 Okay, hi everybody. Sorry, small problem with Windows and Macintosh, but I use Linux usually. Okay, so we are really happy to be here and to be at the Blender conference. I will present you what we do at Karno Computing. Karno Computing develops something very special. We hit people with your rendering, so I will explain you how and why. So first I will explain you the history of this project. In fact, ten years ago I started to work at a large bank in France. And they were using high-performance computing to perform their risk analysis. And they had one of the biggest data centers in the world as a private corporation. And in fact, personally I had a hobby at home. I had many computers in my room and I wanted them to be silent. So I was very interested in all low-performance computers that has no fans, et cetera, et cetera. And once I had a stupid idea, I thought if I can have very powerful computers that perform computation like I do at work, and I can keep them silent, it's a heating system, because everybody here knows about the energy you consume for the rendering. So today more and more computation are used all over the world and they are mainly performed in data centers. Here you have a picture of a very nice data center. Every five years the energy consumed by the data center is doubling, so it's a really big problem since today about 2% of world electricity is consumed by data centers. So they are working hard to be greener, like putting a green carpet like this. So that's the reality of the cloud today. We don't blame them because they do a lot of effort and data centers are really useful, but there is really an energy problem. So 10 years ago the idea of creating a heater with these servers was a bit stupid, but in 2009 I thought it was really a good idea because I saw that there was really a big energy problem. So in 2010 we founded Carnot Computing to design and create the first prototype of the digital heater, we call it the Curad, and you have all the features of the Curad, in fact it is the features of a computer and the features of a heater. You have computing power, you have bandwidth, you don't have that in a normal heater, you have heating, heating you have it in a computer and in a heater, and you have no noise. This is perhaps what is the most interesting thing because without noise you can put that in everybody's home. If you go into the data center you will feel the heat but also the noise. And today this is not only a project, last year we deployed 350 units in Paris mostly, so we deployed heaters in social housing, we have 100 households equipped with this system, and also in engineering school in Paris. So inside the heater you have what we call a raw power, in fact this is bare metal logic. We have only CPU and memory, we have only CPU and memory, and everything is booted on the network, so this is what is called bare metal approach. Each node is Intel i7 at 4 GHz with 16 GHz on each node. And all the data are transferred encrypted and we do absolutely no storage in the heater. When the job is performed it is performed inside the heater and then the heater will shut down their node to perform another job. So let's talk about what I call the payloads. The first year we developed the prototypes, as the queue writes it's our infrastructure, and then we work on what we call the queueware or the queue network, it is the API to distribute the jobs on the different heaters. And what we call the payloads are software as a service, and the first payload we did, maybe it was more than two years ago, it was Blender. It was really by, it was not a choice, it's just because Blender was the easiest real application we can use in our system. And we went to banks and we showed them Blender to tell them we can do the same with your computation, because Blender is something you can really show to people. And we work also on other payloads. So I will explain in more detail what is a payload, but first I will show you a demonstration of our service on Blender. So it is accessible at this URL. You can try it for free, you have 25 hours to try. And I will explain to you how it works on a quick example, if I can help. Is the Windows guy here? Yeah, does somebody know about Windows 8? I don't use this, so... It's Windows 8. Sorry. Oh, it's not in English. I would be like, ah, yeah. Wow. Ne. It looks like no. No? Ne, okay, okay, no. Okay, if you arrive the first time here, you can register or just sign in. Not sure. Wow. Okay, okay, okay. I'll watch right here. Yeah. We're nearly done. Okay, so here you have the interface you will find on the website. We are not a Blender specialist, but we discuss with some people to help us make it clear and simple. So here you have all the projects that you can prepare on the website. So here I already prepared an example, hopefully. And if you enter a project, when you create a new project you have no files. If you want to upload some files, you can go here directly. And you can upload a zip file, or a blend file, or a texture file, create directory, etc. So here in this example, we took the Kaminandes movie. And what was great is that it was working at the first try because the links were well organized. So here are the files. And when you upload your files here on the interface, you can see all the blend files that are in your project. And when you select a blend file, you can choose the scene you want to render. You have some default buttons to choose the resolution, but here you have the default settings of your blend file. You can also choose the version you want to use. You can also subtitle your frames so that it renders quickly. I will explain you how. And you can change everything here. So I will start the rendering here. I hope everything will be okay. So here it started. So here you have the human time. So I just launched it 10 seconds ago. And here you have more than 200 CPUs that are working on this job. And in fact, the heaters in the houses are booting. This is bare metal. So with Linux, it's booting quite fast. I will explain you during the rendering how it works. We use a Docker container system. So at 30 seconds, they are probably booted now. And they will start to work on the rendering of the file that I just chose. Here you have all the elements concerning your rendering. So we know it is styled in 64 parts. You have the first frame and the last frame, the version of Blender you are using, Cycle or Blender Renderer, the resolution, and the number of minutes that will be consumed by the job. You can stop the job anytime you want and you can download the temporary results or the final results here. And here we will see the thumbnails of the partial image. So here you can see that you're starting to have computation time. So here it's human time. Here it is computation time. It already started because we used the output of the command line of the Blender or Cycle Renderer. So we know that we are in progress. And the progress is shown here. And then as soon as we have tiles available, they will be shown here also. So I think in... Oh, I have internet problem. No, it's okay. So that's the magic. It's here. The tiles are starting to come out. The first nodes that have finished their tiles are sending their tiles back to our server. And we gather them on the server to show them I can click on one if I can here. So here you have a preview. This is HD here. We can also output open EXR and do the preview also. So it's quite comfortable for the user. So here you can see that in two minutes we have performed more than two hours, nearly three hours of computation. So during the rendering, I will explain you a little bit about the framework. Everything you do here through the interface, you can also do that with the command line through a shell. We have a shell we developed at Carnot that is very easy to use. So if you want to experiment other things. What can be also very interesting in our framework using Docker is that if you have really a bleed version of Blender, you can push them on our Docker repository for the moment it's not public, but we can collaborate really easily. You just have to call us and it can be interesting to see together the new version you need to push and to be always with bleeding versions. Today we have a model that is quite simple because we offer free credits when you register. You can have 25 hours of free computing but we are really looking for a good model with Blender and we are really thinking about a way to do that is to say if the movies are open we can do the renderings for free. The problem with that is that many people will render cubes with completely hairy things and it takes us a thousand hours of computing and they say, yeah, but it's open. So we don't want to do that. So we want to maybe choose movies that can be good projects to improve the platform, to improve also the pipeline. In Docker we can put Blender but we can also put command lines especially for baking or things like that. So we want to improve the platform and make it more and more easy to use. Our motto is to say that with Blender we can do things we can absolutely not do with other payloads. We tried that with C4D. It was a big pain to create the payload and the command line from C4D is just telling you I crashed because something is missing. So for our clients it's not very good but with Blender we can really do nice things and imagine a great service better than with any other close source software. We are really, really proud to be here. One of the movies that won the prize yesterday was performed on our platform. This is Jonas from France also and won the design award and another candidate also did that. So we are really open if you have new ideas and we see many things very interesting here. It was our first time here at the Blender conference and we really enjoy talking to everybody if you have idea or movie projects we can help. We will be very happy. Thank you very much. I can answer the question while it is finishing. What is the business behind it? People get a free heating or do they have to pay for it? No, I forgot that. People who have these heaters at their home they are heated for free because inside the heater we have an electrical counter and we refund them for all the electricity that is consumed by the heater so they are heated for free. What if it is too warm already? What if their house is already too warm? They can't turn it off. In fact they have a thermostat on the heater so we have more or less computing power but we regulate the process of frequency so even in summer we can keep them in low power mode and we keep between 30 and 50% of our computation power and the other side during the winter wintering is coming we have much more power so we have extra power to give and it's a good time to give computation to we do that for research we work with the university in biotech in France but it's also great to do that with open movies. But you predict that in the next few years if this model will work we will have tens of thousands or maybe even hundreds of thousands of those things or it's unlimited in fact. In fact the high performance computing universe is quite large and it's growing since 40 years so it's an alternative way to deploy computing power and what is specific at Carnot is the bare metal approach we don't do only computation for 3D we do also for biotech or also for insurance that do risk analysis so when the processor finished to work they can start on something completely new it's not a problem. One last thing, how do you work without having a hard drive for a storage? It only has a little... In fact we have a small hard drive in each apartment and when I start the job here so here it is finished when you start a job the data are sent to the different apartment that are ready to compute and when the data are available we reboot or boot the different nodes and they boot in PXE, it's network booting and everything is performed in RAM but we still have an encrypted cache in each house so for example if you relaunch a job most of your data will be already available on site so you won't have to repush everything so that's a way to go quite fast so this is why it is interesting to interact with Asset Managers so that you can pre-push your data so that it is already to run when you click on the button One render shop for like in English it needs many files and textures and everything so you do need some kind of storage the website is hosted in the data center where we have storage so all the data here are stored for example here on Caminandes all these data are replicated here on this website here you can see all the lives and the shots here so every node will use the web server so that's why you don't need local storage so the nodes, the render, the heaters they will load all the data they will pull the data and say I want to compute, I'm ready to compute and pull the data but once they did that once the data are close to them in cache so here I will just show when it is finished so you have the quick facts about how many how many computation you perform and the wattage also that was used and you can also download all the results here you can choose the frame you want to get or you can download everything you will receive a zip file I don't know about the bandwidth here here it is coming here it's quite fast here hey don't hesitate to come it's as we say data center is a little bit like intensive farming and we do organic farming thank you very much what is the price of such a... ah sorry today it's about 25 cents per hour of computation we invoice by the minute so this is our base model because we know it is cheaper than what you can find on Amazon or things like this we have an advantage also on carbon footprint that is interesting large companies but today our model is like this on this side it is a PayPal you have different package you can try it for free first but we really want to do nice things because we have some people paying and they are doing really shit and sometimes and we have people who do great things and we like Jonas that won yesterday and if you have a nice project we are really happy to discuss with you thank you