Voices of Video

So You Want A VPU? Here’s The No-Drama Way To Plug, Play, And Push To Your CDN

• NETINT Technologies • Season 3 • Episode 33

Share episode

Tired of choosing between ripping out your video stack or standing still? In this episode, Kenneth Robinson, Director of Field Application Engineers at NETINT, walks through a practical playbook for deploying Video Processing Units (VPUs) at any stage of growth - from retrofitting a live server to scaling across edge, cloud, and hybrid environments.

We start with hardware options that match real-world use cases:
 • T1U hot-swap modules that slide into existing servers - no downtime required.
• Prebuilt 1RU systems with up to ten T1Us and your choice of ARM or x86 CPUs.
• Compact mini servers built for SDI ingest and edge workflows.
• Cloud-based VPU instances via partners like Akamai, CDN77, and i3D.net - including test credits for quick starts.
• Hybrid configurations that keep steady-state on-prem and burst to the cloud for overflow or redundancy.

Then we cover software integration. Choose your path: a native API for granular control, or use FFmpeg and GStreamer for faster deployment. Kenneth explains how NETINT’s SDK is being upstreamed into both projects - simplifying maintenance and keeping features current without custom patches.

Next, we dive into two advanced capabilities that redefine efficiency:
 • Multi-layer AV1 encoding for personalized overlays or targeted ads inside a single bitstream.
• Multiview encoding that lets the player dynamically stitch camera feeds without re-encoding - perfect for multi-angle sports or live events.

Finally, not every team has a full dev bench - so meet Bitstreams, NETINT’s no-code interface for managing transcoding workflows. Build templates, monitor load and health, convert captions to WebVTT, and push to multiple origins with RTMP, SRT, HLS, or DASH.

Kenneth closes with a preview of the customer-driven roadmap: WHIP/WHEP contribution, RTP, SMPTE 2110, audio-level control, RIST, and NDI, all prioritized based on real-world feedback.

If you’re exploring AV1, chasing lower latency, or planning hybrid expansion, this walkthrough gives you concrete choices and clear next steps - from card to cloud.

Download the presentation: https://info.netint.com/hubfs/downloads/VPU-Deployment-Options.pdf

Stay tuned for more in-depth insights on video technology, trends, and practical applications. Subscribe to Voices of Video: Inside the Tech for exclusive, hands-on knowledge from the experts. For more resources, visit Voices of Video.

Voices of Video:

Voices of video. Voices of video. The voices of video. Voices of video.

Kenneth Robinson, NETINT:

My name is Kenneth Robinson. I am the director of field application engineers at NetEnt. So what I do and my team does is help customers deploy VPUs into the market. So the question is, how do you deploy a VPU? There's a couple different options. The first option is to upgrade your current servers. So we have a couple different form factors that allow you to take existing servers or repurpose old servers and utilize the NetEnd VPU in them. So a most common form factor is the uh T1U. It's a U.2 form factor. It's the size of a hard drive, it's hot swappable. So when you upgrade an existing server, all you have to do is just plug it in. You don't have to uninstall the server or power it off. You can do this when the server is running. Now the second option is if you don't have any service upgrade, is to buy a server that comes pre-populated with our quadra T1U cards. So we have a one RU form-pacter server that comes with up to 10 T1Us pre-installed. And you have a choice of an ARM CPU or an x86, either Intel or AMD. Now we also know that uh high density decoding, encoding isn't always required for every single application. So that's when we developed the Quadro mini server. So this is designed for smaller channel counts, more on the edge. It comes with our smaller T1U installed, so it has a smaller uh capacity count. And then it also comes pre-installed with a Black Magic SDI capture card. So you can use this to capture SDI sources in different studios and uh create all your ladders and send it straight to your CDN without having to backhaul it back to your uh main location. And then the the other option, the third option, is a cloud deployment. So these are great for people who don't have their own servers, don't want to manage their own servers, they want to just purchase or rent um our VPUs that are in the cloud. And we're currently deployed with uh Akamai, CDN77, and i3D.net. And with Akamai, you can actually log in now, and we have a promotion code that will allow you to get credits and be able to test the quadrant VPU very easily. Um so if you need that code, just come see us afterwards and we will get you set up with that. Now there's actually a bonus one that I never mentioned. I said there's three, so you get an extra one for free. And that one is a hybrid deployment. This is great for on-prem um and then being able to scale to the cloud when you need the extra capacity. Or you could use the cloud for like a redundancy or backup option if one of your servers in your lab goes down, not your lab, in your production environment goes down, you can just run something on the cloud just temporarily. So now that you have the hardware, how do I use it? How do I run it? Well, you need some software to run it. So we have three different ways that you can integrate with our cards. The first way is using our own API. Um, using it and that requires a little bit more work, a little bit more knowledge. So the other options are using FFmpeg or GStreamer. Now we support a couple different versions of those ones, and we're always working to maintain the latest releases as as they come out. And the other thing we're actually working on, we're very excited about, is that we're actually upstreaming our SDK fully into both FFMpeg and GStreamer. Um and we're working on that now, and it's going to roll out in pieces just because there's a lot of code and the open source community just you know doing a full push like that. It's just a lot of work from them. So we're just doing it in the incremental steps. And we're also going to include all the features that we've developed and all the bug fixes that we've done for those different um what they call it, FFMpeg and Gstreamers. So everything we have will be pushed to the cloud. Now we've developed a couple different features within our SDK that I just kind of want to touch on in the last year. So the first one is multi-layer AV1 encoding. So this allows you to have your main video and then overlay different uh graphics or ads, and then on the player side, they can choose which layers they wanted to see. So you have one bit streams with multiple layers, and then uh one device could have one ad, and another device could have a completely different ad. The other one that we also developed was uh a multi-view encoding. So this is really cool because the boundary picture pixels of the video is actually all iframes, which means then you could stitch together the video on the player side without any bleed over and allow uh the the end user to control which uh video sources he hit he or she would like to see. And and this is also really good, especially if you have say an event with a hundred cameras and you want to be able to show a two by two matrix of all possible combinations. That's like 8,000 different combinations. So this allows you to just encode the hundred channels and then the end device will control what they want to see. So now that we've covered like the SDK package, not everyone has a development team or the resources to develop their own app from scratch. So we develop what we call bitstreams. This is uh a user interface, it's code free, it's point and click, allows you to set everything up that you want. It has many different features I've listed there, um like VOD recording, it's based on templates. We can convert uh subtitles into WebTT for like HLS. Um, and then we also support pushing to different origin servers. Now, some of the specifications that we're currently working on from an input standpoint, like RTMP, SRT, dash HLS, uh the list goes on. We support all three different codecs that the Quadra supports, uh 264, 265, AV1, and then on the output side, SRT, multicasts, pushing to different origin servers. It's all there. We continually work on this. Which also leads to is we're working on it, what's coming up. So currently in 2025, by the end of the year, we're hoping and scheduled to get uh WIP support, um RTP, audio level control, some new different codecs, and also SEMPT2110 is also something we're working on. And then in the new year, we'll work on me working on, you know, maybe like RIS, NDI, Azixes, um, more statistics, DOM. So this list is actually kind of, you know, it's it's in the air because we really work with our customers to develop the features that they want. We work closely with them to help um define what the feature is, prioritize the feature, get it out to them, get them tested. So, you know, nothing set in stone. If there's a feature you're missing, we're more than willing to consider it and look at it and um potentially schedule it into a feature release. So just a kind of a quick overview of what BitStrings is and what it looks like. You know, we have a dashboard which gives you the different metrics, shows you what's running, it shows you current error status, current load, um, what's been installed. And then we have what we call an encoding page. This is just uh a table view of everything that is configured. It could be running, it could be it could be in an error state. Um, and then we also have a more detailed view where it has more historical logs of what's going on. And then we have at the template, and this is the heart of bitstreams. This is this is what everything is surrounded around. It allows you to create uh what resolutions you want, what configurations, what encoding settings, what parameters you want. It's all done on the template, and that template gets applied to one stream or as many streams as you see fit. So that's it. If you have any questions, please see myself or Craig here at the booth.

Voices of Video:

If you'd like to find out more, please visit us at netint.com.

Head Shot

Mark Donnigan

Host
Head Shot

Per Nybom

Host
Head Shot

Jan Ozer

Co-host
Head Shot

Anita Fejter

Producer
Head Shot

Kenneth Robinson

Guest