Hacker News Reader

Comment by esjeon | Hacker News Reader

> Then, run the frame analysis pipeline, which will divide the video into separate video scenes (1s each, or 1fps) > (…) > Frames analyzed 57,537

Aha, it makes total sense. This number sounds much more reasonable than “669 GB”, since the actual total size of processed frames would be like 10-30 GB.

(Not downplaying anything. Doing-at-home always requires some math on practicality)

> Total compute time 67h 40m 42s

I’m just curious tho — is there any paying options that can accelerate this kind of process? Just spin up GPU instances?

View on HN

> Aha, it makes total sense. This number sounds much more reasonable than “669 GB”, since the actual total size of processed frames would be like 10-30 GB.

The reason why is “669 GB” is the total raw footage size when I'm doing the video processing, I downscaled each frame to 720p to make the video processing much faster and I don't need full original quality in order to get accurate results (as far as I know and experiment with).

> I’m just curious tho — is there any paying options that can accelerate this kind of process? Just spin up GPU instances?

For now, I found that NVIDIA GPU for example RTX 3060 with 12GB Vram was much faster than my M1 Max. (still working on optimizing for speed and accuracy).

What PAYG providers do people here recommend? Most powerful machine at home is an M1 MBA (16GB), so I too am interested in short term options where I can still benefit from the privacy of local models

Runpod

Seconding runpod.

They were having availability issues with GPUs (of course) but especially their UI where you'd customise a template only to try to start a pod, the GPU be unavailable and the UI reset forcing you to make the changes all over again.

But they have fixed that since, now starting a pod is more from a live page where as GPU availability status changes it updates in realtime/if your deploy fails you just try again - your customised env vars etc are still there.

Plus they also addressed the GPU availability problem as something they're working to fix and it's understandable seeing as nobody can get their hands on GPUs atm.

Yep. Go to vast.ai, spin up a cheap GPU instance, add a bit of code to the project and let it run it finish in just a few hours for like ten bucks.

But it's not as fun as running local model right here on your computer on your own desk. It feels like magic.

Reply on news.ycombinator.com