error.

Feels like it's not really a viable option. 1 fps would be wonderful, it can't even do that and I don't know why. Feels like a broken marketing thing instead of a tested and viable tool to be used.

I've successfully used SpeedWarp on UHD footage to turn 25p to 50p. It was a few weeks ago but the almost 4 minutes clip took around 2 hours if I remember correctly.
This was on a RTX 3070, so 8GB of VRAM.
It's definitely not fast but the result was very good and the time was manageable.
I wouldn't add other effects to it though. Either rendering them beforehand or after the SpeedWarp.

Even my humble Radeon 580 can do it, but it's really slooooow.

How can you all get better performance (1fps is better than me) when using cards that are less effective cards than the 2080ti?

Mine is stuck at 0fps and when trying to export, the "remaining time" just increases until the end of the universe.

cache?
i have 2x 2080ti in the machine i was working on
just finished a show shot ArriRAW open gate, finished @ DCi4kScope
node cached at the end of the tree
then set optical flow/speed warp and cached the color output
caches took a few min, didn't note the time, was not excessive tho
export to Dci4kScope DNxHR444 went at around 120Fps (about normal for this machine)