Vegas Pro 22 AI Fx SUPER slow to render

Sean-Lester wrote on 3/4/2025, 5:52 AM

With my system normal renders for 4k 59.94fps video are around 30 fps, at 1080p 29.97fps video I get around 80 fps. When I use an AI effect on 4k 29.97 video render drops to 1.75 fps. Task manger is showing the following usage: CPU 25%, memory 17%, GPU 50-65%.

Why is the render so slow with a lot of "power" not being used?

Here are my system specs, it is an HP Envy w/ Win11

Processor   13th Generation Intel® Core™ i7-13700H 1.70 GHz
Installed RAM   64.0 GB (63.7 GB usable)
Hard drive 2 TB PCIe® NVMe™ M.2 SSD
System type   64-bit operating system, x64-based processor
Graphics (integrated) Intel® Iris® Xᵉ Graphics (external) eGPU w/ NVIDIA RTX 4060 8GB memory
Ports 1 USB Type-A 10Gbps signaling rate (HP Sleep and Charge); 2 USB Type-A 10Gbps signaling rate; 2 Thunderbolt™ 4 with USB Type-C® 40Gbps signaling rate

Thanks

Comments

Dexcon wrote on 3/4/2025, 5:54 AM

What exactly is the AI FX?

Sean-Lester wrote on 3/4/2025, 5:57 AM

they are all slow like this, especially upscale

Gid wrote on 3/4/2025, 6:20 AM

@Sean-Lester Yep same here, different PC's but similar results. This is a 1920x1080 29.97 MP4 clip in a 3840x2160 29.97 project. Vegas AI Upscale - Sharp, render 3-4fps with Magix AVC, Magix HEVC & Voukoder.

Howard-Vigorita wrote on 3/4/2025, 11:07 AM

@Sean-Lester AI Upscales over 1.1x seem to always be increasingly gpu intensive as the scale goes up. Doing that to an fhd clip as an Event FX in a 4k project is a 2x upscale, even without any cropping. Only way I know to speed it up there is to use a faster gpu with a greater amount of faster vram. If you can tolerate bringing the scaling down a little, you could try setting it explicitly by putting the AI Upscale into Media FX (in Project Media) instead of Event FX... should go a little quicker if you set the scaling to 1.5x and should look better than no upscale at all. A trick I use regularly in 4k multicam projects which include fhd ptzs, is to set scaling there to 1.09x which has negligible delay.

If your workflow leads to multiple renders of the same clip, you could at least confine the delay to the beginning of your workflow by baking the upscale into a pre-rendered intermediate. Doing that up front as a separate step might even be faster overall by letting other gpu fx processing go faster in later steps. If you try an AI Upscale transcode without any other processing, you could also try clocking cpu versus gpu performance by selecting no gpu in video prefs... the selected gpu in video prefs is the gpu used for AI and selecting none will cause AI to use the cpu. If your cpu and ram are super fast, they might outperform a slow gpu for that processing step.

Sean-Lester wrote on 3/4/2025, 11:41 AM

@Sean-Lester AI Upscales over 1.1x seem to always be increasingly gpu intensive as the scale goes up. Doing that to an fhd clip as an Event FX in a 4k project is a 2x upscale, even without any cropping. Only way I know to speed it up there is to use a faster gpu with a greater amount of faster vram. If you can tolerate bringing the scaling down a little, you could try setting it explicitly by putting the AI Upscale into Media FX (in Project Media) instead of Event FX... should go a little quicker if you set the scaling to 1.5x and should look better than no upscale at all. A trick I use regularly in 4k multicam projects which include fhd ptzs, is to set scaling there to 1.09x which has negligible delay.

If your workflow leads to multiple renders of the same clip, you could at least confine the delay to the beginning of your workflow by baking the upscale into a pre-rendered intermediate. Doing that up front as a separate step might even be faster overall by letting other gpu fx processing go faster in later steps. If you try an AI Upscale transcode without any other processing, you could also try clocking cpu versus gpu performance by selecting no gpu in video prefs... the selected gpu in video prefs is the gpu used for AI and selecting none will cause AI to use the cpu. If your cpu and ram are super fast, they might outperform a slow gpu for that processing step.

Thank you, I will gives these things a try!

Sean-Lester wrote on 3/4/2025, 1:36 PM

@Sean-Lester AI Upscales over 1.1x seem to always be increasingly gpu intensive as the scale goes up. Doing that to an fhd clip as an Event FX in a 4k project is a 2x upscale, even without any cropping. Only way I know to speed it up there is to use a faster gpu with a greater amount of faster vram. If you can tolerate bringing the scaling down a little, you could try setting it explicitly by putting the AI Upscale into Media FX (in Project Media) instead of Event FX... should go a little quicker if you set the scaling to 1.5x and should look better than no upscale at all. A trick I use regularly in 4k multicam projects which include fhd ptzs, is to set scaling there to 1.09x which has negligible delay.

If your workflow leads to multiple renders of the same clip, you could at least confine the delay to the beginning of your workflow by baking the upscale into a pre-rendered intermediate. Doing that up front as a separate step might even be faster overall by letting other gpu fx processing go faster in later steps. If you try an AI Upscale transcode without any other processing, you could also try clocking cpu versus gpu performance by selecting no gpu in video prefs... the selected gpu in video prefs is the gpu used for AI and selecting none will cause AI to use the cpu. If your cpu and ram are super fast, they might outperform a slow gpu for that processing step.

Update, tried different configs and the results were worse... less than 1 fps

Not knowing anything about Deep Learning Module I guess I was hoping I was missing a setting that would allow it use more of my available resources