Voices of Video

Trust, Footprint, and Milliseconds: The Real Levers of Live Streaming

• NETINT Technologies • Season 3 • Episode 37

Share episode

What if your live stream could feel instant, look sharper at the same bitrate, and cost less to run as you scale?

In this episode of Voices of Video, we unpack how a fast-growing live gaming platform rethought its architecture, moved beyond a fully public cloud setup, and built a global hybrid model that delivered measurable results: lower egress spend, higher transcoding density, and a clear lead in end-to-end latency.

We start with a hard truth many streaming teams face too late - public cloud convenience gets expensive at scale. Our guest walks through the investigation that led to choosing a single global partner with true backbone control instead of stitching together regional providers. From there, the conversation gets tactical: identifying egress and transcoding as the biggest cost levers, shifting transcodes to dedicated VPUs on bare metal, and simplifying operations compared to GPU-heavy deployments.

The surprise? Hardware encoding matched - and in some low-latency profiles beat - software quality at the same bitrate. Higher density translated directly into dollars saved, without sacrificing visual quality.

Then we focus on what viewers actually feel. By running WebRTC over a backbone that carries traffic as far as possible before handoff, the platform shaved seconds off delivery compared to competitors - an advantage that only becomes obvious when a goal scores or a clutch play lands. We also dig into visibility: turning opaque networks into transparent systems with traffic-path insight, ACL hits, and attack telemetry. That visibility enabled faster response, fewer surprises, and stronger trust with users.

Finally, we explore the product layer. Predictive, card-based overlays only work when streams are tightly synchronized. Better quality and tighter sync led to longer watch times and stronger monetization - proof that infrastructure decisions directly shape business outcomes.

If you care about building low-latency, high-quality live video without lighting money on fire, this conversation is a practical playbook.

Topics Covered:

• why teams move away from a full public cloud model
 • selecting a global partner with backbone control
 • real cost drivers: egress and transcoding
 • moving transcodes to VPUs on bare metal
 • WebRTC for ultra-low-latency delivery
 • opening the network “black box” with traffic visibility
 • predictive overlays that demand tight sync
 • trials, fast support, and iterative validation
 • improving quality at the same bitrate
 • simplified operations with consolidated vendor management

Links & Resources

NETINT Technologies
👉 https://netint.com

NETINT Video Processing Units (VPUs)
👉 https://netint.com/products

Voices of Video – Full Podcast Library
👉 https://netint.biz/podcast

This episode of Voices of Video is brought to you by NETINT Technologies, delivering purpose-built video encoding solutions designed for scale, efficiency, and real-world streaming workloads.

If this episode was useful, subscribe, share it with a teammate fighting egress bills, and leave a review to tell us what you want us to unpack next.

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. Voices of video.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Yes, this is working. Hello everyone. Welcome. Welcome here at our uh booth share with NETINT Technologies and um during the next twenty to thirty minutes, depending how deep dive you go on the technical bits, we're gonna um tell you the whole journey about the from from going from a purely cloud-based platform to a hybrid-based platform and all the disadvantages, advantages, hardware, VPUs. We're gonna dive into all of that. So I hope you guys uh uh will enjoy this and maybe can learn a little bit from it. So um yeah. Starting with the introductions, I'll first introduce quickly myself. I'm Stefan Ideler, co-founder of i3D.net, I've been with the company for more than 21 years, and still very much a true technical geek on many of this, and that's why it was such a pleasure to work with Sebastian, who will introduce a little bit about himself.

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

Hi, and thanks Stefan. Hi everyone, I'm Sebastian, I'm uh the head of media infrastructure at Noice. I joined in uh January 2014, so nearly two years ago. Um, and uh my background was video, video compression, video delivery before I feel working at Twitter, working at the video on the man's side of their business. Um I know a lot about video, and that's also annoying uh speaking to Jordan and build out their video pass from the video series. Um that's some easy about me. I think get to know me a little bit more the next 20 minutes. Yes.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

So would you mind telling us a little bit about um how you started engaging with us at the first time? Uh how did the process go? Or and also maybe about the reasoning why you were looking to go from uh yeah, from a full cloud scenario to uh what uh what you ended up using.

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

Yes. So when I joined noise, everything was running in Google Cloud or in the public cloud environment. Um and that was because it's just easy for a startup to start developing in a big cloud, and it's just simple, everything, everything worked, then voice now launched, and suddenly a lot of data traffic uh started to happen. And but then typically it tends to be the case is that things start to become very costly, very expensive. And so uh when I joined noise, uh one of the few things I had to look into is like how can you reduce the total cloud? Um, and then the most obvious one was the amount of data. Um at a startup, you always try to achieve a holistic curve, uh growth. So we know that once noise started to pick up, it's gonna be all exponentially more extensive. Um very early on, after a few weeks already, when I want to join noise, we've started to discover the industry. What are the different opportunities? Um, and then out of that investigation, a couple of providers um were very interesting, very promising. What is important for a company as noise is that you've got a global footprint. Yeah, so it might be a very nice partner in some obscure country, but if they can't deliver very stable connection to the rest of the world, then the thermular fortune of them. And so me, we didn't want to go with like five, six partners at one slow. We want to go with one market because we're a small company, we don't want to spend a lot of it, a lot of effort in managing all of those relationships. We just want one single point of contact, and if there is a problem, they've got to fix it solution. Very, very bluntly said. And so out of that investigation, we found uh ITV came out on uh on top uh for us for a number of reasons, um which I love to delve into by the way. Um one of the reasons obviously from our perspective, the price point was very, very good. The more importantly is the uh human touch. So the moment you start talking with simple at different companies, if they understand your problems, if they think a long way to use them, if they start bringing you different solutions, not necessarily from a seller perspective, but um also from the idea of like, hey, if is this not a problem for you?

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Exactly.

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

Then you're like, oh yes, actually that is a problem, but not right now. Stephen holds your horses, but like we we can we we got it, it's it's why we it gives a lot of trust in within your partner that they think along, but also that they have solutions for you when those problems are gonna arise because you don't want to do the same exercise for a different part of your staff, uh for a smaller piece of the whole pie of the infrastructure to find another vendor and that yeah, and so that was one of the the main important reasons for us that uh we decided to uh uh to go together with the IT PBI backpunk.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Thank you so much for for all those those compliments. And so I totally agree. I I truly believe what makes a difference with with engaging with our our our clients and partners is to more sync in as a partnership. We are on the in the Slack channels together with all of my technicians, your technicians. We just make the brainstorm happening when we see uh a technical challenge. Um some of them we we might have already solved, and then we can suggest to them some ideas to you, like like we did. Um, but we also uh uh did some new things that you together. Um uh of course, one of the reasons why we're here uh at this booth. Would you mind sharing a little bit about that?

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

Yeah, that's right. One of the things you brought up, and and from my background, I was super interested in uh one of the biggest polar costs for for start to be in GPS CV is the transcoding as well. Yeah. Um and so although our initial idea was to move our streaming spec out of the public cloud for the reduction in the bit rate, because that's where your exponential cost uh structure comes from, or the cost factor comes from. Um I already was thinking about the transcoding as well. Because the transcoding that still leaves me public cloud, you still have data going back and forth. Back and forth. Sure, that data is only one expert creator, not per viewer, but still that's expensive in the transporting cost in itself is also extensive. So it's a double warming if you can move that into into the uh with that metal as well. So on the first, the second meeting, you guys already said, like, oh, what are we gonna do about transporting? Interesting as can you offer because I've got a few IDs myself, which I really would love to explain if they're working because they're very effective, they're very efficient, and from use from your point of view, they're gonna have like limited footprint in the data center. Yeah, so and so I think and that's single meeting. We already laid out the the the blueprint for like 18 months of collaboration or so, 12 months of collaboration or so. Um like I needed no instead of like let's do the data egress first, and then secondly, let's be interfaced coding. That's basically how we all started to get to work together with our strength of I2P and NetB. And I said, Thank you.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

For us, it was also a really nice case to to also firstly believe now, especially in in partnerships. Net is also one of our partnerships, they're very happy about. And for the engineers in the data center, it was just yeah, you open up a server, plug in, plug in the net in cards. I mean, don't want to speak badly about it, but it's really easy to to to install. You don't need to reinstall the machine we provision, now we close it up, and you're good to go. So for us, it also proved to be a very easy uh solution to use in in our data centers uh worldwide. Uh because of uh the traditional GPUs, it always becomes a mess with cabling, power cabling, and the big chassis fitting five of them in a box. Uh with GPUs, it's really it's a lot more simpler for operational reasons as well. Um at I3D, of course, we originate and are very strong within the gaming industry. And uh, aside from the human touch and and being everywhere globally with uh the compute, what's also very important for us is to make sure that our network connectivity to all of the ISPs, to all of the users, is as much under our own control as as possible. So we try to carry the bits and bytes as far as we can ourselves before handing it off to the ISP. Um, did that have any positive impact or experience for for your point of view, for noise, maybe?

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

So that very much is one of the most important aspects what we were looking for in Athens on the investigation. So global footprint, global network. Um, and then we needed to, before we could test anything, we need to have a decent amount of trust that you could deliver it because we were a live streaming platform, we were using not HLS like many other people, we were using WebRTC, which comes in its own host of problems, but also interesting opportunities because we get really ultra-low weight if you do it well, but on the network it needs to be there as well. And we we found that with IXP there was no problem. Like the moment we almost literally flipped the switch and months we started to have traffic through I3D, we saw that the metrics were either the same or better, and more likely better than the same for most of them. Yes, and so we also were comparing, of course, our own politics with competitor products, and we noticed that consistently there were three, four seconds faster than them in there in the delivery because of the RepRTC protocol, which is inherently out of the protocol, but also the stability and that you can only do if you've got a partner that has a metric. So obviously, I don't think that you know what goes into the network, and that's not my business. Then I need to trust that my partner. And I think that very much was a point like we saw in our method that it was good, said there's no business for us to to to uh go and investigate that well, your problems are because that's your business.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Exactly. Now that that is exactly right, and and for many companies, the networking part of their providers is a gigantic black box. They they you just have to trust it that it works. And then yeah, if it doesn't work, what's important is that you get quick interaction with the team to figure out what it is. Um one thing we learned from this also is that we now also uh created a new feature on our end which gives the client complete insight of all the traffic that is happening on his or her servers. You can see it's precisely from where all the countries called from, if it travels our backbone from and then which locations it traveled to backbone from, um uh uh if it hit any ACLs, if it hit any attacks, and we believe offering that visibility so you don't even have to create a ticket or do the slack conversation anymore for those topics, is is is trying to sync along on opening those black boxes and exposing uh that valuable information. I'm not really sure that um Hebrew you would like to tell a little bit about what Noise actually did, because I don't think we actually talked about this during this past fifteen minutes.

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

So uh Nois is a live streaming platform very similar to Twitch and all platforms. Um and where we try to differentiate is by offering an engaging on top of the live game stream, engaging for the viewers of the live stream. So the game itself is a card game, it's a predictive card game. Um I'm not a game designer, but a lot of mechanisms go into the card games. It's not as simple as the tunes. That's not all there is uh partly of noise. Um so if there's 50 sort of events that you can detect, which is eventually not noise five to me, they detect events in games, in game stream. If you can detect 50 events and you can only select three or four different things that can happen by random, and if you guess correct, then you get points. You suddenly got like a very interesting game dynamic if you want to guess the right thing. You want to guess them very quickly so you get more points. You are frustrated because we cannot guess for the right thing, so you're gonna set 10 coins or points to get a resource of all of these kind of game dynamics tours into it, which led to a very engaging and very entertaining way of consuming uh game streams, which then if you look into the metrics and if you look into the engagement, people watch longer, if it's more engagement, spend more money, all of this kind of things that you would like to have as a business uh are suddenly there. But the only way you can do that is about having the game or a video stream that's really stable and smooth as a high quality in the first place. So if you've got stuttering, you've got buffer rate, if you've got low frame rate, but the audio and video are not synced, that's a poor scheme to start with. So your table sticks needs to be there, and everything on top of that makes it that it's better than a competitor. And so that's why for us NetLabs Work processing was super important because that is adding to the edge of noise, because if you've got a live video game stream with live predictions in a card event, and something happens, somebody scores a goal or shoots somebody else in the head, and the card commonly scores five or ten seconds later, or even worse, the card scores. With the game is like in five seconds because it's brutally in a throughout or something along those lines, then suddenly you know it's gonna happen, or no, it's not gonna happen because the the card scores, but that's a very trap experience. From a user perspective, you don't really want to have a high so everything needed to work, everything needed to be signaled. And so that makes it a very interesting technical challenge. Um, but once it's nailed, it's a very interesting experience for the user for the it's users. So our partners there needs to also open a team to make sure that they can support that they can.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Thank you for for the additional context. What really struck to me is ultimately what was really important for you was also that end user experience. If if the end user is happy, they engage more, they will eventually monetize more on the streams, and it's ex precisely the same from our origins as well. If the gamer is happy, the game session doesn't crash, the game doesn't have packet loss, the connection is optimal, the the the the match feels very smooth, there's no cheaters, no DDoS, then those gamers are way more likely to buy the next DLC, buy the next skin, do the next micro transaction. Uh in the end, now I think about it, it's all about keeping users engaged. And it's exactly it's not just here in the games industry, it's also in in the on the live streaming industry. And I think when I look here at at this this this show, I think it all matters in many, many other verticals. Um so yeah, this is why we as I C D um we and we we are big in gaming, but that's why we also are here at IBC to explore all these other verticals to see if we can make a difference by offering amazing end user experience. Again, Sebastian, I want to thank you very much for uh being here and sharing your story. Are there any last words you want to say or things that you might have missed during the conversation?

Sebastiaan Van Leuven, Head of Media Infrastructure at Noice:

Yeah, maybe I want to be a bit more exholistic about certain parts of how we work together because I think um a lot of it is said about trust. Correct. And like we had to trust you because we have to look into your statistics and stuff like that. But at the same time, one one part that instilled at first was the idea that we could get a box, a demo machine, and we could start testing. And no questions asked, yeah, sort of, right? I mean with a reasonable, um and that allowed us to actually do that initial testing, it's like hey, these guys, their network is as good as they intend it to be. So there is no reason to doubt that that network is coming being performing on words at scale, and so that allowed us to then focus on what really mattered is like how can you now solve the problematic part for residual engineering part? How can we get a cloud infrastructure that needs to keep our willing, that's quite complicated already, extract the video streaming stack there, integrate that in the same cloud environment, but let us run on a different cloud on a metal project. And then once you can test that with a single or two devices, um, and there is no down version, and you don't have to, you're not taking up the bills for the signal test. And you got a question, and the guys respond to you insight within like five minutes, and you got a more difficult question because guys are then let us get back to you tomorrow. You need to ask a few people here, then the sort of support level that you get as a starter is like priceless, yeah. Like that you only get like a very high enterprise grade support level on the big bulk, yeah. And so that really helped us move forward, it really helps us move quicker. And it also goes into relatively simple questions for us that we have no idea about how exactly they were both studied all in networking engineering and these kinds of things, but how exactly does it work in the wild because it potentially has an impact on how the things are around us from our perspective. Yeah, understanding that allows us to have better in monitoring better visibility into what we're not, and already identify what potentially could be problems or developing problems, even though there was a thing in the question about metric protocols that was a problem for us. Still, there was like a very visible answer coming back in a very, very quick uh amount of time. And so that's really helped. I think in the next phase of the collaboration, once you fly the streaming steps, once I ether stuff going over IP the it was just another box switch. In this case, there was threatened transcodes, and we could just test them. And that was really helpful because now I could just test how many loads can I push through a machine like that? When is the machine gonna start to struggle? What are the policy limitations? What is the quality of the input? And nothing in itself allowed me to take some time to do that investigation and say, Yes, this actually gives me an edge compared to software transport. Yeah, um, this makes sense, and then with from my point of view, it comes down to the amount of hours transcoded per dollar on a on a month, right? And if you can open me a machine with a much higher density than CPUs on a price that is like in ratio much more interesting. Should I use it for simple operation possible? Because you can you can sell more for the same square meter, exactly. And and I don't have to pay a switch per hour transport and at the same time, and that's a win. That's something that I didn't know at front. And I was saying I was also skeptical about that. And that's why I wanted the investigation, the quality of the encode for uh low-latency encode for nettings was better for the same bit rate than I had the software lens code. So it's not that we compromised on the quality, we didn't compromise all the amount of data that you were sending. It was a wing win on our perspective for us. And that we couldn't have in any big cloud provider asking them to open a right engine. And and and and put in a few nothing machines there so that we can have something and then put it in confliction. Not only is possible with only the parting and the bio, that was a piece of IT. And so eventually also launched that into our in cohort platform. So at the end of the day, we are running our web part in in multiple locations too. I3.net transcoding was happening in I3D.net, so there was no reason for us to actually get the data back into the public. Enough for storage purposes. Exactly. So we did have all the firmovers implemented and we didn't need them. So that's also a waste of time. But of course it's time to say you also have this. But it just shows how well these things actually work and that eventually the the the quality that we got was far exceeding much to get us.

Stefan Ideler, Co-Founder and Chief Innovation Officer at i3D.net:

Thank you so much for those those those final final words. Either now or after after this talk. Good. Thank you.

Voices of Video:

This episode of Voices of Video is brought to you by NetInt Technologies. If you are looking for cutting-edge video encoding solutions, check out Netinth's products at netinth.com.

Head Shot

Mark Donnigan

Host
Head Shot

Anita Fejter

Producer
Head Shot

Sebastiaan Van Leuven

Guest
Head Shot

Stefan Ideler

Guest