Voices of Video
Explore the inner workings of video technology with Voices of Video: Inside the Tech. This podcast gathers industry experts and innovators to examine every facet of video technology, from decoding and encoding processes to the latest advancements in hardware versus software processing and codecs. Alongside these technical insights, we dive into practical techniques, emerging trends, and industry-shaping facts that define the future of video.
Ideal for engineers, developers, and tech enthusiasts, each episode offers hands-on advice and the in-depth knowledge you need to excel in today’s fast-evolving video landscape. Join us to master the tools, technologies, and trends driving the future of digital video.
Voices of Video
Quality, Cost, Latency: The Hidden Battles in Video Delivery
Starting with signal reception at massive satellite dishes, Martin Azpiroz, director at Bold MSS walks us through the complete delivery chain, from transcoding and packaging to content protection and playback. We explore Bold MS's fascinating technical evolution from CPU to GPU to their current ASIC-based encoding approach.
This shift hasn't just improved video quality; it has also dramatically reduced hardware failures, server footprint, and power consumption. As Martin explains, some providers have discovered that server energy costs can reach 30% of the hardware's replacement value annually, making these efficiency gains financially significant.
• Bold MS offers a complete software stack handling everything from ingest to playback
• Their transition from CPU to GPU to ASIC encoding has dramatically improved density and reduced power consumption
• ASIC-based encoding provides better resilience when dealing with satellite signal issues
• Implementing CMAF and low-latency technologies reduces streaming delay to just 1-2 seconds versus cable's 8-10 seconds
• Their encoding approach varies by device type, with mobile streams typically capped at 3.5 Mbps using H.264
• Smart TV usage is growing rapidly alongside continued mobile dominance in Latin America
• All premium content requires DRM protection using Widevine, Fairplay, and PlayReady
• Their CDN strategy includes both their own infrastructure and overflow to third-party providers during peak events
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.
Martin Azpiroz:Voices of Video.
Jan Ozer:Today we talk with Martin Azparaz, a director at Bold MS, who are systems integrators focusing on providing new end-to-end solutions to pay TV, internet service providers and telcos. Among other topics we'll be tackling today are how choosing an encoder can impact signal quality and robustness, particularly when receiving signals via satellite. Thanks for joining us, Martin.
Martin Azpiroz:Hi Jan. Thank you for inviting me. It's a pleasure to be in Voices Video.
Jan Ozer:Thank you, thank you again. Why don't you tell us a little bit about your company, where you're located, where you're moving to, who you serve, what your products are?
Martin Azpiroz:Okay, bolt MSS is a company, as you say, is a system integration and software development company. We develop platforms, always focused on streaming video, internet and all the world that is around us. We are based in Uruguay, but our main focus is, until this year, in all Latin America. We have customers in many countries of Latin America, so we started our company back in 2011 and started developing customers in Argentina, ecuador, brazil, costa Rica, honduras, peru, chile, uruguay itself. So we've been a lot. This year we are incorporating the state, so we are trying to reach that market. That seems pretty good for us. Our platform works on live and VOD. We transcode, we translate, we put on CDNs that content and we have CMS applications and we have CMS applications basically end-to-end solution for ingesting the content and delivering to the subscriber. At the end of the day, that is the goal of our customers is get content on the subscriber's iPads.
Jan Ozer:Who are your typical customers? What type of businesses businesses are they in?
Martin Azpiroz:yeah, we work on b2b environments. Our typical customer pay to be operator or telcos, so company that's want to have platform to to help the customers and subscribers to reach content over OTT.
Jan Ozer:Okay, and where are they getting their signals from? What are the origination streams you're working with?
Martin Azpiroz:Yeah, from many sources. Usually in the paytip industry they get all the signals from satellites at the head ends. They have the biggest C-band dishes and the ARDs at the head end. We take the signals from there, from the same point that they use for the traditional pay-to-view operation, and from there we do all the magic to put that content over RTT on the subscribers. But there are many sources like SRT is very popular now so many programmers or many content owners start delivering content on SRT to the pay-to-view operators. So we take that input, simple OTT channels that are on the internet. We take that input, sdi as an example. Sdi if they have local content and local content generation. We can take sources from SDI files, video files, from BOD content. That's the kind of sources we handle.
Jan Ozer:Okay, and you talked about an end-to-end solution. Are you selling consulting, or do you have your own software programs, your own software stack, or is it a combination of both?
Martin Azpiroz:The software stack. Basically, we sell or we license software that we developed. So we provide platform as a service or platform as a product from end to end, not only the consulting or advising our customer what is best or what is needed, we provide the product itself.
Jan Ozer:And the major components. I mean, do you go from transcoder to media management to player, or you know where are your products in that stack?
Martin Azpiroz:Okay, perhaps we can share a workflow that we prepared A workflow and we're going to explain a bit. This is a typical workflow from Ingest to client layer. So at the left we have the Ingest layer where all the inputs we described are listed. We work on the processing layer, distribution layer and client layer, client application. So in the processing layer, the present layer, we have CMS for VOD channel for VOD content and linear live streaming platform, that is, lsp for live channels. We use ASICs from NetInt to do the transcoding on the processing layer. It's very important for us on that layer, the integration with NetInt. Then on the over-distribution layer we have CDN software where we place all the live or VOD content and is distributed through different edge. We can work with third party CDNs also. Then on the client layer we have the application itself, so the application that runs on the mobiles or on the TV to render the videos or the content itself.
Martin Azpiroz:Okay, that's impressive so basically what you see there in blue and orange is from bold. What do you see in green? Or gray are from third parties. Okay, or gray are from third parties.
Jan Ozer:Okay, what codecs are you using to distribute at this point, and what might a typical encoding ladder look like for you?
Martin Azpiroz:With codec. We support H.264, habc, vp9 for distribution, but we packetize over HLS Dash, microsoft Smooth Streaming, other less popular than RTMP, rtp, several ones, but the most used packager are HLS Dash and the most used is H264 over that.
Jan Ozer:Okay, how much 4k and HDR screens are you getting at this point?
Martin Azpiroz:Well, we work with 4k HEVC. I think that since three years ago we're working. We have a couple of customers developing. In fact, in our region there's not too much 4K content to provide to distribute, so it's been developed and been deployed 4K in the region. But our product works fine with 4K. We have a couple of customers that have a few channels on 4K.
Jan Ozer:Okay, what about ATV?
Martin Azpiroz:Sorry, we expected in the past that 4K will be a killer application, but it's growing slow in the region.
Jan Ozer:How about ATVC? What's your percentage of H.264 versus HEVC at this point?
Martin Azpiroz:I think 5% HEVC 5.
Jan Ozer:Okay.
Martin Azpiroz:Just 5% is pretty low, so we want that to increase, but the markets and the content providers are not providing that content.
Jan Ozer:Okay, tell us about the migration that you've worked through on the transcoding side from CPU to GPU to ASIC and how that's impacted operations and robustness and stability from the signals you're receiving from satellites.
Martin Azpiroz:Okay, we started using CPU for transcoding back in 2011, 2010, 2012, for I think that our first product on the street was in 2013 with CPU. So we started developing in a couple of years before on VOD. It works well for VOD, nonlinear contents demands less hardware, but the quality was good. Then the live OTT arrived back in 2015 or 16 and everybody needed live TV over OTT. Then CPU. We realized that wasn't enough to put a medium-sized pay-to-view operator with 120 channels lined up. We needed $300,000 of servers or more. So it wasn't good with CPU. So we started using CPU. It was a big improvement. It was a big improvement with every improvement. Then new problems arrived, like energy consumption, density, cost. But was a good experience with CPU.
Martin Azpiroz:For a couple of years I think from 2016 to 2021, 2022, we worked with GPU. Then last year we started working with ASICs. It was a new stage for our company in transcoding live content. Of course we use also for POD, but the killer application is live content because better quality, much better density, much better density. With less servers you can handle a channel lineup of 100, 150 channels, four or five profiles, better quality, better build rates. So the energy consumption is much better. Of course we have servers with five years running GPU much better. Of course we have servers with five years running GPU and we don't have servers using ASICs with five years. But every some months when GPU fails or our software is resilient our platform is resilient or failure of some cards, but GPU works with a very high temperature.
Martin Azpiroz:They are not specifically designed chips for transcoding. They can do a lot of things besides the transcoding, so they consume a lot of energy. They consume a lot of energy. They consume a lot of energy In the computing world. When you consume energy, that means heat. When you mean heat, it means failure. So every some months when a CPU fails and you need to replace, that's a nightmare for systems that are running 5,000 kilometers from your office. But when we started using ASICs, they don't consume energy, the consumption is pretty low. They don't produce heat. They are fanless. They are less critical failure points, so critical failure points. So at this moment, in all our deployments we didn't have any, not one, card broken. So we don't have five years running, but we don't have one card broken in quite some time. So for us.
Jan Ozer:It's important on reliability on the platform. Okay, that's good to hear. What's it look like programmatically? How are you talking to the cards, whether it's GPU or ASIC? Is it FFmpeg or?
Martin Azpiroz:are you talking to the API? Yes, FFmpeg, FFmpeg. We have some particular implementation of FFmpeg. We are working since 14 years with FFmpeg. We are working since 14 years with FFmpeg, so we have pretty good knowledge deep inside video transcode technology. But it's pretty similar. I think that the first time we had the first ASIC card in our hands, it took two weeks to have a system running, not properly, not finished for production, perhaps a couple of months to have it in production, but in two weeks it was running and you were changing the technology, the whole approach of the technology and into which has something running. In a product that is so complicated and have many options to do, fine tuning is very good. So that was something that we evaluated very positive in the migration.
Jan Ozer:So the transition within your software layer was pretty straightforward. Two weeks is not a terribly long time.
Martin Azpiroz:No, it took a couple of months to have it ready for production, but it's for sure not a long time in deep software development when you work deep inside the video transcoding.
Jan Ozer:You had talked about the impact of your encoding technique on signal robustness, particularly the signal coming in from a satellite. Could you talk about that and what you saw with CPU, gpu and ASICs in terms of the stability of the signal?
Martin Azpiroz:Yes, we found ASIC a bit more resilient. You know, when you receive satellites there are so many factors that can impact on the stream, on the stream Solar flares, solar storms. That impacts on the, on the, on the receiving, the receiving, the channel, the satellite signal. So the Satellite input usually have some problems attached with the technology On CPU is very robust. On CPU is sometimes when you don't have the proper stream, at the end of the day to do transcoding you need to have the elementary stream pure, the elementary stream, with all the PCI table detached and all the. Sorry I might be technical, but when you have the elementary stream pure, there needs to be when you, when that source is not robust is you need to work on that, on that video, to need to work a lot. And we find the nice X a bit better than previous analysis so that translates to fewer failures or fewer reboots, or fear, fear, fear, fear failures.
Martin Azpiroz:Sometimes you need to work in parallel with with the same. When you, the clock is being desynchronized from the stream. So if you have more hardware available because you're more dense and you can work better, to have the second stream arriving to replace the one that is failing. I don't want to go deep in technical what we do, but, yes, fewer failures and easy to to solve the problems okay, so you've.
Jan Ozer:You've you talked about being, you know, serving customers. Since around 2013 you met, you started off with CPU based transcoding systems. You must be seeing a lot of customers kind of needing to upgrade. What do you, what are you seeing with those customers and what are you recommending on how they go forward?
Martin Azpiroz:You know, most of them are working on GPU. None of them are working on CPU, perhaps some legacy BOD transcoding that they don't need. They receive 100, 200 titles per month. They don't need more than a CPU and they don't need to upgrade. But most of them they are already CPU and we are encouraging them to use ASIC because it's all what they have. It's all they need to replace servers, and the new technology we are working with is this one with all the good features or improvements that we've been describing.
Jan Ozer:Okay, and what are they seeing regarding their power costs? Is that going to drop significantly between GPUs and ASICs, or is it going to be pretty similar?
Martin Azpiroz:It depends. There are two approaches. All of our companies are thinking in green, so we are very concerned about concern about reducing our energy footprint in the society. So on that approach, for us it's important that once we change technology, or when we migrate from one technology to another, it should be greener. That approach is good for us and it's good for our customers. Most of our customers are aligned with that motto.
Martin Azpiroz:But the second approach is the economical. The economical if you spend more money, you know a full server with with 8 CPUs, with 8 CPUs, 2 Gold Xeon perhaps. It consumes something like $3,000-$4,000 a year of energy, depends on the country the cost of energy Latin America is. There are many differences between countries on matter of energy cost, but in some countries one server can consume 30% of the cost to replace the server per year. So you could replace the server. You have a new server, you have a new technology and it pays in two or three years. It pays just with the reduction of energy. Okay, that's a good, that's a good. That's the second approach of being less energy consumer.
Jan Ozer:What are you seeing as the impact of sports on how you configure your encoding ladder? Are you looking at more 60 frames per second footage for football, or is football not a significant component of the overall video mix that you're distributing?
Martin Azpiroz:Well, sports is very important, I think, I will say in our region, but I think in the whole world. So we are a huge fan of soccer and nothing can be worse than watching your team on Sunday and your neighbors shout the goal before you watch it. That perhaps could be the worst experience you may have. The ball is still running and your neighbor goes. So we implemented CMath and low latency, low latency technology to do that. The latency of the transcoding part is important. Asic has a good transcoding part, good latency, and that so improves a bit, improves some milliseconds that when you're talking about 20 seconds it doesn't seem to be much, but when you are talking on CMUF, on sub-second, on sub-second, end-to-end, from the incest to the delivery, some milliseconds matters.
Jan Ozer:So your customers, they have both cable TV and streaming. Is that a lot of them?
Martin Azpiroz:Yes, usually all of our customers that pay TV, not Intel customers, but in pay TV they have cable TVs and OTT and subscribers that are watching OTT, usually using SMAF and low latency technology. Usually they have the goal before the cable set-up box.
Jan Ozer:So tell me about the latency for those two mediums.
Martin Azpiroz:It depends from customer to customer. It depends how many transcoding they do inside the head end for the cable operation. So if they do multiplexing I missed the word, sorry, I miss the word. Sorry, I shot the multiplexing online on the fly sometimes they add more, more latency. So it depends on the head end of the cable, of the cable operation and sometimes they have eight seconds, 10 seconds from glass to eye and in OTT today you can have one second or two or even two seconds.
Jan Ozer:And that's with low latency CMAP you're using.
Martin Azpiroz:Yes.
Jan Ozer:Oh, that's nice. What are you noticing about signal robustness? When you're at that latency level, are you seeing a lot of failures if there's a drop in?
Martin Azpiroz:bandwidth. Yes, when you approach to sub-seconds, you lose resiliency, the robustness goes down. So sometimes it's not important to be in sub-seconds. If your competition or not your competition in terms of commercials, your competition, that is, your subscribers watching on the TV set with the set-top box on cable, is 8 or 10 seconds, you don't need to be sub-seconds. You don't need to be sub-second Right. You don't need to be sub-second Because any problem on sub-seconds impacts on glitching, impacts in the video, because you don't have time to recover and so your benchmark is the settop box on the cable TV. So you need to be equal or below.
Jan Ozer:It's your neighbor's set-top box, not your….
Martin Azpiroz:Yes, particularly your neighbor's set-top box. Ring the door at your neighbor's.
Jan Ozer:So what is the platform you're primarily distributing to? Is it mobile? Is it smart TV?
Martin Azpiroz:Mainly smartphones, mobile, but Android TV is growing a lot. Android TV is growing a lot and in the last year it's growing Tizen and WebOS is growing Tizen and WebOS so. Smart TV is growing much, much more than mobile, but it's still 80 mobile, 8% or something like this web, and 10% 12, smart TV that are not Android.
Jan Ozer:What's the? Are you 5G down there 4G? What's the typical viewer watching on? What connection?
Martin Azpiroz:In our region 4G but works well with 5G. As a matter of fact, I have 5G and use perfect, but most of the subscribers from our customers have 4G, lte or even 3G. So the ladder of bandwidth is very important when you're talking about mobiles, because it's easy when you have 400 megabits of IC megabits on 5G, but when you have two megabits you need to have a proper ladder to go down on the bit rate to have a good video streaming there.
Jan Ozer:So what are you doing at the lower bit rates? I guess you're going to very small resolutions, are you also?
Martin Azpiroz:dropping Depends on the customers. Depends on the customers.
Jan Ozer:Okay.
Martin Azpiroz:Who really knows the network is our customer. So we define the bit rate ladder with the customers. We have some, we do some inquiries with customers, get feedback from them what the average speed of Wi-Fi, of mobile, of, and we design and we design. We have some customers that start in 400 kilobits, customers in small cities with bad infrastructure, the first, the first, the first bit ready 400 kilobits, but usually is 1.1, 1.2 megabytes the first one and 3.5, even 5 or even 8 megabytes the first one and 3.5, even 5 or even 8 megabytes for 720 and the highest one on 4K, something like 8 to 10 megabytes, hevc.
Jan Ozer:Okay, you may not know this, one of the. I just published an article looking at ways to reduce bandwidth costs, and one of the suggestions was to use different manifests for smart TVs and mobile devices. So your smart TV may go. Is that something you're seeing? Is that something your customers are doing?
Martin Azpiroz:Yes, we do that.
Jan Ozer:Yes, we do that, we offer to our customer the request that even if you are on Wi-Fi on the mobile or you are on LTE or the mobile network, you can have different manifest. Okay, that's pretty advanced stuff. What's your top rate that you see for a smartphone? What would you recommend For 1080p video? When does additional bandwidth?
Martin Azpiroz:3.5 megabits, h.264.
Jan Ozer:H.264. What about ATVC? How much will that reduce that?
Martin Azpiroz:On mobiles we don't recommend because some phones will try to use ATVC and some phones don't have the proper hardware to decode that and do it by software and it drains the battery of the mobile. So we don't recommend manifest higher than that or hvc for for mobiles. Now if we are talking under tv or smart tvs, of course okay, that's because we are talking about devices that are attached to the electrical power, so consumption for for them is not important. But for a mobile we are not doing HEVC. I think in a couple of years all the phones or most of the phones deployed on the market will have hardware decoding for a TVC.
Martin Azpiroz:So okay there's sometimes there's right management issues. The requirements from the studios to have a TVC or mobiles requires level level different level of wine. So there's a lot of ingredients on that decision. Today they are not doing HVC on mobiles, but they are doing HVC on smart TVs. It seems pretty logical to have 65 inches or 85 or 80 inches TV set, to have a 4K HVC, rather than watching on a 6.5 inches mobile or 6.7.
Jan Ozer:Is mobile 6.7. Okay, I noticed in the software stack that you presented earlier that you've got DRM. What percentage of content?
Martin Azpiroz:goes out with DRM among your customers 100% 100%, yes, 100%.
Martin Azpiroz:And you have customers that? Yes, because it's pretty easy. All the premium content requires DRM. All the premium content requires DRM and most of the non-premium content requires DRM too. Perhaps if you have a local channel and you are the owner of our customer paytb operator has a local channel. They are the owner of the right pay team operator has a local channel and they are the owner of the right of the channel. They don't want DRM. But you are applying DRM to all channels. Why don't have one? From the technical point of view, it's not wise to have different technologies on different channels. We apply to all channels. Sometimes the content sorry, sometimes the content right owners update our platform and they want to see everything protected from end to end.
Jan Ozer:So how many DRM technologies are you typically using to accomplish that?
Martin Azpiroz:We use Widevine, we use Fairplay and we use PlayReady.
Jan Ozer:Same provider for each. Do you have your own servers or are you using a service provider?
Martin Azpiroz:We are noticing from the provider, usually our customers. We are a strong partner from Nagra so usually our customers have our Contego Legacy multi DRM or or Niagara servers for multi DRM. So we usually use that technology from them. But we are not using that. We have custom. We using easy the RAM by the DRM.
Martin Azpiroz:So okay you have the API. We have all of those API implemented. We have two engineers that are CWIP from Widevine, certified and also certified by Apple, so we can talk Widevine language or fair play language ok, are you looking at any protection schemes beyond DRM, like watermarking or anything like that?
Martin Azpiroz:not at the moment. We we planned to do some watermarking with NAGRA in the past but didn't work in our region. It's not required, but we are ready to do so. We'll do that. But didn't work in our region. It's not required, but we are ready to do so. We'll do that. It's very interesting. Watermarking technology is very interesting for us.
Jan Ozer:What about the CDN side? I noticed you supported multiple CDNs as well. Are you using one CDN per customer in a region, or do you have auto-switching, where you're detecting service and making decisions based on that information, or how does that work?
Martin Azpiroz:well, we have a software for doing some CDN. If you want to buy, if we want to buy a whole solution and a closed environment solution, you can have your own server for CDN and you deliver on that. If you want to approach in more OTT, open to the internet solution, we can use our CDN and do some offloading on third-party CDNs like CloudFront, leaseweb, akamai, whatever. For us we have, I think, more than 10 CDNs integrated. So once you meet there's some logical integration, you meet the the one with in the contract with akamai. You can put more, more, more content, more sorry more subscribers on cloud front. As an example, if you have 10 gigabits with akamai and you are reaching that, you you do the overflow to other.
Jan Ozer:CDN automatically. That's okay for us. We do that. Is that necessary? I mean, are most of your customers serving one?
Martin Azpiroz:country and a single CDN does a good job with that, or do you really need all the…. Sometimes it's necessary. Yes, If you are streaming football and your national team is playing. All CDN help is needed Because, as an example, you have a system developed by OO, the project to have 50,000 subscribers concurrent and from one moment to the other you have 500,000, half a million you need to overflow to other, to third party.
Jan Ozer:Okay, what do you? Moving to the United States seems like a very interesting strategic move. You've got a lot of expertise developed in Latin America. How do you see that helping you as you move into the United States?
Martin Azpiroz:At the moment we are incorporating a company there. It's hard to do business from Latin America to the States we are incorporating there, from Latin America to the States we are incorporating there. We usually go to all the trade shows, all the major trade shows CCT, nab. So we have contact with many prospect customers there. But usually we have friendly contact with many proper customers there, not a commercial approach. Now we want to do a commercial approach to those prospect customers that we already know that will be our approach.
Jan Ozer:Okay, I mean your software stack is impressive. You really do have a very capable end-to-end solution, and so you should have good luck in the United States. That's all I have. I appreciate you coming on with us today. I appreciate you going into some of the topics we went into. It's pretty fascinating stuff, and thanks for your time.
Martin Azpiroz:Thank you, jean. It's a pleasure to talk with you. You know a lot of this technology, so it's good to have a chat an informal chat with you and discuss about what is going in this technology.
Jan Ozer:Okay, let me know if you get to NAB, because I hope to be there and we can touch base.
Martin Azpiroz:For sure we to be there and we can touch base. For sure we will be there.
