NVIDIA Silently Increases GeForce NVENC Concurrent Sessions Limit to 3

[btarunr]

NVIDIA has reportedly increased the concurrent sessions limit of its NVENC hardware video encoder on GeForce graphics cards to 3, up from 2. This means up to three different apps could use NVENC simultaneously, or an app (such as Premiere Pro) could use up to three sessions of NVENC for faster live previews during video editing. NVIDIA's Quadro family graphics cards can have practically unlimited NVENC concurrent sessions. The company recently updated the NVENC support matrix page showing a "Max # of concurrent sessions" increase from 2 to 3. The first screenshot below shows the updated page, and the second one shows a Web Archive snapshot from 2 weeks ago. NVIDIA last updated its GeForce drivers late May with 446.14 WHQL, so you might want to update your drivers.



View at TechPowerUp Main Site

 

[AsRock]

Excuse my ignorance, so nVidia has got some artificial software\firmware code blocking the ability to have more ?, or has it all improved some release ?.
Or are they getting a little twitchy with RDNA2 on it's way ?.

 

[Verpal]

Excuse my ignorance, so nVidia has got some artificial software\firmware code blocking the ability to have more ?, or has it all improved some release ?.
Or are they getting a little twitchy with RDNA2 on it's way ?.

It is artificial limit, no evidence decision is influenced by RDNA2, but we can expect Nvidia to provide more for its customer when AMD up their game.

 

[Solaris17]

no evidence decision is influenced by RDNA2

Not given that these limits have been imposed since NVENC was a thing.

Nvidia NVENC - Wikipedia

en.wikipedia.org

 

[zhongfu]

Excuse my ignorance, so nVidia has got some artificial software\firmware code blocking the ability to have more ?, or has it all improved some release ?.
Or are they getting a little twitchy with RDNA2 on it's way ?.

You're right, it's an artificial limit in the drivers, and it's possible to patch it out

 

[Kokorniokos]

Will this be in the studio or gaming drivers? Or both?

 

[ZoneDymo]

I dont really understand this, what is the point of being able to run it multiple times?
Like is it not better to encode something using the full available ermm power? instead of spreading it out?

orr is there some cap anyway, that it already sis working as hard as it can on that one task due to how the software works, sooo might as well have multiple instances?

like when using it for livestreaming, the more it can encode, the less bandwith would be needed so I would think having the card dedicated to doing all it can for that 1 task is better then making it do multiple tasks lazily

 

[kayjay010101]

I dont really understand this, what is hte point of being able ot run it multiple times?
Like is it not better to encode something using the full available ermm power? instead of spreading it out?

orr is there some cap anyway, that it already us working as hard as it can on that one task due to how the software works so might as well have multiple instances?

like when using it for livestreaming, the more it can encode, the less bandwith would be needed so I would think having the card dedicated to doing all it can for that 1 task is better then making it do multiple tasks lazily

If the transcode uses less than the NVENC chip can provide, like a single Plex stream, then you can transcode multiple streams at the same time. It's not that useful for livestreaming, but for Plex trancode streaming on a server a single card could easily deliver upto 8 streams on the GPU, which could be useful if you run something like UnRAID where multiple people are watching shows/movies at the same time without having to bother the CPU.

That being said, users have been able to unlock the NVENC limit for a long time now, and have 8 simultaneous streams on servers providing Plex streams.

 

[Roph]

Excuse my ignorance, so nVidia has got some artificial software\firmware code blocking the ability to have more ?, or has it all improved some release ?.
Or are they getting a little twitchy with RDNA2 on it's way ?.

Hard to imagine Nvidia being twitchy about VCE, its quality is absolute garbage compared to NVENC, especially for H.264. The latest VCE (which they pointlessly renamed to VCN) has even worse H.264 quality than their 3rd gen VCE (Tonga, 285 / 380). Even that is still worse than first-gen kepler H.264 NVENC.

One of their multimedia engineer team members was somewhat active in github issue comments in the past, the gist I got is that AMD simply won't allow a larger transistor budget on their GPUs for media encoding, so they have to make do. They view media encoding more like a checkbox affair, instead of any focus on actually improving output quality.

 

[Octavean]

Too bad prices are so inflated now.

I was looking to buy a Quadro Pro P400 for Plex transcoding on a server because it was relatively cost effective at about ~$70 USD, doesn't require power beyond what is provided via the PCIe slot and is single slot / single height. Unfortunately now its typical for a new Quadro Pro P400 to cost north of ~$115 USD.

 

[DaveBetTech]

I think this means Plex can now transcode three streams instead of two. I hope that is what this means. This would be awesome.