Thanks Jeff, for this good overview. The conclusion is very clear.
Looking back at the communications we had a few weeks ago on performance, please allow me the following questions for clarification. Mainly about the Windows and PD settings, I am wondering whether these settings make any difference on the results of your comparison?

• Which driver do you use: Studio or Gaming, and would it make any difference?
• Did you use any of the windows (System, Display, Graphics) settings? They can be defined per app individually: 1) let Windows decide, 2) Power saving GPU: intel (in my case), 3) High performance, GPU is NVIDIA…., and 4) Don’t use optimisations for windowed games…..
• And the PD-settings for hardware acceleration, did you use either or both the Enable OpenCL technology, and/or Enable hardware decoding (I am assuming that the latter has been used😊)

Thanks

Quote Thanks Jeff, for this good overview. The conclusion is very clear.
Looking back at the communications we had a few weeks ago on performance, please allow me the following questions for clarification. Mainly about the Windows and PD settings, I am wondering whether these settings make any difference on the results of your comparison?

• Which driver do you use: Studio or Gaming, and would it make any difference?
• Did you use any of the windows (System, Display, Graphics) settings? They can be defined per app individually: 1) let Windows decide, 2) Power saving GPU: intel (in my case), 3) High performance, GPU is NVIDIA…., and 4) Don’t use optimisations for windowed games…..
• And the PD-settings for hardware acceleration, did you use either or both the Enable OpenCL technology, and/or Enable hardware decoding (I am assuming that the latter has been used😊)

Thanks

In my testing the user really has very little control of anything, CL's AI Style approach reminisces of shadowfile generation. In fact, AI Style transformed video didn't even match source video bitrate. You have no control that I found, simply accept what the black box does.

I don't use the Studio driver, but no, it will make no difference with PD and never has. I've seen nothing from Nvidia that the quality control that goes into studio drivers for certain applications even includes PowerDirector, they definitively state some other relevant applications for quality control (Adobe Premiere Pro, DaVinci Resolve...), but those apps utilize the GPU rather differently. This evaluation was initially done with 536.67 driver, and then repeated with 537.13 driver, no performance change as would be expected.

The Windows settings you refer to are laptop type settings, this evaluation was done on a desktop. But I doubt any mentioned settings would play a roll, when the iGPU is enabled, PD will use iGPU for everything except encoding when user can specify choice. Here, you have no user choice for encoding, so the AI Style process will simply use the CUDA cores of your Nvidia GPU to apply the style when selected in AI Styles. I also did not notice any performance difference with the Windows GPU hardware scheduling flag, it does however change component load levels which is rather typical with the setting.

The PD settings for hardware acceleration will make no difference, much like the shadowfile generation process. The AI Styles process uses the Nvidia CUDA cores only, it did not utilize the dedicated Nvidia hardware blocks of NVENC or NVDEC. They must know all the PD issues with GPU decoding/encoding, so rather than fix them and allow user control to utilize in shadowfiles or AI Styles process, they apparently elect internal control and use software (CPU). As such, the speeds I listed will very some user to user based on their CPU. However, this is only a minor part of the overall transformation time and hopefully someone's not dropping a RTX4070 into a very poor performing CPU box. The referenced numbers should still provide a valid GPU comparison when deciding GPU upgrade merits if they like using AI Styles.

Jeff

Thanks Jeff, very helpful.
Much appreciated!