Nvidia Boosts Hardware Video Encoding Performance For Some Consumer GPU's

[ThePointblank]

Tom's Hardware recently noticed that in a recent update to an Nvidia support page, a number of consumer GPU's have had their video encoding performance improved, from only being able to encode 3 video streams to now 5 concurrently:

 

https://www.tomshardware.com/news/nvidia-increases-concurrent-nvenc-sessions-on-consumer-gpus

 

Quotes

Quote

 

Nvidia has quietly removed some of the concurrent video encoding limitations from its consumer graphics processing units, so they can now encode up to five simultaneous streams. The move may simplify the life of video enthusiasts, but Nvidia's data center grade and professional GPUs will continue to have an edge over consumer products as now Nvidia does not restrict the number of concurrent sessions on them. Obviously, the speed of encoding can suffer with more simultaneous encodes.

 

Nvidia has increased the number of concurrent NVENC encodes on consumer GPUs from three to five, according to the company's own Video Encode and Decode GPU Support Matrix (opens in new tab). This is effective for dozens of products based on the Maxwell 2nd Gen, Pascal, Turing, Ampere, and Ada Lovelace microarchitectures (except for some MX-series products) and released in the last eight years or so.

 

There is no change that affects the number of NVENC and NVDEC hardware units activated in Nvidia's consumer GPU's; the capability was always in the silicon, but was disabled in software, while Nvidia's workstation and data center focused GPU's technically could support 11–17 concurrent NVENC sessions depending on the quality and hardware.

 

It appears Nvidia has changed their stance and reduced the cripple hammer on hardware video encoding, and this should be positive news for people with a lower budget and want a video editing workstation or those working on a laptop, as depending on their needs, may no longer need a workstation GPU to multiple concurrent encodes.

[emosun]

is this a streaming thing or rendering thing

[GuiltySpark_]

The Encode/Decode support matrix link is a little wonky so this will work: https://developer.nvidia.com/video-encode-and-decode-gpu-support-matrix-new 

[SquintyG33Rs]

5 minutes ago, emosun said:

is this a streaming thing or rendering thing

both, either... just a question if the software stack uses nvenc acceleration.
Streaming does in almost every case. video editors are more varied and more annoying with only being able to accelerate certain parts.

[Levent]

Ooh, nice. GTX970 did get a boost.

 

I have actually hit this limit in Jellyfin couple of times. Still, having a limit is dumb has hell but ill take what I can get.

[Antrikshy]

For the lazy, here's the full list of devices (under "NVENC - Encoding").

 

I think the change is "Max # of concurrent sessions" getting bumped from 3 to 5. Models as old as the GTX 745 are on there.

[LAwLz]

10 hours ago, emosun said:

is this a streaming thing or rendering thing

I'd say it's neither. 

 

What they did was increase the maximum number of video streams that can be encoded from 3 to 5 (possibly with a hit to performance if all streams are used). 

 

When rendering a video in for example Adobe Premier, you're only encoding one stream.

When you're streaming a game, you are usually only encoding one stream as well. 

 

I see this more benefitting people who maybe rip DVD/Blurays and want to encode 5 movies at a time. 

I doubt this change benefits more than a very, very small subset of users. It's still nice of them to remove the restriction, although I don't think they should restrict it to begin with. I don't see much point in it. 

[leadeater]

3 hours ago, LAwLz said:

I doubt this change benefits more than a very, very small subset of users. It's still nice of them to remove the restriction, although I don't think they should restrict it to begin with. I don't see much point in it. 

There was only two thing I could think of, first being broadcasting and re-encoding cameras or video footage but I don't think many will be using the affected GPUs here. The other was security camera servers but checking the documentation of Milestone that only uses the decode path for motion detection so isn't applicable. Few things are springing to mind that could involve multiple encoding streams.

[leadeater]

@LAwLzOh actually I forgot an obvious one, Plex transcoding. That can use multiple encoding streams.

[The 8-Bit Time Traveller]

Never understood why NVIDIA chose to shackle their own GPU's like this. It gave Intel an opportunity to sneak in with incredible quicksync hardware. My plex server runs on a  i7-10710U intel NUC with a bunch of VM's and docker containers. I've seen it peak at 9 1080p HVEC transcodes without breaking a sweat, some people have tested them at over 20 concurrent transcodes.

[Fasterthannothing]

I've been running unlocked limiters on consumer Nvidia GPUs for years. 

https://github.com/keylase/nvidia-patch

 

Plex even bundles the patch with their Linux software now 

[Kisai]

On 3/24/2023 at 4:26 PM, emosun said:

is this a streaming thing or rendering thing

It's more of a "they increased the limit to 5, regardless of the device", where as The quadro's are either unlimited, or 3 if they were weaker devices (x50/x60 part equivalents.) Previously they were all 3 or 0 (chips with no encoder) and only the quadro's were unlimited. And previous to that they were all 2 or 0.

 

Realistically there's no reason to limit the decoding. The encoding however should be restricted when the resources are likely to conflict. So under normal circumstances, you'd likely never hit this, it's something you'd run into in video editing or using the GPU as an encoder for plex or security cameras. 

 

So my guess the reason why is because of Intel, cause the iGPU's alone can do 14-24 h264 encode streams without breaking a sweat. 

 

 

But there's a bit of dishonesty there. The number of streams may only be representative of the front-end of the encoder. With Quicksync, it's pretty clear that it consumes CPU resources when you give it settings that the "hardware" encoder doesn't like. And that is the same with Nvidia (consuming CUDA resources mainly.)

 

So technically, there has never been a hard hardware limit, it's always been a software limit, when you hit that hardware limit on nvidia, you get penalized in speed when it's uncapped. The software limitation has been more of a bane for people who are streaming (Eg twitch and youtube) since they may be in situations where the decoder is always invoked (eg discord/zoom) and then uploading to multiple points.

 

That said, I doubt anyone will notice except for twitch/youtube streamers, since many/most do simultaneous stream+record, and thus when they try to collab with someone, eg via skype/discord/zoom the encoder for that has had to resort to software.

 

However it's always been a thing that the "encoder overloaded" in OBS happens if the GPU is already pretty maxed out by the game being streamed.

[Quackers101]

for new ones too? didnt they promise 5 with the current ones and it was limited?