Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

general questions about Neat Video
Post Reply
CraigT
Posts: 6
Joined: Tue Dec 17, 2019 6:48 am

Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by CraigT »

I've been using Neat Video for 7+ years now with both Sony Vegas Pro/Magix Vegas Pro and VirtualDub/VirtualDub2, with much satisfaction.

I hope to build a new pc in a few months but am not sure what to focus on, CPU or GPU or both, and had some questions. While I’m fairly sure I’m going with the Threadripper (2nd or 3rd gen), the Ryzen 9 3900X was the clear winner in terms of CPU speed.

Current build:
Intel i7 5930K 3.5GHz 6 core (40 lane mobo chipset), 32GB RAM, EVGA GTX 980 Ti

Intended Use:
Handbrake or some other transcoder
Plex server - nothing requires transcoding so this should not have much overhead
Security Cameras – still tbd, up to 16 8MP cameras, no transcoding, should not require much overhead
VirtualDub2 with Neat Video
Vegas Pro with Neat Video

1. CPU vs GPU Quality:
Has anyone found using the GPU to be of any lesser or different quality than the CPU? I found artifacts (large blocky areas of solid black) on 2-3 frames when using the CPU+GPU as well as GPU only, but never with CPU only. I discovered later on that I was using a very outdated video driver which has since been updated. But since those bad results I have not attempted using the GPU again at all.

2. CPU vs GPU vs 2x GPU Speed:
From multiple benchmarks it appears the AMD Ryzen 9 3900X takes the lead for CPU speed. However this CPU's board's chipset only supports 24 lanes, which I believe would not allow full potential for one or two GPUs plus 3x m.2 SSDs.
It also appears the GeForce RTX 2080 Ti takes the lead for GPU speed, albeit a huge financial hit.
I have found two benchmarks using two GPUs (one had different GPUs while the other had two of the same), but I believe the chipsets did not have enough lanes to fully support both GPUs and significantly limited the results.
Has anyone tested two identical GPUs on an Intel board with 40+ lanes or on a Threadripper board with 64 lanes?
The AMD Radeon RX 5700 XT looks interesting since it's roughly 20% slower than the 2080 Ti, but less than half the price.
I would really like to see the benchmark of two AMD Radeon RX 5700 XT cards on a mobo with enough lanes to fully support both GPUs. If this could get close to double the fps of one GPU, it would be much faster and cheaper than the 2080 Ti, plus it would free up the CPU for other tasks.
I also can't wait to see how the Threadripper 3970X compares to the Ryzen 9 3900X.

3. Has anyone tested if the benchmarks stay relatively consistent ratio-wise between different CPUs/GPUs for different Neat Video settings? I max out the settings in quality and radius, and it does take considerably longer than the default settings. I’m curious if these max settings change the ratio of the benchmark results between CPUs/GPUs, basically changing the winner or separation for GPU and/or CPU speed.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by NVTeam »

I recommend to check our recent article about CPUs and GPUs and also check the NeatBench results in this recent review of AMD CPUs.
We couldn't directly test those CPUs yet, but the results presented in that review look very promising.

Normally there shouldn't be any difference in quality with CPU vs GPU processing, unless of course there is some problem with the GPU drivers. Some drivers contain bugs and if we find out those, then we try to report the problem to the developers of those GPUs to make the bugs fixed as soon as possible. If you encounter any GPU-specific problems causing reduced quality or artifacts, please send such cases to support at neatvideo.com

Thank you,
Vlad
CraigT
Posts: 6
Joined: Tue Dec 17, 2019 6:48 am

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by CraigT »

Thanks for the info. I had read your article on CPU/GPU but was not aware of the benchmarks on legitreviews. Both were helpful.

Using NeatBench, my CPU was 11.6fps while the 980Ti was 23fps. The Radeon RX 5700 XT benchmark provided was roughly a 50% gain, which might be all I need instead of a new system, and at a decent price.

So I attempted another GPU-only run with the latest video drivers, but after using AviSynth Compare() I was able to quickly find both numerous splotches of solid black in the frame as well as repeated frames where one frame might repeat itself 10 times even tho the following frames were vastly different. So I'm hesitant with buying a newer video card and using GPU, but wow would that reduce project time considerably.

I could send a request for the video issue to you, but do you think they would focus on a 4-1/2 year old GPU?

Is anyone else having these issues with the 980Ti?
Is anyone getting acceptable results using any of the newer video cards running GPU only?

Thanks,
Craig
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by NVTeam »

Do you see (using just your eyes) any problems in the output of Neat Video when using GPU?
If you do, please send such reproducible cases to support [at] neatvideo.com and we will check that from our side
(run the same render on our machines, reproduce the problem and then trace it back to the source).
It may be not specific to one old GPU model.

Thank you,
Vlad
albino1
Posts: 20
Joined: Fri May 17, 2019 8:28 pm

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by albino1 »

I sometimes get random splotches with some sources if I have Dust & Scratch removal turned on. I would check to see if you have that enabled and then try turning it off.
CraigT
Posts: 6
Joined: Tue Dec 17, 2019 6:48 am

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by CraigT »

Thanks again for responses.
I provided a bug report.
The issue can black out 1/4th the entire frame, so it is obviously (painfully) visible, while other times it can black out very small boxes, requiring a comparison tool to easily find, especially since it might only corrupt 3 frames on an hour long video.

I bought a new video card and will install and try it probably Monday/Tuesday. If it works, I won't have much reason to use this 980 Ti card again.

albino1 -- Thanks for the tip, but I don't use Dust & Scratch. While when it works it's impressive, it usually corrects detail that I don't want it to, regardless of the settings I've tried.
CraigT
Posts: 6
Joined: Tue Dec 17, 2019 6:48 am

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by CraigT »

I installed a new video card, AMD Radeon RX 5700 XT, and the original missing parts of the frame are gone when rendering GPU-only.
This is 3x faster than my CPU-only renders.
It's actually faster than 3x, since I can run HandBrake using the CPU on a previously cleaned up video while the next video is running Neat Video using GPU only with a small performance hit. So it's really closer to 4-5x faster than using the CPU-only for the entire start-to-finish process.

I am still having errors where Neat Video identifies multiple frames as copies of one another even though they are not identical frames.
This is causing previous frames to replace following frames for up to 14 frames.
As one image fades to black, followed by 10-20 frames of black, followed by credits scrolling up, the first 14 frames of the credits remain black, then pop on the screen where they should be in frame 15, but frames 1-14 of the credits are gone. I have repeated this issue numerous times with multiple video sources.
GPU-only has this issue while CPU-only does not.
GPU-only has a graphical display issue when selecting a different frame at the bottom of the preview screen.
This issue existed with both my original GPU as well as my new GPU, both NVIDIA and AMD.
I will submit a bug report.
NVTeam
Posts: 2745
Joined: Thu Sep 01, 2005 4:12 pm
Contact:

Re: Questions on CPU/GPU Quality & CPU/GPU/2xGPU Speed

Post by NVTeam »

Thank you for reporting and sending additional details to support. We are working on this issue.
Post Reply