Topaz Video AI 4.2.1.0.b

I tested with Proteus and Nyx: the estimation works here. Be careful if you change the model, sometimes there is a download of the model (sometimes it’s an update?). It takes time (examining the model storage directory).

2024-03-11_14-48-33
2024-03-11_14-49-18
2024-03-11_14-49-50
2024-03-11_14-50-25

On the other hand, I did a test on a 30 minute TL. At the start I have 6 fps and GPU VRAM usage ~10/24 GB, at the end I only have ~3 fps and +23Gb GPU usage (no GPU FULL message). Slow increase in VRAM usage. It seems that there is not enough freeing up of GPU usage. (indications obtained from the Windows control panel).

1 Like

Mine did not change on my Mac .

I suggest if it is possible that during the download of a model, a popup type information for icon or? is displayed to inform the user to wait.

Curious how fast Aion can be with TensorRT acceleration

So apparently THEMIS deblur will be working on RyzenAI / 7000/8000 Mobile NPU, very interesting. Have I missed this feature for releases, or is it a sneak peak sort of deal?

That being said, AMD’s laptop ‘partners’ are very very bad with enabling RyzenAI. I have a Lenovo LOQ 15 7840HS + 4060 laptop, and despite my hardware absolutely having it, Lenovo has not enabled the ‘IPU Device’ despite users giving plenty of requests and feedback to AMD (and other manufacturers) over the past 6 months:

1 Like