AI Upscaling langsam nach GPU-Tausch

Follow
Avatar
AudialsOneHD

Ich hatte bislang eine AMD Radeon 5600 XT in meinem Rechner und die Upscale-Zeiten waren stark verbesserungswürdig. Da Nvidia allem Anschein schneller ist und Audials exklusiv für NVidia für einige Upscaler auch CUDA und TensorRT anbietet, habe ich ein "Upgrade" vorgenommen. Da mein Netzteil keine schnellere Grafikkarte zulässt und ich auch eine leise Karte haben wollte, fiel meine Wahl auf eine MSI Geforce RXT 4060 Gaming X, welche im Allgemeinen ca. 50% schneller sein sollte. 
Die Leistung entspricht aber nur ca, 60% dessen, was ich vorher erreicht hatte. 

 

Waren meine Erwartungen falsch oder hat Audials noch irgendein Problem, da es mal eine AMD im System gab? Hat irgendjemand eine Idee?

Leider habe ich meine alten Zeitmessungen weggeschmissen, da ich nach der Entscheidung des Upgrades nicht mit dieser Situation gerechnet hatte.

Die einzigen noch vorhandenen Werte sind die Prognosen für ca. 30 min 720p, 30fps für das Upscalen auf 1080p mit Antares. Diese sind von 3h00m auf 5h20m gestiegen.

0

Comments

4 comments

Sort by Date Votes
  • Avatar
    AudialsOneHD

    Am System kann es nicht liegen. Irgendetwas in Audials bremst. Siehe dazu die Benchmarks von Topaz Video AI mit der Radeon und der Geforce. Da ist die Geforce bis zu 4x schneller:

    Topaz Video AI  v5.3.3
    System Information
    OS: Windows v10.22
    CPU: AMD Ryzen 9 5900X 12-Core Processor              63.923 GB
    GPU: AMD Radeon RX 5600 XT  5.9605 GB
    Processing Settings
    device: 0 vram: 1 instances: 1
    Input Resolution: 1920x1080
    Benchmark Results
    Artemis        1X:     02.89 fps     2X:     ERR fps     4X:     00.31 fps     
    Iris        1X:     05.40 fps     2X:     03.23 fps     4X:     00.68 fps     
    Proteus        1X:     05.47 fps     2X:     02.27 fps     4X:     00.72 fps     
    Gaia        1X:     00.84 fps     2X:     00.36 fps     4X:     00.50 fps     
    Nyx        1X:     01.02 fps     2X:     01.18 fps     
    Nyx Fast        1X:     03.85 fps     
    Rhea        4X:     00.28 fps     

    ```

    CPU: AMD Ryzen 9 5900X 12-Core Processor              63.923 GB
    GPU: NVIDIA GeForce RTX 4060  7.7705 GB
    Processing Settings
    device: 0 vram: 1 instances: 1
    Input Resolution: 1920x1080
    Benchmark Results
    Artemis        1X:     10.91 fps     2X:     08.19 fps     4X:     02.33 fps     
    Iris        1X:     11.48 fps     2X:     06.61 fps     4X:     01.90 fps     
    Proteus        1X:     10.50 fps     2X:     07.84 fps     4X:     02.70 fps     
    Gaia        1X:     03.47 fps     2X:     02.38 fps     4X:     01.62 fps     
    Nyx        1X:     04.22 fps     2X:     03.70 fps     
    Nyx Fast        1X:     07.88 fps     
    Rhea        4X:     01.29 fps 

     

    0
    Comment actions Permalink
  • Avatar
    AudialsOneHD

     it's possible that your expectations were a bit high, considering the performance increase is around 50% rather than doubling.

    I do not have a performance increase at all. Audials is with the Geforce significant slower on my system. I tried standard driver and studio driver. Restarted  several times.

    My last hope is to deinstall and reinstall Audials 2025.

    0
    Comment actions Permalink
  • Avatar
    AudialsOneHD

    It seems so whether Audials updates internally the predicted time and adjust it after using the engine. The time changes only for the used upscaling (eg for 480p to 1080p but not for 480p to 720p).

    So the calculation of predicted time is at the beginning not very accurate.

    After upscaling a short anime video with 430 frames from 480p to 1080p the predicted time changed:

    Deneb: 1:55 to 1:05 min

    Polaris: 2:30 to 1:45 min

    Pollux: 8:00 to 4:20 min

    Sirius: 6:00 (no change)

    Antares: 2:20 to 1:10 min

    Rigel : 24 min to 70 min (!!!)

    So in most of the cases the predicted time dropped and will be certainly more accurate the next time.

    My conclusion: I cannot judge only by the predicted time. I have to run them.

     

    Btw - after testing a real film (from old digital camera) the time adjusted again and went up.

    0
    Comment actions Permalink
  • Avatar
    William090

    It sounds like you were expecting a performance boost with the MSI GeForce RTX 4060, but the results are falling short. The switch from AMD to NVIDIA should, in theory, yield better results given NVIDIA’s optimizations for CUDA and TensorRT, especially for upscaling tasks in Audials. However, performance can be impacted by several factors, especially if there are remnants of previous AMD configurations that conflict with NVIDIA’s drivers or settings. To ensure a clean slate, it might help to check that all AMD drivers and any registry entries are fully uninstalled before reinstalling the NVIDIA drivers.

    I’ve encountered similar issues with hardware on my travels, like during my trip to Kauai, Hawaii. I rely on GPUs for capturing and editing high-quality images and videos of the island’s vibrant birdlife, like the beautiful red-footed boobies and the native Nene goose(https://hawaiiislandkauai.com/). Any performance lag in editing or upscaling can make a noticeable difference, especially when working with detailed visuals. 

    If possible, try resetting Audials’ settings or even reinstalling it to see if that improves compatibility with the new GPU. Alternatively, consider using a benchmarking tool to measure the card’s performance outside of Audials; this can help determine if the slowdown is software-specific or due to a deeper system setting.

    0
    Comment actions Permalink

Please sign in to leave a comment.

Didn't find what you were looking for?

New post