CUDA hardware acceleration in VS19

[Matt8890]

Hi I've been talking with the Corel Support desk for VS18, which they have confirmed it doesn't support CUDA, however VS19 does. I only have the trial version of VS19 and so far I have not seen any difference with CUDA hardware acceleration if anything it has made the clips more jerky and laggy. Interested in other peoples experience. I thought this might have been the solution to not needing proxy files which are a pain for long clips. I have a GTX1050 card and it shows no utilisation for encoding and decoding and never gets above 10% overall performance.

[1jhill]

That’s interesting I’ve heard it does make an improvement in render times for pinnacle which is what new VSP 2019 seems to be using (in terms of engine). I was looking to get an RTX 2060 as my next card instead of AMD because of CUDA support but if it’s not making a good improvement then maybe not. Still hoping they implement OpenCL so buyers don’t have to pick and choose.

[BikerDave1]

Back in Jan '18 I posted I was hoping Cuda v6.1 could be implemented for GPU rendering in CVS but alas no-go still. At this point I am thinking it must either be compatibility or licensing issues. I would find it hard to believe Corel would purposely keep the GPU throttled at the bottom. One would never know since Corel uses this 3rd-Party Forum platform for its Support and no employees post here.

[iNate]

That’s interesting I’ve heard it does make an improvement in render times for pinnacle which is what new VSP 2019 seems to be using (in terms of engine). I was looking to get an RTX 2060 as my next card instead of AMD because of CUDA support but if it’s not making a good improvement then maybe not. Still hoping they implement OpenCL so buyers don’t have to pick and choose.

OpenCL has absolutely nothing to do with Encode Acceleration. OpenCL is for effects, image processing, etc.

NVENC, QSV, and VCE are done by completely separate chips. These chips just happen to be packaged with GPUs (or iGPUs) with specific processors or Graphics cards... but they don't do graphics processing. They just do media Encoding or Decoding - depending on which chip you use.

They can do flawless OpenCL, and if you don't have a GPU or CPU that supports the Encode or Decode tech they implement support for, you will still get no Acceleration.

Currently, AMD CPU + AMD GPU is a very bad combination for almost all consumer editors (except PowerDirector, and VEGAS Movie Studio for Encoding) as most NLEs do not implement support for UVD and VCE. QSV and NVENC is predominant.

As far as VS2019 not giving big boosts over 2018, this is expected, because 2018 already had support for NVENC Encoding. The only thing they seemed to do was extend this support to H.264 and QuickTime files, instead of "only HEVC" like in VS2018. If you ran 2018 on a PC with QSV and an NVIDIA GPU, the benefits in VS2019 aren't going to be great, because you aren't going to see massive boosts in Encode Speed moving from QSV to NVENC, unless you have a really high end NVIDIA Pascal/RTX card paired with an older Intel CPU (like a Haswell i5, or something).

I believe I've already given you this information in another thread, but you are still posting this as if you refuse to believe it...

Pinnacle Studio does automatic render caching, so the final renders are often quite fast especially if you render the final video from the cache. This means that you don't have to render all of the effects, transitions, etc. from scratch when you render out the video. This also improves playback performance (though most consumer NLEs are capped at FHD playback, anyways).

[Davidk]

It's highly likely that the particular version of Nvidia (owner of CUDA) display adapter you use isn't supported by the display driver you have installed.
If you search wikipedia on "CUDA", you'll find an interesting and lengthy discussion of that.
Synopsis:
- explanation of how it works (basically, api in display driver to enable programs to use the computing capacity of the adapter to help the main program/cpu do it's job)
- the latest software isn't backwards compatible on all models of older hardware
- about 19 versions of software and 287 approx versions of hardware likely to be involved with CUDA.
And you'll note that there is nothing in the selection of the VS2019 nvidia acceleration option to specify just which model of hardware/driver you have, nor does it say the combo you have is not supported.

But on testing (rendering a project file with all the usual SD video/photo/music/voiceover/text components, play time of roughly 4 mins) using 2019 setup alternatively for both nvidia and Intel versions of hardware acceleration in 2019, those settings did not make any significant difference to the result. Counting the seconds - about 10 sec slower with hardware accel on as opposed to off. What did make a difference was changing the OS priority for the program itself from normal to realtime - and even then the apparent improvement was only about 6-10%.

There are a number of user forum threads on this 'hardware acceleration' topic.

[iNate]

GTX 1050 supports CUDA. It has 640 CUDA Cores, and supports everything CUDA has to offer, including Encode and Decode Acceleration for all supported formats (for Pascal Generation Cards).

Hit GPU isn't the problem.

Read my reply above... The only time you will see activity on the card is when Rendering Media, and you will only see that for "Video Encode" in task manager when you click on that GPU. You might see some activity for 3D and Copy, depending on your configuration... But, beyond that... not much.

VSP doesn't have a CUDA-optimized Enging for Playback and Effects Rendering. It only uses NVENC for Encoding supported formats; and maybe NVDEC for Decoding (haven't tested this part).

VS2018 supports much of the same CUDA as VS2019. If you render HEVC from VS2018, it will Accelerate the Encode with NVENC. I have tested this. I did a lot of testing around this, and was one of the people who started the lengthy discussion about it after VS018 was released.

The big issue with VS2018 was how it's locked to the Intel iGPU in Optimus Setups, which causes applications/plug-ins it spawns to perform terribly, as they're locked to the same GPU that the NLE is locked to (as spawned processes). VS2018 isn't allowed to run on the Nvidia Discrete GPU. The Nvidia Control Panel doesn't let you change what GPU it runs on. Only the Intel iGPU is enabled for it.

Maybe that part has changed for 2019 The other parts won't change, until the NLE gets a new video/effects engine. Aside from rendering, everything uses the default tech in Microsoft Windows, DirectX Video Acceleration, etc.

[Matt8890]

Thanks what I take away from this is that hardware acceleration is coded for rendering the final project. not for decoding the clips, in my case FHD GoPro clips in a way that they play without jerking or freezing. For now I a hearing i will need to continue to use SmartProxy or find another editor
It's a shame because the CUDA acceleration in Pinnacle Studio 22 decoded and played and scrubbed the clips in realtime. You could see the GTX1050 got to about 40% in performance manager
The challenge with a number of the GoPro clips is that they are 8 mins long and it takes hours for smart proxy to make the files. In many cases I want scrub through quickly and determine which bits i want and move on. many of them are from the camera filming while driving on holidays.

[asik1]

Browse thru this comparison
https://www.youtube.com/watch?v=297stHVZJac

[Davidk]

Hard to do that if you don't speak german . . . .

[asik1]

Hard to do that if you don't speak german . . . .

Whats there to understand ? just view the last few minutes. his laptop is faster.

Shortly our forum germans will give us the final points