Comments

john_dennis wrote on 12/10/2018, 2:27 AM

Wayne said:

"If anyone is interested in getting "into the weeds" of the VCE encoder..."

I'm already deeper into the weeds than I thought I'd ever be.

Nick said:

"I sampled the length of the histogram at 4 points in the video and took an average."

I've observed that there is a pronounced degradation from an index frame to the last frame before the following index frame. If you sample more often and sample the frame before the index frame you'll more consistently capture the largest differences.

This project has markers one frame before an index frame.

bob-h said:

"There are no B-frames (h.264) in VCE 3.4. Only in VCE 2.0,3.0,3.1"

I stopped using them a few months ago. I noticed that none of my camera video had B-frames and said, "Hmmmm." But, I don't have to answer to anyone.

john_dennis wrote on 12/10/2018, 2:48 AM

bob-h said:

"I downloaded a chart a long time ago from a streaming group/forum, where the quality of encodes was rated for 3500kbit/s"

Was that 3500 kbps or 3500 bps? BruceUSA's VCE output is 25.2 Mbps. I tried to match that bit rate.

I think BruceUSA and Magix have a winner.

fifonik wrote on 12/10/2018, 4:38 AM

The VCE result is pretty good. I downloaded source, rendered with x264 CPU (25.0 Mbps, used CRF 23.2 to get there) and analized crops* with the tool I mentioned earlier.

Blocking (x264 is slightly better):

SIMM (the VCE is slightly better overall):

P.S. I'm using free version of the MSU tool so I cannot analize full frame. I cropped 1024*704 region somewhere in the middle for comparison.

P.P.S. @john_dennis it's 25.0, not 25.2. The last one is overall bitrate, but we comparing video codecs. I know it's not big difference at all.

Last changed by fifonik on 12/10/2018, 4:42 AM, changed a total of 1 times.

Camcorder: Panasonic X920

Desktop: MB: MSI B450M MORTAR TITANIUM, CPU: AMD Ryzen 3700X (not OC), RAM: G'Skill 16 GB DDR4@3200 (not OC), Graphics card: MSI RX580 8GB (factory OC), SSD: Samsung 970 Evo+ NVMe 500MB (OS), HDDs: Seagate & Toshiba 2TB, OS: Windows 10 Pro 1909

NLE: Vegas Pro [Edit] 11, 12, 13, 15, 17

NickHope wrote on 12/10/2018, 7:14 AM

@Former user What versions of CPU/Intel HD Graphics and GPU/NVENC did you use? How's the speed of QSV vs CPU?

Your results seem to imply that QSV may be the preferred option for someone who wants fast AVC rendering on a budget.

Former user wrote on 12/10/2018, 7:26 AM

Hi Nick, see my signature. In the Red car and Running man tests in the long i9 9900k thread I discovered that the best combination going forward for my PC system is HW acc. = Nvidia and Qsv for rendering.

https://www.vegascreativesoftware.info/us/forum/intel-i9-9900k-processor-review-using-vegas-pro--113429/?page=2

https://www.vegascreativesoftware.info/us/forum/intel-i9-9900k-processor-review-using-vegas-pro--113429/?page=3

For these tests, this thread, I used Nvidia for HW Acc. on laptop and PC, I only have Qsv on PC, Qsv rendering with Nvidia gpu for HW Acc. for me, is the best combination. On the laptop i’ll just use nvenc for rendering as I have up to now for casual video.

I didn’t test the speed for this test but the above threads clearly show that Qsv is the quickest render.

“How's the speed of QSV vs CPU” 18s vs 1:48s in the Running Man project.

Former user wrote on 12/10/2018, 7:50 AM

@fifonik “Some encoders produces frames shifts. So when doing diff you should check if you are comparing exactly the same frames.”

Thanks fifonik, maybe i’ll check it out later.

Later. I discovered whats happening with the Nvenc test. I’ve updated my previous post with the details, see above.

Former user wrote on 12/10/2018, 7:56 AM

I wonder if it matters, i.e. the Nvenc discrepancy? Maybe it’s a bit like identical twins. When you see them one by one they look identical from memory, its only when you see them together that you can spot the small differences.

If the for example, Nvenc render, doesn’t have artifacts etc that take from the video then it may not matter.

Maybe there’s nothing at all wrong with the nvenc render, its just different?

In this case one duplicated frame.

john_dennis wrote on 12/10/2018, 8:36 AM

@fifonik

"@john_dennis it's 25.0, not 25.2. The last one is overall bitrate, but we comparing video codecs. I know it's not big difference at all."

I wish that was the only mistake I'd made this week. At least, I matched the audio bit rate, 192 Kbps.  

Former user wrote on 12/10/2018, 1:38 PM

I found that the Nvenc issue of duplicating one frame, appears to be frame rate related. I had no problem with media that was either 25fps or 23.98fps, project set accordingly. However when I mistakenly used 29.97 in the Running Man 23,98 project I encountered the same issue. It shows up in the BruceUSA project for this reason, its a 29.97fps media project.

wwaag wrote on 12/10/2018, 2:21 PM

The duplicated and overwritten frame problem is clearly a bug in the Vegas implementation of NVENC rendering. Using Happy Otter Scripts, there is no duplicated frame and the preview window remains dark throughout the test clip. The easiest way to test is to add a Timecode Fx (e.g. absolute frames), then render and add back to the timeline which HOS can do automatically. Here are a couple of screenshots at 2:29 showing the frames of original and rendered are the same. Also a screenshot of the preview window and scopes when Difference is chosen.

 

Former user wrote on 12/11/2018, 12:50 PM

Hi @wwaag, have you confirmed the issue also with Nvenc 29.97 fps? 

 

wwaag wrote on 12/11/2018, 1:46 PM

That was an NVENC 29.97 render.

john_dennis wrote on 12/11/2018, 2:47 PM

I started a Vegas Pro 16 trial to some extent out of my curiosity about the efficacy of the AMD VCE encoder. My first effort was discouraging, since I had green artifacts in the upper part of the Preview and rendered output as shown:

I updated my Radeon RX480 video driver to Radeon Software 18.9.3 and the render worked as expected, taking 39 seconds

which was an improvement over the first render which took 53 seconds to produce a defective video.

 

BruceUSA wrote on 12/11/2018, 3:00 PM

I started a Vegas Pro 16 trial to some extent out of my curiosity about the efficacy of the AMD VCE encoder. My first effort was discouraging, since I had green artifacts in the upper part of the Preview and rendered output as shown:

I updated my Radeon RX480 video driver to Radeon Software 18.9.3 and the render worked as expected, taking 39 seconds

which was an improvement over the first render which took 53 seconds to produce a defective video.

 

Never got such issue here. I love my Vega card. At the movement there is no such card can match it's performance on Vegas and encoding quality.

john_dennis wrote on 12/11/2018, 3:14 PM

Never got such issue here. I love my Vega card. At the movement there is no such card can match it's performance on Vegas and encoding quality.

I run all these trials on a "gold" system image that I restore from a backup. I have to be really bored or have a failure to cause me to update my video driver. Without either of those, I'll keep the driver that I installed when I built the system. This is from experience. I've been burned from changing drivers and from not changing drivers.Since my "gold" system works OK up to Vegas 14, I'll have to test for a performance improvement before I update the driver in the system image.

I can match the encoding quality of the AMD VCE encoder, but and here's the big but, not at the same bit rate. I haven't even looked seriously at render times though I kept all the data.

I'm very impressed with the AMD VCE encoder at low bit rates. Thank you for posting.  

Former user wrote on 12/12/2018, 2:07 AM

 

@BruceUSA Any chance of sharing the original clip and your VCE-rendered clip to eliminate the downscaling and Vimeo's recompression? Could make a nice little benchmark for others to try NVENC, QSV, x264 etc..

@NickHope Is this a new “known issue” i.e. adding a duplicate frame after 2s 27f while rendering from 4k 29.976fps to 4k using Nvenc? @wwaag and I can confirm using the @BruceUSA benchmark. I see you probably can’t confirm as you use an Amd card, maybe someone else can confirm also?

NickHope wrote on 12/12/2018, 2:16 AM

 

@BruceUSA Any chance of sharing the original clip and your VCE-rendered clip to eliminate the downscaling and Vimeo's recompression? Could make a nice little benchmark for others to try NVENC, QSV, x264 etc..

@NickHope Is this a new “known issue” i.e. adding a duplicate frame after 2s 27f while rendering from 4k 29.976fps to 4k using Nvenc? @wwaag and I can confirm using the @BruceUSA benchmark. I see you probably can’t confirm as you use an Amd card, maybe someone else can confirm also?

@Former user It's not listed yet. It would need more testing with various test media to be sure it's repeatable and to define the parameters. i.e. Does it happen with all media? At the same frame on all media? All resolutions? Which NVENC versions/drivers? Once per event or repeatedly? etc..

Former user wrote on 12/12/2018, 2:40 AM

Thanks. I agree that it needs much more confirmation. Later i’ll generate a different piece of 29.976 and test using Nvenc and see if it repeats, as a start. Its an Nvenc issue though, not a Vce issue. So far it appears on my laptop and PC, with different Nvidia gpus and different Nvidia drivers.

NickHope wrote on 12/12/2018, 2:58 AM

Oops, yes NVENC, not VCE. Will change my comment.

Former user wrote on 12/12/2018, 5:49 AM

@NickHope

Ok, I tested 3 new source video clips from a GH5, A7R II and a Samsung S7 edge, all 4K UHD 29.976fps.

The first 2 exhibited similar duplicate frame result issues, i.e. the 2s;27th frame

 and 2s;28th. frame are the same. the S7 had no issue, but it is VFR.

I tested Pana and Sony on my Laptop and Desktop which still have 2 different Nvidia driver versions, PC=4.1694 and Laptop=4.1722.  The graphics cards obviously are different also.

Just a note, I had to upgrade the laptop driver to 4.1722 as I was initially getting half length and twice length video renders using Nvenc with the GH5 sample.

I disabled the SO4 .. .dll.  This had the effect of moving the dublicates frames to a different position i.e. 3s;20/21f for the Sony and to 4s;00/01f for the GH5. 

Former user wrote on 12/15/2018, 11:49 AM

@NickHope

1) Does it happen with all media? 2) At the same frame on all media? 3) All resolutions? 4) Which NVENC versions/drivers? 5) Once per event or repeatedly? etc..

Hi Nick, as you know @wwaag has also confirmed with the same @BruceUSA media.

Item 1) No but the exception was 1 piece of vfr media, the other 3 pieces of different media do have the same issue. 

Item 2) Yes the same frame, if So4 .. .dll is used, otherwise it varies.

Item 3) At 4k UHD & 1920x1080 FHD, 29.97fps resolution.

Item 4) I’ve listed my graphics drivers in my signature, but I used a total of 3 different Nvidia drivers.

Item 5) Once only per event.

Item 6) Etc?

 

Former user wrote on 12/16/2018, 3:58 AM

I’ve now tested with a FHD test clip, it gives the same result.

bitman wrote on 12/17/2018, 2:10 AM

@Former user I will see if I can reproduce on my system, give me a couple of days, pretty busy at work now.

Current system: VP 17 (build 421), VP16 (suite) build 424, (uninstalled VP 13,14,15), Vegasaur, Magix Video Pro X (VPX11), Corel VS ultimate 2019, a lot of NEWBLUE plugins, Titler Pro 6, Mercalli 4.0, Respeedr, Vasco Da Gamma 12, VASST stuff, Production Assistent pro3, Boris Continuum 2020, Davinci Resolve Studio 16,...

  • OS: Windows 10 Pro 64, version 1909
  • CPU: i9900K stepping R0 (since October 2019), previously, der8auer i7-8700K (advanced edition), default speed (no overclock), Cooler: Noctua NH-D15s
  • RAM: G.Skill Trident Z 3200C14 DDR4 64GB, XMP set to profile 1 in BIOS
  • Videocard: NVIDEA RTX 2080Ti (Founders edition), NVIDEA studio drivers
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive (games & APPS): Samsung NVMe SSD 2TB 960 pro

  • Current Video source work drive: Samsung NVMe SSD 2T 970 EVO plus

  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB

  • MOBO: Gigabyte Z370 Aorus Gaming 7, BIOS F14
  • PS: Corsair HX1200i, Case: Silverstone fortress 2,
  • Misc: Logitech G910, Evoluent Vertical Mouse, shuttlePROv2

 

Former user wrote on 12/17/2018, 5:04 AM

@bitman aok, no hurry at all, thanks.