Topaz Video 1.4.0

The English fonts are pretty good! All other languages ​​will be treated as totems by MINI.

I will test this, thanks

Great catch, @kmf-songokuj5! This explains exactly why my RTX 4090 is crawling at around 0.6 fps with the Starlight Precise model. I was going crazy trying to figure out why my hardware wasn’t being fully pushed.

It makes total sense. The Neuroserver is basically treating our newer architectures (like my Ada Lovelace SM89 and your Blackwell) as an older Ampere (SM80) card. By forcing it to run on BF16 instead of utilizing native FP8, they are leaving a massive amount of Tensor Core performance on the table.

In generative AI models like this, dropping to 8-bit quantization (FP8) would practically double the throughput and cut the VRAM footprint down significantly, without any noticeable loss in visual fidelity.

Hopefully, the devs see your post and push an update to compile the Neuroserver natively for SM89/SM100+ with FP8/FP4 support. Right now, it feels like we are driving a Ferrari in first gear. Thanks for sharing the discovery!

12 Likes

LOL the thought of getting 1.0 FPS or above on this model is insane.

Good afternoon from the NFL Black Magic department.

3 Likes

Yes, it checks the CUDA compute level as well as the VRAM, OS, and other factors.

You should see it under the Starlight model when looking at the model variant. Precise 2.5 is available to all users with an active subscription to Topaz Video.

The team is aware of MKV issues with the model and in other settings. They have been working on getting MKV supported.

6 Likes

Precise requires 12GB of VRAM to operate, even with the Neuroserver.

2 Likes

Ok, that raises another question. Precise 2.5 requires 8.9, but there are people who report that they’re able to run it on Ampere GPUs that are 8.6?

Please pull the logs and send to the support team from you Mac for review. help@topazlabs.com

To gather logs, please select Help > Logging > Get Logs for Support and attach the zip file to your reply.

Here is a video to help with the steps of how to collect the logs.

**Update

The model was built for GPU cards with 8.9, and that is what is supported. Some users have reported they can run with 8.6 but that is not guaranteed as not all cards running 8.6 will be able to run it.

We will try and get that more clear later today or tomorrow on the docs page.

2 Likes

Any chance the frame ghosting issue might be fixable?

what is it?

Sure, but then you could only run the models on Nvidia Blackwell if they use FP4 and only on ADA and above for FP8. There are many people who are running Topaz Video on 20XX & 30XX series cards.

It is a CMake (a C language compiler) environment variable. If you don’t do programming you won’t have it.

Could this be due to interlaced input video?

No, I’m working with progressive video.

It has been mentioned to the research team and devs to see what can be done. We are working on replication across multiple sources to help understand what is going on here and causing it.

3 Likes

Pls release any beta for Precise 2.5 for Mac :heart: