Announcement

Collapse
No announcement yet.

TMPGenC Settings for UHD60p output?

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • TMPGenC Settings for UHD60p output?

    I'm trying to output UHD 60p, and the files that TmpGenC is generating don't play well with either Edius or VLC Media Player. In Edius, the files stutter like mad, but CPU use never goes over 15% or so. VLC doesn't even try to show more than the first frame. Neither program has problems with the 150Mbps HC-X1 files they came from. Files upload to youtube fine, but I can't play them, which is sort of sad :( I don't have this problem with L4.2, but L5.2 files seem too aggressive in some way that I can't really determine.
    Example bad file: https://drive.google.com/open?id=0B5...HYtd2pWbVp1LXM
    Example good file: https://drive.google.com/open?id=0B5...3dWbnhWUDlBck0

    I noticed that the default GOP structure is using 250 (?!) frames, which seems ... pretty long? Does anyone know how to peer into these files and tell me how to convince TmpGenC (or some other tool) to create a better format?

    Thanks!
    2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

  • #2
    Originally posted by dnavas View Post
    I'm trying to output UHD 60p, and the files that TmpGenC is generating don't play well with either Edius or VLC Media Player.
    I fixed my VLC issue by turning off hardware decoding. Not sure why VLC is fine with my Panny files, but not with TmpGenC, but there you have it. VLC uses ~18% CPU playing the stream. Sadly, Edius uses 48% playing the Panny streams, and half that with the TmpGenC-output streams (with stuttering). Any pointers welcome! Does anyone else out there have problems playing the linked "bad" file I uploaded (above)? (turn off hardware decode -- I don't have Intel QS :>)
    2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

    Comment


    • #3
      Your bad file plays just fine in Windows Media player, MPC-HC x64, and VLC on my system.....and Edius. Edius also uses 100% of the cpu to play that file.
      This file runs the cpu at 90%:
      4k60ptest.mp4

      See how this plays for you.
      Last edited by Jerry; 03-15-2017, 05:59 PM.
      Jerry
      Six Gill DV
      www.sgdvtutorials.com
      If you own the Tutorials and you need help, PM me.

      Vistitle YouTube Channel
      https://www.youtube.com/channel/UCMVlxC8Am4qFbkXJRoPAnMQ/videos

      Windows 10 Pro up to v.1803 Tweaks for Edius Users
      http://sgdvtutorials.com/WIN%2010%20...%20V.2.0.0.pdf


      Main System:: Azrock Z97 Extreme 6, [email protected], 32gb ram, NZXT Z63, Win10 Pro 64, Samsung 850 pro, E7.5/8.5/E9 on separate SSD drives, 2TB Sabrent M.2 NVME, BM MINI MONITOR 4K, 12tb RAID 0 on backplane ,2 BD, Dual LG 27GK65S-B 144Hz monitors, GTX 1080ti SC Black.
      Second System: EditHD Ultimax-i7, X58, [email protected], Corsair H80, Win764, 24gb ram, Storm 3g, Samsung 840 Pro 256, 4tb and 6tb RAID 0 on backplane, GTX 980ti Classified, Edius 9, Apple 30", Samsung 24", dual BD.

      Comment


      • #4
        Originally posted by Jerry View Post
        Your bad file plays just fine in Windows Media player, MPC-HC x64, and VLC on my system.....and Edius. Edius also uses 100% of the cpu to play that file.
        This file runs the cpu at 90%:
        [ATTACH]16868[/ATTACH]

        See how this plays for you.
        Oddly, I'm not having a problem with CPU use per-se, but rather that Edius refuses to use more than 2-4 threads to decode. Your test file uses 24% of the CPU, and stutters :( -- buffer never gets over 2 frames or so. It's also puzzling to me that VLC (with hardware disabled) uses only half the CPU resources to produce a completely stable output.

        I may try a reinstall. I know my Spark card isn't happy in the new system, and maybe there are other pieces of Edius that aren't as well?


        Originally posted by More4K
        Reduce number of reference frames to e.g. 3. Reduce GOP size to 60. Reduce number of B frames to 2. Also set slices=4 (then set slices to 8 and check if Edius uses more cores) to help threaded decoding. Also set tune=fastdecode. Also set maximum bitrate to something like 100Mbit and mode to VBR.

        Panasonic file uses only 2 ref frames, fixed GOP structure and size at 30, 2 B frames.
        Files are quite different.
        You are using default x264 setting which are made for maximum quality and push things a lot. You need to take control over mentioned settings. The slower x264 setting you use the more it pushes parameters, so you have to overwrite them.
        Thank you -- that helps a lot. I had tried using a GOP size of 30 with 2 B fields (that's how I'm usually running for output to youtube), and had tried reducing ref frames from 5 to 2, but none of that made a difference. I also tried VBR and smaller and larger bitrates. I'll try changing slice settings and tune, which are parameters I'm not familiar with.

        Thanks both of you!

        -Dave
        2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

        Comment


        • #5
          Winner! Couldn't find "tune", but setting slice=4 makes an absolutely enormous difference. <70% CPU, and full buffer. That's still 3-4x VLC, but it's at least usable! 8 slices and higher bitrates used more CPU. 2 was unusable, just like 1 was. Not sure why 8 was oversized, but 4 seems like the sweet spot. Thanks!
          2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

          Comment


          • #6
            Originally posted by More4K
            Depending on the decoder number of slices should about match number of cores in your system
            It looks like Edius attacks each slice with more than one thread, just not enough of them for UHD. ATM, I see about 500 threads, and I don't have (or know of) a great tool to pick out which ones are seeing work. I'm guessing it uses 4, which seems like a reasonable constant if you're going to use a constant....

            [ed: I suppose it's also possible that Edius is decoding more than one frame at a time?]

            You should have tuning presets. Even with slices being the main one you still should reduce GOP=60 and number of ref frames to e.g. 3.
            Yes, I reset back to my standard GOP of half a second (30 for 60p), 2 b-frames, and auto-ref. I believe that this works, though I can't verify given that my machine is busy encoding (why yes, VLC is quite stuttery right now ... color me surprised :)). What I can say is that something with these settings is causing my HD encodings to hard-fail ("Failed to export") after hitting ~33% (I guess 2/3 through first pass of a 13 minute clip). I may need to reset slice to 1, or otherwise muck with settings. That seems like a TmpGenC issue, anyway.

            This test shows that Edius AVC decoder is not optimal and why people with many cores not really getting good performance for AVC based formats.
            Yes, there seems to be a large performance gap between VLC and Edius in general, and a thread/core management issue which should be addressed. I can't imagine working on 8k like this. The performance gap may be explained by having to decode to some internal standard, but I do think it's large enough to warrant a peek as long as you're in there removing constants [hint hint :) ]
            2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

            Comment


            • #7
              VLC is merely a media player and not an NLE. That's like comparing apples to oranges.
              Your concern should be with the files working smoothly in Edius.
              The 3 files in this thread play at full buffer on my machine. I see it more of a system issue than encoding.
              Jerry
              Six Gill DV
              www.sgdvtutorials.com
              If you own the Tutorials and you need help, PM me.

              Vistitle YouTube Channel
              https://www.youtube.com/channel/UCMVlxC8Am4qFbkXJRoPAnMQ/videos

              Windows 10 Pro up to v.1803 Tweaks for Edius Users
              http://sgdvtutorials.com/WIN%2010%20...%20V.2.0.0.pdf


              Main System:: Azrock Z97 Extreme 6, [email protected], 32gb ram, NZXT Z63, Win10 Pro 64, Samsung 850 pro, E7.5/8.5/E9 on separate SSD drives, 2TB Sabrent M.2 NVME, BM MINI MONITOR 4K, 12tb RAID 0 on backplane ,2 BD, Dual LG 27GK65S-B 144Hz monitors, GTX 1080ti SC Black.
              Second System: EditHD Ultimax-i7, X58, [email protected], Corsair H80, Win764, 24gb ram, Storm 3g, Samsung 840 Pro 256, 4tb and 6tb RAID 0 on backplane, GTX 980ti Classified, Edius 9, Apple 30", Samsung 24", dual BD.

              Comment


              • #8
                Originally posted by Jerry View Post
                Your concern should be with the files working smoothly in Edius.
                Of course it is. That and throughput.

                The 3 files in this thread play at full buffer on my machine. I see it more of a system issue than encoding.
                Indeed. Ryzen IPC is just shy of Broadwell in most cases, and my proc is running about 27% slower than yours (if that 4.7G is all-core). If the "bad" file uses only 50% (four threads) of your 4790k to keep your buffer full, then I'll be most upset at my system :) I think the strange thing is that Edius is choosing to use a total of 4 threads to attack the decode problem (I never see more than 25% CPU use, even when stuttering).

                [ed: I agree, there IS a system problem of some kind. That said, changing the encoding should not alter thread use this way. That is an Edius issue.]

                Something I haven't eliminated is that I need to reinstall. My Spark card is only half recognized (Spark driver says it's happy, system says it has an unknown card in the system), and when I try to force it on (in an HD project, of course), I get a blue screen. I have the same problem with my capture card, fwiw. This is what happens when one does not install a system from scratch.... I'm just not looking forward to the reinstall process of everything under Edius ;/

                [ed: Changed comparison of clock speeds, as 4.7Ghz posted speed was higher than the default 4GHz I thought 4790ks all-core ran at]
                Last edited by dnavas; 03-16-2017, 05:59 PM.
                2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

                Comment


                • #9
                  Until that Ryzen system is created from a clean install, any results should be considered suspect.
                  If you just slapped in a hard drive from a different system and platform and didn't build from the ground up, it's no wonder you are having issues.
                  Jerry
                  Six Gill DV
                  www.sgdvtutorials.com
                  If you own the Tutorials and you need help, PM me.

                  Vistitle YouTube Channel
                  https://www.youtube.com/channel/UCMVlxC8Am4qFbkXJRoPAnMQ/videos

                  Windows 10 Pro up to v.1803 Tweaks for Edius Users
                  http://sgdvtutorials.com/WIN%2010%20...%20V.2.0.0.pdf


                  Main System:: Azrock Z97 Extreme 6, [email protected], 32gb ram, NZXT Z63, Win10 Pro 64, Samsung 850 pro, E7.5/8.5/E9 on separate SSD drives, 2TB Sabrent M.2 NVME, BM MINI MONITOR 4K, 12tb RAID 0 on backplane ,2 BD, Dual LG 27GK65S-B 144Hz monitors, GTX 1080ti SC Black.
                  Second System: EditHD Ultimax-i7, X58, [email protected], Corsair H80, Win764, 24gb ram, Storm 3g, Samsung 840 Pro 256, 4tb and 6tb RAID 0 on backplane, GTX 980ti Classified, Edius 9, Apple 30", Samsung 24", dual BD.

                  Comment


                  • #10
                    Originally posted by More4K
                    4.7GHz clock is going to make huge difference (even compared to 3GHz) in Edius and not only.
                    Yep, I do multi-threaded programming for a living (not video encoding, though), enough to be somewhat suspicious of some of those benchmarks... but that's long and tangential. Instead, I think I'll just leave it where Jerry rightly pointed out that I need to do a clean install.

                    It's on the list. When I built my system, they were out of the drive I wanted. I used an old drive to get the OS up and running, and then hacked USB3 drivers onto my existing Win7 install (yeah ... that -- stop sniggering). What I really want is a Windows 10 install on an m.2, though there is a certain degree of cleanliness to Windows 7 that I will miss. Call it old-age/nostalgia, though the fact I have nostalgia over a Windows install is giving me a case of the creeps :>

                    Anyway, life, work, and timing contributed to a concatenation of circumstances that leaves me where I find myself today :|

                    Anyway, I *do* appreciate the help with the slice setting! Saved me many hours. When I have clean Edius numbers for Ryzen, I'll send them along. Given that Skylake-X has been delayed, it's likely to be an interesting choice come May/June.
                    2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

                    Comment


                    • #11
                      Skylake X, along with the name change, kept it at 14nm. Ice lake is scheduled to be 10nm in 2018. I was looking at the Skylake E, but I'm not buying until Intel is at 10nm and PCI-E 4.0 is implemented.
                      Jerry
                      Six Gill DV
                      www.sgdvtutorials.com
                      If you own the Tutorials and you need help, PM me.

                      Vistitle YouTube Channel
                      https://www.youtube.com/channel/UCMVlxC8Am4qFbkXJRoPAnMQ/videos

                      Windows 10 Pro up to v.1803 Tweaks for Edius Users
                      http://sgdvtutorials.com/WIN%2010%20...%20V.2.0.0.pdf


                      Main System:: Azrock Z97 Extreme 6, [email protected], 32gb ram, NZXT Z63, Win10 Pro 64, Samsung 850 pro, E7.5/8.5/E9 on separate SSD drives, 2TB Sabrent M.2 NVME, BM MINI MONITOR 4K, 12tb RAID 0 on backplane ,2 BD, Dual LG 27GK65S-B 144Hz monitors, GTX 1080ti SC Black.
                      Second System: EditHD Ultimax-i7, X58, [email protected], Corsair H80, Win764, 24gb ram, Storm 3g, Samsung 840 Pro 256, 4tb and 6tb RAID 0 on backplane, GTX 980ti Classified, Edius 9, Apple 30", Samsung 24", dual BD.

                      Comment


                      • #12
                        Not on topic but I think related. Looking at Z270 boards from Gigabyte they do not appear as comprehensive as the Z170. The Z170's can have Thunderbolt 3 and HDMI 2.0 as an example both unavailable on the Z270 boards as far as I can tell.
                        Ron Evans

                        Threadripper 1920 stock clock 3.7, Gigabyte Designare X399 MB, 32G G.Skill 3200CL14, 500G M.2 NVME OS, 500G EVO 850 temp. 1T EVO 850 render, 6T Source, 2 x 1T NVME, MSI 1080Ti 11G , EVGA 850 G2, LG BLuray Burner, BM IP4K, WIN10 Pro, Shuttle Pro2

                        ASUS PB328 monitor, BenQ BL2711U 4K preview monitor, EDIUS X, 9.5 WG, Vegas 18, Resolve Studio 17


                        Cameras: GH5S, GH5, FDR-AX100, FDR-AX53, DJI OSMO Pocket, Atomos Ninja V x 2

                        Comment


                        • #13
                          Originally posted by Ron Evans View Post
                          Not on topic but I think related. Looking at Z270 boards from Gigabyte they do not appear as comprehensive as the Z170. The Z170's can have Thunderbolt 3 and HDMI 2.0 as an example both unavailable on the Z270 boards as far as I can tell.
                          TB3 appears to be supported via an add-on card. The inclusion of an HDMI 1.4 port is weird, but I suppose no one is going to game on a 4k monitor running Intel graphics, so maybe they save a dime or something?

                          Closer to topic: picked up different memory from Fry's that was on the vendor list for my GB board. At least it posts at 3200, but I haven't found it stable under load. That said, some of the weird problems I was having generating my HD files are gone when downclocked. At least I'll be able to finish!

                          If it isn't already obvious, Ryzen has a teensy little problem with their BIOS/memory controller.
                          2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

                          Comment


                          • #14
                            Some analysis I've been able to do:

                            Edius has 4 high-load threads active for single-slice videos (at least on my system). There are a number of other threads that come in and out and appear to be doing random bits of work, but they are not significant contributors to load. Altogether, they account for a little more than 25% load on a 16-thread system.

                            On a dual-slice video, Edius has 9 high-load threads active. Why 9 instead of 8? Beats me. They consume a little over 50% of the CPU, which continues to be insufficient to show the video at full speed (but it's close).

                            On a quad-slice video, Edius has 17 high-load threads active. Why 17 instead of 16? Again, I dunno. I counted half a dozen times, just to make sure I wasn't over counting.

                            I'm running under Windows 7, I don't have a Windows 10 system I can use to compare. Additionally, I did a brand spanking new install of Win7 & Edius to make sure that the tests I did for Bassman weren't invalid. There was no discernable difference in speed.

                            Neither install was happy with the HDSpark (in non 4k mode, of course). Trying to force the use of HDSpark in either system resulted in a black screen on the Windows box very shortly after (for the fresh install, this followed the monitor complaining about an incompatible signal). At least the fresh install did correctly activate the card on boot (HDMI on the other end happily received a black screen, instead of no content).

                            If I had an i7 to compare it to, I could probably work to figure out why Jerry's system is fine with the single-slice video. I assume it's a combination of faster RAM and higher clocked CPU, but it's also possible that Intel's QuickSync h/w is contributing to better throughput. I continue to hope that GV will better utilize gfx card h/w sometime in the future so that we don't have to make a tradeoff between i7s and higher core-count processors. Processors aren't getting any faster -- we have to go wide, or just throw in the towel.

                            Bassman also passed a few videos from his PX270 and RX10 cameras to me. Diving videos, which I was amused at, given I had just come back from Victoria (Canada) taping my daughter's diving meet up there. In general we believe that the Ryzen system is somewhat more performant than his i7, but not to such an extent that it's worth upgrading for him. The PX270 videos hit the CPU pretty hard, and getting three of them up on a screen is a real struggle.
                            2950X 32GB Win10, Gigabyte GTX 970 (IXOC), dual EA275UHD, QNAP 873 (w/10GbE). Video source typically HC-X1 or Panny TM700; Audio DR-701D and Rode NT4. Edius 8.5 WG w/ Vistitle & TmpGenC.

                            Comment


                            • #15
                              I tested using:
                              QS
                              CPU
                              TMPGE Movie Plugin with QS pass through and CPU

                              All options play fine on my system.

                              Everything you posted, plays perfect on my system, including when I have QS deactivated.
                              Albeit, there isn't much head room. So, it does have some effect not using QS decode.
                              16 threads should more than make up for QS decode. I know it does on the 5960x and the 6950x because
                              I have built systems with those two chips.

                              I think it's the components, the build, and the tweaking that has my system running the way it does.
                              Last edited by Jerry; 03-22-2017, 04:45 PM.
                              Jerry
                              Six Gill DV
                              www.sgdvtutorials.com
                              If you own the Tutorials and you need help, PM me.

                              Vistitle YouTube Channel
                              https://www.youtube.com/channel/UCMVlxC8Am4qFbkXJRoPAnMQ/videos

                              Windows 10 Pro up to v.1803 Tweaks for Edius Users
                              http://sgdvtutorials.com/WIN%2010%20...%20V.2.0.0.pdf


                              Main System:: Azrock Z97 Extreme 6, [email protected], 32gb ram, NZXT Z63, Win10 Pro 64, Samsung 850 pro, E7.5/8.5/E9 on separate SSD drives, 2TB Sabrent M.2 NVME, BM MINI MONITOR 4K, 12tb RAID 0 on backplane ,2 BD, Dual LG 27GK65S-B 144Hz monitors, GTX 1080ti SC Black.
                              Second System: EditHD Ultimax-i7, X58, [email protected], Corsair H80, Win764, 24gb ram, Storm 3g, Samsung 840 Pro 256, 4tb and 6tb RAID 0 on backplane, GTX 980ti Classified, Edius 9, Apple 30", Samsung 24", dual BD.

                              Comment

                              Working...
                              X