[FFmpeg-user] RES: A question on the Quadro P6000 and maximum simultaneous NVENC encodes in FFmpeg

[Yang Zhang]

If you didn't specify -hwaccel option, then the full gpu transcoding
is disabled. for example:

1. full gpu transcoding.
$ ffmpeg -hwaccel cuvid -c:v h264_cuvid -i test.ts -c:v h264_nvenc ...

2. gpu transcoding via memory copy.
$ ffmpeg -c:v h264_cuvid -i test.ts -c:v h264_nvenc ...

3. cpu decoding and gpu encoding.
$ ffmpeg -i test.ts -c:v h264_nvenc ...



On 8/24/18, Dennis Mungai <dmngaie at gmail.com> wrote:
> Is it possible to disable hwaccel decode explicitly in an FFmpeg instance?
>
> Say, -hwaccel none such that no hardware-accelerated decoder instance is
> initialized on launch?
>
>
>
> On Fri, 24 Aug 2018 at 01:14, Yang Zhang <yzfedora at gmail.com> wrote:
>
>> Yes, 25 hd1080p channels use -hwaccel cuvid options to do full hardware
>> acceleration transcoding. other channels use the GPU encoding only. this
>> should can let you transcoding more channels. and fix the stutter problem.
>>
>> You can use "nvidia-smi dmon" to check the usage of NVDEC/ENC chips.
>>
>> On Fri, Aug 24, 2018 at 6:02 AM Dennis Mungai <dmngaie at gmail.com> wrote:
>>
>> > So, to clarify:
>> >
>> > If you were to disable hwaccel decode (and rely solely on software-based
>> > decoding) BUT utilize NVENC on a system with a beefy processor and
>> adequate
>> > RAM, you should be able to eliminate the bottleneck, correct?
>> >
>> > Because what you're implying (by the bottleneck at decode side) would be
>> a
>> > limitation in how FFmpeg handles filter performance (hwupload, to be
>> > exact), an issue that can be circumvented by eliminating hwaccel in
>> decode.
>> >
>> >
>> >
>> > On Fri, 24 Aug 2018 at 00:57, Yang Zhang <yzfedora at gmail.com> wrote:
>> >
>> > > The problem is even you using the Tesla P100, and you can see it has 3
>> > > NVENC chips inside it. but it has only 1 NVDEC chip, this is very very
>> > > important. the bottleneck is at decoding side. this is why you got
>> > freezing
>> > > or stutter problem. basically, perhaps you can use Tesla P100 encoding
>> 75
>> > > hd1080p channels, but you can not decoding more than 25 hd1080p
>> channels
>> > on
>> > > it.
>> > >
>> > > On Fri, Aug 24, 2018 at 5:51 AM Dennis Mungai <dmngaie at gmail.com>
>> wrote:
>> > >
>> > > > No.
>> > > >
>> > > > This is as tested on the Tesla P100.
>> > > >
>> > > > On Fri, 24 Aug 2018 at 00:26, Pedro Daniel Costa <
>> > > > portalnet2 at outlook.com.br>
>> > > > wrote:
>> > > >
>> > > > > Ok
>> > > > >
>> > > > > But with the TESLA models, did you have any issues over 40
>> > > > > streams?
>> > > > >
>> > > > > I am planning on budge project to run minimum 100channels
>> > > > >
>> > > > > -----Mensagem original-----
>> > > > > De: ffmpeg-user [mailto:ffmpeg-user-bounces at ffmpeg.org] Em nome de
>> > > > Dennis
>> > > > > Mungai
>> > > > > Enviada em: quinta-feira, 23 de agosto de 2018 13:54
>> > > > > Para: FFmpeg user questions
>> > > > > Assunto: [FFmpeg-user] A question on the Quadro P6000 and maximum
>> > > > > simultaneous NVENC encodes in FFmpeg
>> > > > >
>> > > > > Hello there,
>> > > > >
>> > > > > For users with this specific card (and FFmpeg installed), kindly
>> > > clarify
>> > > > > on the following:
>> > > > >
>> > > > > (a). How many simultaneous encoder sessions can you run on this
>> card?
>> > > > >
>> > > > > (b). Have you ran into any issues, such as stuttering and dropping
>> > > > streams
>> > > > > with multiple concurrent encodes?
>> > > > >
>> > > > > I ask because of a recent case where adding more than ~40
>> concurrent
>> > > > > encodes would make the output "drop" and stutter massively, as if
>> > > FFmpeg
>> > > > > itself was hanging, despite the GPU having more than enough VRAM
>> > > (24GB?)
>> > > > > and resources (under ~11% utilization according to nvidia-smi),
>> > > implying
>> > > > > some sort of artificial limitation in effect despite the GPU
>> support
>> > > > matrix
>> > > > > specifying otherwise:
>> > > > >
>> https://developer.nvidia.com/video-encode-decode-gpu-support-matrix
>> > > > >
>> > > > > Your feedback on this is appreciated.
>> > > > > _______________________________________________
>> > > > > ffmpeg-user mailing list
>> > > > > ffmpeg-user at ffmpeg.org
>> > > > > http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>> > > > >
>> > > > > To unsubscribe, visit link above, or email
>> > > > ffmpeg-user-request at ffmpeg.org
>> > > > > with subject "unsubscribe".
>> > > > > _______________________________________________
>> > > > > ffmpeg-user mailing list
>> > > > > ffmpeg-user at ffmpeg.org
>> > > > > http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>> > > > >
>> > > > > To unsubscribe, visit link above, or email
>> > > > > ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>> > > > _______________________________________________
>> > > > ffmpeg-user mailing list
>> > > > ffmpeg-user at ffmpeg.org
>> > > > http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>> > > >
>> > > > To unsubscribe, visit link above, or email
>> > > > ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>> > > _______________________________________________
>> > > ffmpeg-user mailing list
>> > > ffmpeg-user at ffmpeg.org
>> > > http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>> > >
>> > > To unsubscribe, visit link above, or email
>> > > ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>> > _______________________________________________
>> > ffmpeg-user mailing list
>> > ffmpeg-user at ffmpeg.org
>> > http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>> >
>> > To unsubscribe, visit link above, or email
>> > ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
>> _______________________________________________
>> ffmpeg-user mailing list
>> ffmpeg-user at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>>
>> To unsubscribe, visit link above, or email
>> ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".
> _______________________________________________
> ffmpeg-user mailing list
> ffmpeg-user at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-user
>
> To unsubscribe, visit link above, or email
> ffmpeg-user-request at ffmpeg.org with subject "unsubscribe".