Some experiments in GPU-Land

My little not-a-DAE-replacement is coming along nicely.

I just did a test of batch processing 123 images using my GPU - boy was I in for a surprize!!! And this really shows why DAE needs GPU support…

When I ran this the first time I thought something had gone wrong (happens when writing programs) - it ran WAY too quick - 47 seconds but no - everything is perfect.

I then told my app not to use the GPU and ran the test again - a somewhat less impressive 8 mins + 15 odd secs.

Finally I ran DAE will do to the same image (1920x1324 image) so a 2k chunk size… 12 mins

Hmm - DAE should be close to my CPU-only timings, not 50% slower…

Full results…

GPU Time = 47.704 secs (0.388s secs per image / 9300 per hour)
CPU Time = 8 mins 13.031 secs (4 secs per image / 898 per hour)
DAE Time = 12 mins (5.9 secs per image / 610 per hour)

The 9,300 images an hour thruput mean that a 30 minute video @ 30fps should take under six hours to render (there would be some overhead I guess - I’ll find out when I get around to processing video)

Which GPU are you using at this point?

6Gb NVidia 3060 in a Gaming Laptop for that specific test run (got it last October)

I’ll have to go see if the Russian PC is at War with anyone as that’s a good test as well with a 1050-ish GPU.

A developer friend has a 2080 Laptop, I’ll have to get him to try it out as well.

I also have access to a remote 24Gb A10G (AWS Server that’s at least twice the speed) - Hmm - I’ll have to try making a Movie on that one… Not tried batch rendering on that one yet, only used it for training and 6k image renders (it can’t quite manage 8k - would need more GPU memory)

I did try a render on my 2013 MacBook Pro - less said about that one the better (29 secs for one 1920x2400 test) - I’ll try a little 8G M1 Mac Mini out for comparison / sanity checks in the next day or two (not expecting too much though - might have to waste 24 bux on a 24G M1 for a day which should be better).

Needs an invite to your alpha

As I said in other post