Is Distributed Computing an option for Starlight?

Topaz Video AI Ideas
1

A Topaz rep told me the Topaz Labs cloud based render farm consists of 6 servers, each with 8 h100GPU’s (80GB VRAM each), with infinityband.

This rig can render Starlight at 0.25fps based on 10 seconds of video rendering in 20 minutes.

While I understand this might not be possible due to bandwidth and latency restrictions over the internet I’m wondering if a DC solution might be possible for Starlight rendering?

Could each client PC “work” on a single frame at a time? If this could work then clients could “earn” rendering credits based on how much rendering they have completed. Combined we have a lot of compute just sitting idle like 99% of the time.

2

If 80GB VRAM is the requirement to run Starlight, it won’t matter how many 32GB RTXs you have in a network.

3

Of course that is true. But generally VRAM reguirements are for optimum performance.

As I originally posted, just a thought to put out there.

1 Like
4

I apologize for this post. After thinking about this even if DC was possible with Starlight considering the monster faces and cartoonish looking results in general, this would still make the effort a monumental waste of compute/electricity.

At this point I don’t think the juice is worth the squeeze.

It’s a good research/development project for Topaz Labs I suppose.

5

Here you can see the Server graded Cloud Hardware :crazy_face:
Servers

1 Like
6

Wow!!! State-of-the-art setup!

1 Like
7

I guess that also explains the long wait times in the queue :rofl:

1 Like