Starlight 2.5 doesn't work. Ver 1.50 (Specs: 24GB RTX Quadro 6000, 48GB RAM, Xeon)

Hi,

After being sold on Starlight 2.5 by all the paid YouTube influencers, I bit the bullet and paid for the full Studio Membership, unfortunately it’s been 3 days of trial and error and problem solving and I’m yet to render 1 single video locally using Starlight 2.5.

Specs are 24GB RTX Quadro 6000, 48GB RAM, XeonW3235. Windows 11.

I’ve tried multiple Nvidia Drivers, Clean installs, Rolled back to version 1.4, fully uninstalled Topaz, adjusted the memory settings to 80% and even 50%, Max Processes set to 1. Changed the output to ProRes instead of H264, Still no luck. Other models work fine but i’m not interested in them as visually they make almost zero difference. It’s just SL2.5 that i want and that doesn’t work. Starlight 1 takes about 1hr30 to render a 25 minute MP4 DVD res video and the output video looks almost identical to the source.

Video is H264 MP4 DVD Rip.

99% of the time, the software just says “loading model” and will be like that for 1-2 hours before i quit.

On two occasions the model loaded after around 50 minutes, then after 2-3 hours, it would render 1-2 seconds. other times when the model does load after an hour, it will say there’s 1 minute left to completion but will stay like that for 45+minutes and the end result is actually it rendered zero frames.

I’ve done multiple full reinstalls, repair installs, tried different drivers, changed the GPU and CPU and still doesn’t work correctly. I installed 1.4 which still didn’t work. 1.3.1 wouldn’t allow me to render locally, said I needed a pro membership or legacy account??

1.5 has fully crashed and just closed on its own multiple times.

I’ve trimmed the video down to a 30 second MP4 file and still the same issue. Rendering it in the cloud takes seconds and the output is incredibly good but very expensive on the credits and limits you to 6 minutes.

I’ve heard of Topaz for many years, I’d never expect such buggy non-functioning software like this from them. Absolutely shocking!

Reading all the similar issues others are having, I know its the software so i have to ask, how on earth did this get released?

At this rate, I’ll be looking at a refund.

Here’s some logs if its any use.

logsForSupport.zip (1.2 MB)

2026-05-11-20-43-2-Main.tzlog (138.1 KB)

Precise 2.5 is designed for a CUDA Compute Capability of 8.9. Some people have reported that it will run on their 8.6 GPUs.

Your Quadro 6000 is a 7.5. It doesn’t meet the minimum requirements.

Indeed, it’s 2018 card.

When I conducted the Starlight 2.5 system check within Topaz Video, my system passed all the steps, including the GPU. I’ll just request a refund.

I don’t believe their system check validates CUDA Compute Capability - at least as of today. That might be why you were able to pass that since technically you do have enough VRAM.

That’s crazy if the biggest bottle neck which determins if your GPU can run Starlight 2.5 is not actually checked, it’s just the VRAM that gets checked. Eitherway, I feel duped.

Please write into support at help@topazlabs.com and the team can assist with that refund if you wish to go that route.

Yes, the in-app checker is looking at CUDA compute capability, but we are seeing some issues in 1.5.0 that the team are working on addressing in the next release that were not being caught.

Thanks, I’m going to try it with an Nvidia A2000 ADA GPU which has 8.9 cuda compute and see how that goes first.

Definitely do this.

As the results of SLP are simply stunning most of the time and thus are really worth the slow speed.

Can confirm. I’ve abandoned SeedVR as a first step in favor of SLP.

I’ve rended a few clips using the cloud service for SLP2.5 and the results are nuts, which is why I’m not giving up on it just yet

So i’ve changed to an A2000 ADA card and version 1.6.0 and rendering now starts but its running at 0.1fps. It has been running for 20 minutes and says there’s 30 minutes left. It a 5 second DVD clip upscaled to 4K. Is 30 minutes for a 5 second clip normal?

So it took 54 minutes to render a 5 second clip and the the results are awful.

Are Topaz purposefully nuking the local render capabilities to push people towards to expensive cloud option?

Can you send the logs and the sample file you ran to the support team for review into this situation? 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.

There is nothing different about the model or how it runs when using a local processing workflow or in the cloud. The model is the same.

hey, sorry, I’ve just started a new thread to make sure this didn’t missed within the traffic of this chain.

1.6.0 - SL2.5 - Very Slow Render + Corrupt Image - Topaz Video / Bugs and Issues - Topaz Community

The log is in this post.

It’s CAD Workstation card, focused on stability and full OpenGL support but not max. performance, the card is power reduced for continuous operation and to keep heat generation low, the card was designed with a TDP just below 75W so that a 12V high power connector is not necessary. Not ideal conditions for good FPS :smirking_face:

Spec RTX 2000 Ada RTX 4060
CUDA cores 2,816 3,072
VRAM 16 GB GDDR6 ECC 8 GB GDDR6
TDP ~70 W 115 W
Target market Workstation/CAD/AI Gaming
Gaming performance ~RTX3060–RTX3070 range Native gaming card