Comskip donator version

Does anyone have any idea if by donating for Comskip version here https://www.comskip.org/ will you get another more complete version? or there is no difference between downloaded one and donator one?

Thank you!

The donator version that comes with MCEBuddy is quite old (version 0.82.003 from 2017-09-30). The current donator version is 0.82.012 from 2022-02-25. The notable thing the more current Comskip donator version has is a newer embedded FFMPEG v4.3, and supports 4k and 10bit video processing. Donator version is what gets you mpeg4 video processing.

However it doesn’t really need to have all kinds of modern codec support, because most input DVR video formats are MP2 or MP4 anyway and almost all transcoding will start with a conversion to MP2 before MCEBuddy triggers comskip. The hardware CPU/GPU support comes from the embedded FFMPEG baked into comskip.

The summary is yes, it will speed up the ad detection step, but overall, ad detection is only a really small part of the overall transcoding process.

In the end, it’s a personal decision for you. Would you benefit? Sure. Are you missing out or taking a huge performance hit if you don’t? Not really.

A couple of updates, we have a new donator version of Comskip in the works. It’s built against ffmpeg 6.1.1 and has support for many more codecs including native HEVC and AV1 decoding with a custom av1 decoder for maximum performance. The advantage (when released) will be MCEBuddy will no longer have to remux to MP2 which will improve the conversion speed as well as end quality. We’re also testing native hardware decoding for hevc and av1 for comskip with that build but there are some challenges related to stability so we’re testing the trade off between hardware decoding performance vs stability. We may possibly provide an option for users to enable/disable hardware decoding.

1 Like

That is some excellent and welcome news to hear!

Comskip is really only about decoding, and since the codecs are intentionally designed to be decoded fast, and with CPU-only availability (no dependency on GPU or even iGPU), I totally understand the stability and testing across all the generations of both CPUs and GPUs by multiple manufacturers (Intel and AMD notably, and adding in nVidia for GPUs). Not to mention motherboards and all the interplay between media storage, memory, CPU cache, and I/O busses (north/south bridges).

It’s not trivial, so I appreciate all the work that goes into a release, and don’t worry about trying to rush it.

Try the 2.6.4 beta version. We now have a new custom 64bit build of comskip donator. It’s much faster than the old comskip and has some nice features:

  • Support for direct decoding for AV1, HEVC, H264, MPEG2, VP9, VP8 and many more codecs and containers
  • Support for hardware decoding (currently limited to Intel QSV and Nvidia CUVID). These only support newer GPU’s, e.g. won’t work with Intel Sandy Bridge. The earliest intel chipset we could get qsv hardware decoding to work is Skylake (9th gen).
  • Support for 8K resolution

With this release remuxing has also been optimized so it won’t remux as much when using our build of comskip (it still retains compatibility for the stock comskip and free comskip builds but it will be slower with those builds), esp when decoding codecs like AV1.

Regarding hardware acceleration for decoding with comskip, you’ll need to manually enable it for now from the Conversion task → Expert settings → Comskip options

Here you can enter --qsv if you have an Intel chipset or --cuvid if you have a Nvidia gpu.

Try it out and let us know how it goes for you. For comparison we have seen no improvement in comskip performance when decoding h264 source videos, ~30%-50% improvement when decoding HEVC/H.265 source videos. The one thing we haven’t been able to test is decoding performance of comskip on AV1 source videos. If you have chipsets which can decode AV1 files feel free to try it out and post your results back to us. If you need an AV1 source video test, just use MCEBuddy’s AV1 profile to create a AV1 video output. The logs will show you the average FPS for comskip (and total time taken). We have seen 1500+fps for HEVC files when using hardware decoding.

Note: The reason we haven’t enabled hardware decoding for comskip by default is becuase we’ve noticed that it tends to hang if the underlying chipset is too old or unsupported, we will likely optimize that in future.

@techpro2004 @SystemIdleProcess

1 Like

Just wondering what the use case is for comskip and other video encoding formats? Is Blue Ray X.265? Do we have 4K ATSC3 OTA DVRs that encode to H.265 and that media has commercials that need comskip to remove them?

Same for AV1 media. I apologize if I’m a bit out of the loop on paid services like YouTube Premium that stream live OTA channels in H.265 or AV1 and there are still ads being injected into the content that people want to remove (via comskip).

Like Bezos can’t just suck it up, tighten his belt a bit, and lay off the avocado toast, frappes and luxury yachts by keeping Prime Video ad-free. I mean he only increased his net worth by 50% during the pandemic and still pays near zero taxes. But I digress.

Does someone have examples of DVRs that have input and output video sources with ads that are in AV1 or H.265? What DVRs and what video sources? Any product recommendations to replace my aging Tivo Roamio OTA (ATSC1.0) and Silicon Dust HD HomeRun Quattro tuners and DVR?

nvdec logs below on a rtx 4000 ada. I noticed there are newer versions of ffmpeg and handbrake available.

nvdec.zip (924.2 KB)

1 Like

arc logs below

qsv.zip (909.3 KB)

1 Like

These are super helpful benchmarks, do you happen to have the logs from the software run (without --qsv or --cuvid)?

The results so far are:

Comskip
(FPS)
Nvidia
(–cuvid)
Intel
(–qsv)
Software
(libdav1d)
MPEG2 688 146
H.264 598 144
HEVC 448 98
AV1 117 339

*QSV AV1 comskip hung and did not complete in your system, likely due to a graphics driver issue causing it to become unresponsive

software logs below.
software.zip (1.0 MB)

1 Like

@techpro2004 What is your source input video, duration, format, resolution, and does it have ads?

Just generic info. e.g. DVD rip, 90 min, MPEG2, 480p, etc.
How did you control for up/downscaling and transcoding quality across the different formats to start with?

I’d like to try a similar benchmark on something that is comparable for my system. I can record OTA 1080i HD, 720p HD or 470p SD. I set my SD HDHR DVR to record the raw HDTV stream. And I am aware there are other media formats, resolutions, codecs, and bitrates on the intertubes if I need to run those.

I recorded a program in channels dvr from my hdhr. I then transcoded it to the various formats in mcebuddy. it was a 30 min mpeg2 1080i. it does contain ads.

Thanks. I’ll try to run something similar over the weekend.
If I understand MCEBuddy correctly, it runs comskip before transcoding on the raw input media (i.e. almost always MPEG2) , marks the vid, cuts, and then transcodes to the target encoding and format.

Or does it transcode first, so that comskip is scanning and marking the output encoding for ads (e.g. H264, HEVC, AV1), not the input encoding (MPEG2).

It just seems to me that the input format is almost always MPEG2, so why would comskip marking ads in the output formats be better?

Or are those times also including transcoding time, not just comskip only processing time (to mark ads)?

Historically it transcodes first and then runs comskip. The reason we’ve got these new codecs and hardware accelerated builds of comskip is to avoid transcoding if possible and thereby make it faster and also reduce other issues which transcoding can incur in specific instance (esp when the recording quality if degraded). This also gives Comskip the best chance to work on the original video

These are great benchmarks. Can you confirm what Intel chipset you’re using for QSV? It’s being reported as a 13th Gen Raptor Lake chipset - which doesn’t sound right - according to the Intel specification Raptor Lake shouldn’t support AV1 encoding but your chipset clearly does.

Very interesting results from @techpro2004

Nvidia (cuvid) clearly out performances Intel (qsv) in decoding with MPEG2, H.264 and HEVC but Intel runs circles around Nvidia with AV1 decoding.

Also the AV1 libdav1d decodes is far superior in performance to the HEVC decoder

*QSV AV1 decoding was incomplete due to a driver hang

The qsv system is a 13th gen desktop, however it also has a pcie arc a310 installed. thanks.

1 Like

One thing about the nVidia 4000 series cards - the 4070Ti and above have 2 NVENC units, so the higher tier cards may have a significant boost to processing over the mid and lower tier cards. I don’t know if that translates into faster transcoding for a single AV1 job (both NVENC units are utilized) or it means you can run 2 NVENC encoding jobs at the same time (maybe even different codecs). There is only 1 NVDEC unit, however.

It may turn out, for transcoding purposes and MCEBuddy, that the higher tier cards with 2 NVENC units may not actually offer a benefit over the lower tier cards. However, it might benefit in a system that also runs Plex, for example, where MCEBuddy can be busy doing its thing with one NVENC unit and Plex can use the other to transcode on-the-fly for a player without tanking performance.

rtx 4080 logs below

cuvid 4080.zip (910.3 KB)

1 Like

Definitely a big improvement here: