Hi Folks,

We're in the process of designing some absurdly large recording workflows, and we're torn as to whether GTX 10 series GPUs are better for vMix than say Quadro M4000 cards.

We're considering on the GTX front, 1060/1070/1080 models, and the Quadro M4000 8GB models.

We are not yet certain if we want to go x264 or NVENC, as I am getting inconsistent information about how NVENC session limitations may impact us. And I also am unsure how we can offload the rendering of x264 at scale.

How much scale? Well, we're trying to design a system that could ingest 8-16x HD-SDI inputs, 1920x1080@60FPS, and have all of them not drop frames. Oh, and then do like 3 of these boxes maybe, and pipe them into each other for some external display flows.

Of course we're going to be throwing super beefy CPUs at them too, considering E5-2637 v4's (2x per system).

So, I'm torn as to what kind of direction we should be heading for GPU offloading.

Also, the resulting broadcasts would be to twitch.

Can we get some advice here please? nVidia specs seem to be insufficient for us to be able to make an informed decision here.

1. That discussion doesn't even talk about vMix, and I care about vMix performance.
2. I was also asking about 1060/1070 models, not just 1080.
3. The Z620 doesn't even come close to the scale of system we're working on. Sorry, not going to work here. ;)
4. Raw/synthetic benchmarks have no applicability with my _very specific_ performance inquiries about things like NVENC/DX/x264 encoding...

Sorry mate, while I know you're trying to help, this isn't what I'm looking for :)

Do you know how x264 offloading is handled with vMix?

Also, I'm a sys admin, I design, integrate and support systems. We're designing beefier than that ;)

What's the max NVENC sessions for these then? :
-Quadro M4000 8GB
-GTX 1060
-GTX 1070
-GTX 1080

Also, the reason I am looking into x264 offloading stuff is in the circumstance where not enough NVENC sessions are available.

3.NVENC LICENSING POLICY

There is no change in licensing policy in the current SDK in comparison to the earlier SDK (Video Codec SDK 6.0). The licensing policy is explained as follows:
The underlying software puts a limit of “two” concurrent encoding sessions on the combined number of encoding sessions executed on all non-qualified cards present on the system.

For example, on a system with one Quadro K4000 card and three GeForce cards, the application can run N simultaneous encode sessions on Quadro K4000 card (where N is defined by the encoder/memory/hardware limitations) and two sessions on all the three GeForce cards combined. Thus the limit on the number of simultaneous encode sessions for such a system is N + 2.
For the purposes of this discussion, non-qualified hardware is defined as any GeForce GPUs or low-end Quadro GPUs

Reference link:
https://developer.nvidia...m/nvidia-video-codec-sdk

Apparently there is no such restrictions on Intel QuickSync but I personally never tried to encode with QuickSync so...

Guillaume

Oh man, some solid info here folks! Thanks for all this! :D

Looks like the M4000 is probably where I should head :3

In regards to the SDK, does it actually need to be installed to use NVENC, or is that baked into vMix? I can't recall this very moment.

Also, those graphs, when it says @30, does that mean 30FPS? Can I expect about half if we encode at 60FPS? Our target is 1080p60.

Also, if I add another, say, M4000, making 2 of them in the system, does that logically "double" the number of concurrent NVENC streams possible? At that point I'm throwing hardware at a problem and it mostly scales linearly? Or is it something else? (Do I need to even "SLI" them, or is it all over PCIe bus?)