[HELP] perf. RTX 3060 ti slower than GTX 1070 - NVEnc

[Dan64]

Just to provide an explanation on why is better to use "NVEnc->Harwdare->Only use encoder". 
Hybrid provides the following description for the option: Config->Input->Use gpu for decoding" 

When enabled Hybrid will add '-hwaccel auto' to ffmpeg decoding calls. [...]. Note that most acceleration methods are intended for playback and will not be faster than software decoding on modern CPUs. Additionally, ffmpeg will usually need to copy the decoded frames from the GPU memory into the system memory, resulting in further performance loss. So this option might not provide any additional speed and thus is mainly useful for testing.

When the decoding is performed directly by NVEnc (by enabling NVEnc->Harwdare->Only use encoder ) the process is more efficient and you can gain an extra encoding speed and be able to encode at least at 90 fps. I really don't understand what is the problem in using this option.

[mimile]

Selur, here is the debug with the option "only use encoder" with the RTX 3060 Ti :
https://uptobox.com/b7ci6dgamzp8

You are right Dan64, the encoding is faster by checking the option "NVEnc-> Harwdare-> Only use encoder" but only for the GTX 1070, not the RTX 3060 Ti because it is another problem I think.

summary test result
GTX 1070 without "NVEnc-> Harwdare-> Only use encoder" and "uncheck use gpu decode": 41 fps

GTX 1070 without "NVEnc-> Harwdare-> Only use encoder" and "check use gpu decode" 24 fps

GTX 1070: with "NVEnc-> Harwdare-> Only use encoder" and "uncheck use gpu decode": 51 fps

GTX 1070: with "NVEnc-> Harwdare-> Only use encoder" and "check use gpu decode": 50 fps
- about 45-50% use "video encoder"

RTX 3060 Ti all time 21 fps !! and 100% "video encoder" use

is your RTX 3060 at 100% during your encoding?

when the RTX is installed on the pc, I have problems with the mouse freezing, and when I type text with the keyboard, sometimes a letter is repeated several times for 1 second.

[Selur]

this time the encoding call looks correct:

"C:\Program Files\Hybrid\64bit\NVEncC.exe" --avhw  -i "J:\Download 10 To\Film 4K UHD\The.Protege.2021.2160p.UHD-001.mkv" --fps 23.976 --codec h265 --profile main10 --level 5.1 --tier high --sar 1:1 --lookahead 32 --output-depth 10 --vbrhq 19769 --max-bitrate 10000 --gop-len 0 --ref 3 --bframes 0 --no-b-adapt --mv-precision Q-pel --preset quality --colorrange limited --colorprim bt2020 --transfer smpte2084 --colormatrix bt2020c --max-cll 1000,923 --master-display G(13250,34500)B(7500,3000)R(34000,16000)WP(15635,16450)L(10000000,1) --vpp-resize auto --output-res 3840x2160 --vpp-gauss disabled --cuda-schedule sync --output "J:\Download 10 To\4K UHD enc\The.Prote test 2_2022-01-07@17_20_47_5810_02.265"

May be something is of with your driver settings. Or some anti virus software is interfering,...
The RTX decoder chip should be faster than the GTX chip.

Cu Selur

[mimile]

Disabling the antivirus does not change anything, I will wait for nvidia to update the drivers.
Do you think Rigaya will have a solution?
Thank you for taking care of my problem  
I will look for other solutions ...

[Selur]

Rigaya might at least know more about the issue.

I will wait for nvidia to update the drivers.

You might also want to check your settings and do not overclock your card, overclocking the card often causes the vpu/vpe to effectivly work slower.

Cu Selur

[Dan64]

You don't have specified the version of NVEnc used. This is the full output of NVEnc on my PC

NVEncC (x64) 5.42 (r2072) by rigaya, Dec 11 2021 13:49:16 (VC 1929/Win)
OS Version     Windows 10 x64 (19042) [UTF-8]
CPU            Intel Core i9-10900 @ 2.80GHz [TB: 4.91GHz] (10C/20T)
GPU            #0: NVIDIA GeForce RTX 3060 (3584 cores, 1837 MHz)[PCIe3x16][472.12]
NVENC / CUDA   NVENC API 11.1, CUDA 11.4, schedule mode: sync
Input Buffers  CUDA, 32 frames
Input Info     avcuvid: H.264/AVC, 3840x2160, 30/1 fps
AVSync         vfr
Vpp Filters    cspconv(nv12 -> yv12(16bit))
denoise(pmd): strength 100, threshold 100, apply 2, exp 1
cspconv(yv12(16bit) -> p010)
ssim psnr (yv12(10bit))
Output Info    H.265/HEVC main10 @ Level auto
3840x2160p 1:1 30.000fps (30/1fps)
Encoder Preset default
Rate Control   CQP  I:20  P:22  B:25
ChromaQPOffset cb:0  cr:0
Lookahead      on, 16 frames, Adaptive I, B Insert
GOP length     300 frames
B frames       4 frames [ref mode: each]
Ref frames     4 frames, MultiRef L0:auto L1:auto
AQ             off
CU max / min   auto / auto
VUI            matrix:bt2020c
Others         mv:Q-pel

In my case is the version 5.42. Your Card is quite new and probably is managed better with the last version:

https://github.com/rigaya/NVEnc/releases....42_x64.7z

Also I suggest to don't use 2-pass encoding but the "constant quantizier" (with CQP  I:20  P:22  B:25) as shown in my previous screenshots.
You get constant quality across the frames and it is a little faster.

[Selur]

Also I suggest to don't use 2-pass encoding but the "constant quantizier" (with CQP I:20 P:22 B:25) as shown in my previous screenshots.
You get constant quality across the frames and it is a little faster.

Constant quantizer isn't constant quality, that's not how quantization works, but yes, constant quantizer is faster.

Cu Selur

[Dan64]

I suggest also to use the Encoder preset "default" instead of "quality", since this preset decrease significantly the speed without a significant increase in quality. If you enable "calc PSNR" and "calc SSIM" in the tab NVEnc->Misc and the option to generate the encoding report under "Config->Internals->Create report file", you can evaluate quantitatively if the increase in quality is significant for you, but first check also the increase in speed.

[mimile]

I did not touch the overclock of the card (Gigabyte eagle RTX 3060Ti LHR), it remained original.
Otherwise I may have forgotten a setting in Windows 10 ...
I know that since the launch of the RTX 20 .. nvidia has implemented a new encoder that allows you to play and stream at the same time without losing fps in games, I think it works differently from previous generations (GTX), the solution is to be sought on that side in my opinion.

Dan64, I am using NVEnc version 5.42.
I obviously know that if I lower the parameters it will encode faster, but that's not what I'm looking for, as long as I stay on my GTX so ...
I want to compress my videos as much as possible without compromising on quality, in UHD with my parameters and at around 18,000 kbit / s that's perfect, below the artifacts / pixel start to appear.

What is not normal is that the GTX goes faster than the RTX with exactly the SAME settings!

I actually tested by lowering the parameters and it goes faster! ... but the "video encoder" still works at 100%, the GTX does not have this problem.
I can even perform 2 or 3 encoding at the same time with the GTX, which will be impossible with the RTX

Dan64, is your RTX 3060 at 100% (Video Encoder) while encoding ?
if "yes", the solution to my problem will increase the performance of your encodings normally.

[Dan64]

Dan64, is your RTX 3060 at 100% (Video Encoder) while encoding ?
if "yes", the solution to my problem will increase the performance of your encodings normally.

This is the ouput of GPU-Z



But I will not use NVenc to perform a benchmark of your GPU, it is better to use specific tools for this kind of check.

I suggest to use this tool: https://www.userbenchmark.com/

Good Luck!