I am curious to know if this is normal ? I am processing a photo using Wonder Beta and it started off saying that it will take 155 minutes to process, it has now beeen running for 51 minutes and the message says 140 minutes to go, it got down to about 130 but has since been steadily climbing again. I am going to let it run out of curiosity.
The computer is a Ryzen 7, 8 core processor with 32gb of Ram, (I attach a copy of the specs).
What I would like to know …. cost out … is what specifications I would need to enable this to happen in a reasonable time, or is it just a lengthy process that no amount of money is going to speed up ?
This will be a major impediment to the use of your program.
How normal this is depends on the resolution of the original image and how much you’re upscaling. More pixels to process leads to longer processing time.
The main limitation of your system is the Radeon RX 580. That gpu came out over 8 years ago and has limited support for AI workloads.
By comparison, the GeForce RTX 5060 Ti 16GB is normally over 3 times faster (usually even faster for AI workloads that use tensor cores) for a similar power draw, plus it has more vram for running larger models.
Yes, Alan and TPX are correct, that is quite a old GPU, and Wonder is a stable diffusion AI model which is very memory intensive. You can test, to optimize speeds to use a smaller version of your file, Use Wonder on it, then upscale it to your final size with a non-generative Upscaling models such as Standard or High Fidelity.
We will continue to optimize the Wonder model as well, to use more of the GPU. Even with a good GPU this model is one that is memory intensive and will not be as quick as non-generative models.
Thanks Ange,
I am wary of upgrading one component only to find it causes problems elsewhere in the system. Would a Nvidia 3060 be better ?
Cheers
Grant
