New multi-GPU in Nuke 13.2 - How speed up training
Overview, testing and showing how to speed up training by using all GPUs simultaneously
00:00 Intro
00:15 Start Training at multi-gpu
01:25 Disable gpus
02:10 Reducing batch-size for one gpu only
03:50 Start Training at one gpu only
04:25 Results
05:29 Understanding how multi-gpu works and needs
06:15 Split Batch Training
07:20 Other ways of multi-gpu
08:15 CopyCat options to play in multi-gpu
08:30 Whats is BatchSize and Epochs
09:55 Crop size and resolution slices
10:53 Articles to go deeper
11:35 Dataset of the tutorial
12:00 Custom and labels for train
Articles:
• PyTorch Lightning - William Falcon
https://developers.google.com/machine...
https://machinelearningmastery.com/di...
https://analyticsindiamag.com/acceler...
/ how-does-batch-size-impact-your-model-lear...
https://towardsdatascience.com/implem...
My PC used on this video, if you need start point for a build search:
AMD Ryzen Threadripper Pro 5975WX
Asus Pro WS WRX80E
128Gb ECC Server
7x GPU 3090 with Nvlink pairs
Nvme raid for disks and caches
Ubuntu 20