Quote Does anyone have any knowledge of how Good the Radeon in particular the 6000 series works with Power director?

I have a GTX Titan X (Maxwell). Its served me pretty good. I can get good work done with it. Some of my edits bring the gpu utilization to 100%. Some use 11gb of vram. Im considering Radeon because the high amount of stream processors and vram, and i use nvenc but i think i could possibly make use of a top of the line radeon. Any thoughts?

Can you describe your typical timeline edit features, source and produce video details or maybe post an example?

The last paragraph in this pdf of a comparison between AMD and Nvidia encode quality maybe of interest, https://forum.cyberlink.com/forum/posts/list/97532.page#392352

This post may also provide some context, https://forum.cyberlink.com/forum/posts/list/118553.page#446186

Comparing CUDA cores (Nvidia) or Stream Processors (AMD), may not be a valid metric for PD21, depending on what editing features you are using. Basic source content timeline playback and encoding don't utilize them.

Jeff

Quote

Can you describe your typical timeline edit features, source and produce video details or maybe post an example?

The last paragraph in this pdf of a comparison between AMD and Nvidia encode quality maybe of interest, https://forum.cyberlink.com/forum/posts/list/97532.page#392352

This post may also provide some context, https://forum.cyberlink.com/forum/posts/list/118553.page#446186

Comparing CUDA cores (Nvidia) or Stream Processors (AMD), may not be a valid metric for PD21, depending on what editing features you are using. Basic source content timeline playback and encoding don't utilize them.

Jeff

I create a slideshow with a 4k prores motion background either from proxy to 422. Its runs at 60fps. I also have a another video or image as a png file or anther proresvideo file over it also at 60fps. THere are frequent cuts and transitions. Also sometimes there is some color grading in color director or a lot, With music and also other tracks for sound design.

Sometimes i have a 4k background as described above and I have 3 other videos that may be 1080p layered over it. They may be mp4 files.

the video gets rendered custom at 4k60fps 3840x2160. As an mp4 file h.264.

Quote I create a slideshow with a 4k prores motion background either from proxy to 422. Its runs at 60fps. I also have a another video or image as a png file or anther proresvideo file over it also at 60fps. THere are frequent cuts and transitions. Also sometimes there is some color grading in color director or a lot, With music and also other tracks for sound design.

Sometimes i have a 4k background as described above and I have 3 other videos that may be 1080p layered over it. They may be mp4 files.

the video gets rendered custom at 4k60fps 3840x2160. As an mp4 file h.264.

I know in a prior thread you said you can't share anything, so from this description, I see everything is pretty CPU focused. ProRes, nor 4:2:2 can be GPU hardware decoded for timeline playback.

Saying something is mp4 provides virtually no insight into challenges. Your 3 other mp4 layered videos could be utilizing GPU if a supported decode format, H.264, H.265. Have you tried looking at your 100% GPU load and which aspects of your timeline editing is creating it?

Additionally, color grading is a CPU intensive task in CD or PD, a GPU does nothing to help.

Your only real GPU centric tasks I see from this description is the final H.264 render, however, this hardware encode process benefit could be totally dominated by CPU doing the color grading so potentially little net benefit. Since this encoding is really the only GPU centric item, in my view, the independent SIP core not associated with CUDA or Stream Processors with Nvidia NVENC is much preferred to AMD VCE for quality and reliable encoding in PD.

Jeff

Quote

I know in a prior thread you said you can't share anything, so from this description, I see everything is pretty CPU focused. ProRes, nor 4:2:2 can be GPU hardware decoded for timeline playback.

Saying something is mp4 provides virtually no insight into challenges. Your 3 other mp4 layered videos could be utilizing GPU if a supported decode format, H.264, H.265. Have you tried looking at your 100% GPU load and which aspects of your timeline editing is creating it?

Additionally, color grading is a CPU intensive task in CD or PD, a GPU does nothing to help.

Your only real GPU centric tasks I see from this description is the final H.264 render, however, this hardware encode process benefit could be totally dominated by CPU doing the color grading so potentially little net benefit. Since this encoding is really the only GPU centric item, in my view, the independent SIP core not associated with CUDA or Stream Processors with Nvidia NVENC is much preferred to AMD VCE for quality and reliable encoding in PD.

Jeff

Thank you. So basically PD only uses gpu in exporting the video and mainly gpu specific effects and that doesn't fall on the stream processors? Is that right?

Quote Thank you. So basically PD only uses gpu in exporting the video and mainly gpu specific effects and that doesn't fall on the stream processors? Is that right?

Did you isolate what editing features are creating your 100% GPU load?

For Nvidia, PD decoding of source video can use NVDEC (Nvidia decoding) when supported, and NVENC (Nvidia encoding) when supported to produce a video, which you said you use in OP, both of which utilize special SIP core not associated with and totally independent of CUDA cores. The GPU support features are here: https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new What the GPU supports is very relevant. Say a user only has source files that are wmv and they only produce to wmv. A Nvidia GPU will do nothing as it cannot decode or encode wmv, in such a platform that would always be done by the CPU.

As you indicate, some PD Fx's are hardware accelerated, like Abstractionism, they use OpenCL technology for acceleration which does run on CUDA cores as well as a few other editing features, like for instance AI Style Plugins. Most timelines don't contain significant duration Fx or AI Style's, so the most user benefit typically comes from these special SIP cores and not CUDA. But, if your timeline uses extensive AI Style's like everything gets Van Gough style treatment, a good CUDA GPU will help. So a RTX 3090 (10496 CUDA cores) will outperform a GTX 3060 (3584 CUDA cores), provided nothing else limits throughput. However, these two GPU's have essentially the same NVDEC and NVENC SIP cores so they will decode and encode basic source video that don't have these unique features applied at nearly the same speed.

The same logic above is true for AMD, NVDEC is UVD (Unified Video Decoder), NVENC is VCE (Video Code Engine) and more recently UVD and VCE combined into VCN, (Video Core Next) in new AMD GPU's since like maybe 2018/2019. A likewise similar analogy for Intel GPU's with Quick Sync.

As both AMD, Intel, and Nvidia have different SIP cores, what video formats each one supports is unique as well. They are not generic accelerators that apply to everything, each has a specialized niche format capability that was built into the hardware.

Jeff

Quote

Did you isolate what editing features are creating your 100% GPU load?

For Nvidia, PD decoding of source video can use NVDEC (Nvidia decoding) when supported, and NVENC (Nvidia encoding) when supported to produce a video, which you said you use in OP, both of which utilize special SIP core not associated with and totally independent of CUDA cores. The GPU support features are here: https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new What the GPU supports is very relevant. Say a user only has source files that are wmv and they only produce to wmv. A Nvidia GPU will do nothing as it cannot decode or encode wmv, in such a platform that would always be done by the CPU.

As you indicate, some PD Fx's are hardware accelerated, like Abstractionism, they use OpenCL technology for acceleration which does run on CUDA cores as well as a few other editing features, like for instance AI Style Plugins. Most timelines don't contain significant duration Fx or AI Style's, so the most user benefit typically comes from these special SIP cores and not CUDA. But, if your timeline uses extensive AI Style's like everything gets Van Gough style treatment, a good CUDA GPU will help. So a RTX 3090 (10496 CUDA cores) will outperform a GTX 3060 (3584 CUDA cores), provided nothing else limits throughput. However, these two GPU's have essentially the same NVDEC and NVENC SIP cores so they will decode and encode basic source video that don't have these unique features applied at nearly the same speed.

The same logic above is true for AMD, NVDEC is UVD (Unified Video Decoder), NVENC is VCE (Video Code Engine) and more recently UVD and VCE combined into VCN, (Video Core Next) in new AMD GPU's since like maybe 2018/2019. A likewise similar analogy for Intel GPU's with Quick Sync.

As both AMD, Intel, and Nvidia have different SIP cores, what video formats each one supports is unique as well. They are not generic accelerators that apply to everything, each has a specialized niche format capability that was built into the hardware.

Jeff

i'll do an isolation. I did a project yesterday and the gpu load was close to 100% but the cpu load was low. Thats because I was doing it in 4k. It seems to me that PD stuggles with 4k work. The timeline was 8 minutes, and 40 minutes in it only reached 3 minutes of the render. I used the bcc grunge effect. On one layer and another video underneath it with cropping used on both videos. I stopped the same render and did it again in 1080p and it finished in about 10 mins or less.

Right now im exporting a podcast. i added in a 4k prores screensaver to a .wav audio file and im exporting in 4k. The GPU is being used (i selected nvenc) and its fluctuating between 20%-40% utilization. t appears that in 4k renders it uses the gpu more.

By the way i have a Ryzen 3950x and a Titan X Maxwell with 64 Gigs of Ram. Exporting to a HDD

Quote
i'll do an isolation. I did a project yesterday and the gpu load was close to 100% but the cpu load was low. Thats because I was doing it in 4k. It seems to me that PD stuggles with 4k work. The timeline was 8 minutes, and 40 minutes in it only reached 3 minutes of the render. I used the bcc grunge effect. On one layer and another video underneath it with cropping used on both videos. I stopped the same render and did it again in 1080p and it finished in about 10 mins or less.

Right now im exporting a podcast. i added in a 4k prores screensaver to a .wav audio file and im exporting in 4k. The GPU is being used (i selected nvenc) and its fluctuating between 20%-40% utilization. t appears that in 4k renders it uses the gpu more.

By the way i have a Ryzen 3950x and a Titan X Maxwell with 64 Gigs of Ram. Exporting to a HDD

Yes, Boris Effects another niche GPU use, so the GPU load will more than likely be from your grunge effect.

Jeff

Hey check it out.

Someone with a 6000 series graphics card did a render test and the radeon beat the Nvidia gpu in h.265 export within powerdirector.

RTX 3090 Vs RX 6900 XT H265 4K Rendering Test - YouTube

https://www.youtube.com/watch?v=EbyXeCoEPuY&lc=Ugy9fh7gKWFAXyPra054AaABAg.9qAhLZXTQO09qAiEqc_orA

Quote Hey check it out.

Someone with a 6000 series graphics card did a render test and the radeon beat the Nvidia gpu in h.265 export within powerdirector.

RTX 3090 Vs RX 6900 XT H265 4K Rendering Test - YouTube

https://www.youtube.com/watch?v=EbyXeCoEPuY&lc=Ugy9fh7gKWFAXyPra054AaABAg.9qAhLZXTQO09qAiEqc_orA

Such a poor comparison, means nothing to me. A few brief comments after scrolling through link.
1) No indication how hardware decoding was set, can make a big difference depending on source video specs
2) One writing to C: the other D:, can make a big difference depending on source file complexity and location and hardware behind C: and D:
3) Encode preview enabled during encoding which can be terrible to do during performance test. Recent PD releases don't even allow hardware encoding when preview is enabled.
4) One wrote a 3.5GB file the other 11.0GB, no comparison of output specs meeting profile specs
5) Profile was 80000Mbps video bitrate, total duration 6:32:40, end file should be somewhat larger than 3.14GB as audio encoding bitrate was not shown. At 11.0GB, RX6900XT did nothing as was specified, output file specifics need interrogation.
6) No comparison of quality
7) Custom GOP used and somewhat unique GOP vs more typical. Comparison with default profile would be advantageous.

This test is basically I did something, here it is, totally inconclusive for a good back to back comparison. I'd like to see a little more vigor in the test and results to see real performance difference to make a ~$1700 purchase choice.

I didn't put in a Radeon card to verify, it's too hard to tell from literature as PD uses custom Boris products, not mainstream, but your use of grunge effect may be only CUDA (Nvidia) based. I'm not sure, I'd have to throw in my Radeon RX 5700 to verify.

Jeff

Quote

Such a poor comparison, means nothing to me. A few brief comments after scrolling through link.
1) No indication how hardware decoding was set, can make a big difference depending on source video specs
2) One writing to C: the other D:, can make a big difference depending on source file complexity and location and hardware behind C: and D:
3) Encode preview enabled during encoding which can be terrible to do during performance test. Recent PD releases don't even allow hardware encoding when preview is enabled.
4) One wrote a 3.5GB file the other 11.0GB, no comparison of output specs meeting profile specs
5) Profile was 80000Mbps video bitrate, total duration 6:32:40, end file should be somewhat larger than 3.14GB as audio encoding bitrate was not shown. At 11.0GB, RX6900XT did nothing as was specified, output file specifics need interrogation.
6) No comparison of quality
7) Custom GOP used and somewhat unique GOP vs more typical. Comparison with default profile would be advantageous.

This test is basically I did something, here it is, totally inconclusive for a good back to back comparison. I'd like to see a little more vigor in the test and results to see real performance difference to make a ~$1700 purchase choice.

I didn't put in a Radeon card to verify, it's too hard to tell from literature as PD uses custom Boris products, not mainstream, but your use of grunge effect may be only CUDA (Nvidia) based. I'm not sure, I'd have to throw in my Radeon RX 5700 to verify.

Jeff

Wow. I must say I don't have much knowledge of #1 (How to set hardware decoding) i just let the software do whatever it does? How does one set that?

I also dont understand gop to be honest. I just know that i've also done 4k60fps in the past and i ended up having to make a custom profile to get it at 60fps.

Quote
Wow. I must say I don't have much knowledge of #1 (How to set hardware decoding) i just let the software do whatever it does? How does one set that?

pref > Hardware Acceleration > Enable hardware decoding, when unchecked, CPU will do timeline decoding. When checked GPU hardware will be used for decoding if the timeline video format is supported by the GPU decoder. For instance, a wmv file will always be CPU decoded as video format is not supported by AMD or Nvidia hardware decoding.

Jeff

You should be able to see the benefits of hardware decoding with this simple test outlined below.

First, we need some higher complexity video for easier demonstration of hardware decoding benefits. To create this,
1) Drop 10 copies of this Skateboard 01.mp4 clip in the timeline so we've got enough content for a good test
2) set pref > Hardware Acceleration, nothing selected
3) Go to Produce tab, select standard NTSC profile, H.264, MP4, with Profile name MPEG-4 4K 4096x2160/30p (50Mbps). Use the “+” to right of profile name to create some customization and on the Video tab select 59.94 for fps and enter 100000 for Average bitrate
4) Produce, this will now be our source video

For the hardware decoding back to back test;
1) Remove Skateboard 01.mp4 clips in the timeline and add the above produced clip
2) Go to Produce tab and select default profiles and use the H.264, MP4 MPEG-4 4K 4096x2160/30p (50Mbps) profile
3) Make sure Fast video rendering technology: NVIDIA NVENC is selected for hardware encoding on the Produce page
4) Produce and record elapsed time
5) Use Previous button to go back to Produce tab area
6) set pref > Hardware Acceleration, make sure hardware decode is selected
7) Produce and record elapsed time, prior specified produce settings should be maintained

If I didn't screw up on directions and things went as desired, you should probably see a reasonable difference between the elapsed times in steps 4 and 7. The timeline hardware decoding and encoding in PD is fraught with bugs/anomalies/idiosyncrasies so care often required to really validate what was done, it may not be as expected. I think the above should work fine on your system with PD21 and your Maxwell based Titan and provide a valid comparison of hardware decoding benefits.

If you do some basic high level monitoring of component load with Task Manager > Performance tab, you should see step 4 having higher CPU load and no GPU decode load relative to step 7 which will have lower CPU load and significant GPU decode load. Both steps 4 and 7 will have GPU encode load but the levels will vary.

Post your times, I'd be interested.

Jeff

EDIT: I should clarify, I'm interested in your results as your system is a little strange. Fairly old GPU technology by most standards with a Maxwell Titan X (although high end in 2015), in a fairly recent and higher end Ryzen 3950x. You might not see big GPU decode benefit as your CPU is rather capable when it's doing the decoding. I'm guessing probably still an advantage though.

Quote
2) Go to Produce tab and select default profiles and use the H.264, MP4 MPEG-4 4K 4096x2160/30p (50Mbps) profile
3) Make sure Fast video rendering technology: NVIDIA NVENC is selected for hardware encoding on the Produce page

I would say that today, everyting should be exported in h265 format. Same bitrate, but better quality.
Of course the hardware encoders need to be latest generation... for nvidia Ampere and Turing GPUs (except TU117). Because they have the most features for HEVC:



https://docs.nvidia.com/video-technologies/video-codec-sdk/pdf/NVENC_Application_Note.pdf

Nvidia NVENC - Wikipedia