Slowdown problem on multiple screens (NVS-810)

Hello community NVIDIA

We are developing an 8x2 VideoWall (16 monitors with 1920x1080 resolution) using mosaic technology with the following resources. Our purpose is to show network video sources using the h264 and h264 codecs, spread it across the panel, be able to show it on a monitor with 1920x1080 resolution or use 4 monitors to obtain 4k resolution.

We currently have these resources for its development.

*** Nvidia Quadro NVS 810 (in total there are 2 graphics cards with 8 outputs each, they are cards with 2 integrated GPUs each)**
*** Driver version: 515.65.01**
*** OS: RedHat8**
*** Desktop: Xfce**
*** Window system X**

Well, the problem we have is that when we start to fill the video monitors, it starts to slow down a lot. We have verified that it is not a CPU load error or a bandwidth error, since it happens to us with both network video and local video. I think that the failure may come from some limitation of the NVS810 card that we are using.

I have read in an NVIDIA user guide that this type of graphics is intended for static displays or slow motion video.

DU-05620-001_v13 | April 5, 2017
Digital Signage is designed to be viewed from a distance and typically has slow moving video or static images. This type of application is ideal for NVIDIA’s NVS 810, Quadro K1200 or P1000 GPUs, which supports 4 displays per card. The NVS 810, Quadro K1200, or Quadro P1000 can be used in smaller workstations.

I have also read in different forums, problems very similar to ours using the same hardware.

I wanted to know if this type of behavior can be normal using the NVS810 graphics and if so, what type or model of graphics is best suited for our wallDisplay?

Greetings and thank you very much.

Hi,

There will be a definite limit to the number of video streams that can be decoded and displayed and likewise the NVS 810 is limited in decode performance relative to other GPUs.

A useful resource to estimate decode performance is our Video CODEC SDK webpage: NVIDIA VIDEO CODEC SDK | NVIDIA Developer

If you scroll down there are performance charts showing the decode performance in terms of # of HD streams. The chart uses our datacenter GPUs, however, this can be used to estimate the performance of equivalent desktop GPU and as an example the T4 would be similar to the RTX 4000.

For the kind of installation you describe we would probably recommend something like two RTX 4000 cards (or even higher) instead of the NVS 810 because the offer significantly more performance and can support decoding the required number of streams. The RTX 4000 also support Quadro Sync II which would allow you to synchronise all the display outputs like the NVS 810 offers.

I don’t know if you have system constraints in terms of available PCIe x16 slots since clearly two RTX 4000s would consume 2 PCIe x16 slots along with an additional slot for the Quadro Sync II card.

You also might be interested in a GTC session we created describing how you could “Build a Super Resolution Video Compositor”: How to Create a Super Resolution Compositor that Scales to 32 Displays | NVIDIA On-Demand

It has a Windows focus however the principles also apply on Linux, even if some of the API specifics are different.

Thanks,
Ian

1 Like

Hi,

Thank you very much for the response to our query.

We have decided to change our current NVS 810 for another one with higher performance,
The answer raises new questions.

Looking at the RTX 4000, I see that it only has 3 display port connectors
To support 16 monitors, can we use 2 of the same card?

Wouldn’t each miniDP output be connected to a single monitor?

Finally, I have seen that the consumption can be around 125W and we do not know if we would go a little fair. I wanted to tell you about the T1000 graph which we had studied for some time.
Do you think that card has enough resources for the use we are looking for?

thank you very much again.

Miguel Ángel.