 Hi everyone, thank you so much for being here. Some of you are working, some others are creating, it's a very interesting thing to be here this year. I've been at the Blender conference last year and it's so great to see how many different backgrounds and people from other countries all come here with Blender in common and each one of them, each one of you creates something different and it's so exciting for somebody like me who created a company all around Blender. It started everything in 2015 when I was a music composer and I was working for a global advertising agency in Singapore and I was sitting beside a 3D guy who was creating things, I think with Maya I guess, so I was very curious about what he was creating just with a computer and basically a keyboard and a mouse and this cube that was becoming something completely different, completely great and I got very curious and I started studying and I started with SketchUp and then I found some limits and I moved to Blender because I didn't really want to spend money from a hobby that I just started and I started getting things, you know, some cool design and then I found that there was the rendering time, there was a huge obstacle and at that point I didn't really want to solve a problem. I just met somebody who had some data centers in the United States based on GPUs and I suggested to do something for rendering and he asked me to run this thing and we ended up with the largest cloud render platform ever built without even knowing and so CoreWeave is a cloud render platform today and it's totally purely based on Blender. It supports Blender in every shape and form. Let me just go ahead for a second and understand what really we make different compared to any other render farm. If I want to put it just in one simple word, we listen to Blender users and we act immediately. We listen and we act based on what are the needs of the Blender users and we came up with two solutions who are currently changing the entire workflow of Blender users for a simple reason because we wanted the creators to stay creative without having distraction because of the render time. So what we have done, we created one product called Cubic, launched this year in January and very quickly Cubic offer 0Q, so you don't have to wait to render your things, one button to start into Blender and you can have up to 80 GPUs per project, parallel rendering, automatic texture, up texture pack. So this means that you don't have to do anything else, you don't have to wrap your textures. It goes automatically. It downloads automatically your render files and it costs 50 cents per hour. I think that a demo playing with Cubic explains much better what this means. So I have Cubic open here and the moment you download it, it automatically installs the add-on for Blender. I also open Blender with completely empty files and what you do, you just activate the add-on like any other add-on. These two new buttons in Blender appear right where usually used to be the old image render and animation render. If I hit this button, everything that is in my scene, objects, texture, simulation cache, everything that you linked files, everything gets wrapped in one single file and gets rendered over 80 GPUs. Let me just pick something. I think I have, let's see, monkey test, let's do something together. I think it makes more sense to create something together. Something really basic. I want to just add some complexity to this. I'm on a Mac Pro, so I'm not going to render view. I'm not not sure what's going on in the scene, but I can imagine that it's going to be a blue text with metallic. And then I'm going to put something dark on on the floor just really to check together what's going on. And I'm going to see, then we push the samples a little bit higher. And what is it? Let's do 500 just for to go a little bit hard on the GPUs. So imagine that I'm a Blender user and I'm working on my thing and I'm ready. I'm ready to, I'm not doing anything unrelated to Blender. I'm just building my thing and whatever it is on my scene. I click on this button and something tells me that I'm okay. Task for Cubic was generated and I can continue working on my scene and do my stuff. While Cubic, once activated with this button, starts working. Okay, let's let's think that I'm still working on my scene. The file is already been uploaded and consider that I'm on the Wi-Fi of this place. Whatever you have at home is for sure better than this. And we are already at 25% actually it's done. So it's a very, very simple basic scene, 500 samples. It's a very simple scene. But this works for any scene, up to 8 gigabytes of memory. So if you have a scene of 7 or 8, it's a big scene. You can render with it. For animation is the same thing. Instead of hitting, okay, this is the result. I can hit this folder. The file is already on my computer and that's my output. So if for the animation, you hit the different button, on Blender is the same process instead of hitting this one, you hit this one. And you have the animation. The part of the rendering simply makes it that each frame goes on a single server with 8 GPUs. Up to 80 GPUs. This means that if you have 50 frames, each one of them takes one minute. You're gonna have 50 frames rendered in one minute. They all come together. So what happened a couple of three months ago, I guess, it's not so long ago. Somebody came in and called us saying, hey, I'm using Cubic, but I have a 12k scene with leaves and trees and old vegetation things. It's for a big projection and I have 600 frames to make. And we said, okay, no problem. If Cubic doesn't work because of the size of the scene or the complexity, we can render for you. We can get it for you. So send it to us. We got it rendered over 1,200 GPUs and it was ready in two weeks. Instead of the normal rental time, that would take probably, I think, half a year to get it done because even on our machine, it was taking 12, 20 hours per frame. So we thought that probably Cubic is a product that it's good if you're doing small, medium scenes, but probably you want to go a little bit bigger when it comes to big projects. So we created Concierge. Concierge is for professional artists. We pushed everything we had to the next level and we continue to improve it. It's something that it's not the single button thing that you kind of like in Blender. There is a couple of more clicks to get your render done, but here you can have access to all the 45,000 GPUs that we have. And all the technology behind this is based on the fastest internet speed in the United States on the maximum level of security and privacy protection, enterprise level, and an interface that is solid as rock because you want to make sure that your work is done without problems. The way Concierge works is as simple as Cubic, but just with one difference which is related to the way you interact with interface. So first of all, there's nothing to download. It's online. So you log in into this interface with your credentials and you have a dashboard where you can control your work. You have a file manager where you can upload your file and launch the render. And then you have a job manager where you can control the job scene process. So the first thing that you want to do, you go in File Manager and then get the Blender, the blend file that you want to render. Let me see the one that we just created. It's the Beacon. Let me see what it is. Beacon 2019 Blend. You drop in the Concierge and then it gets uploaded. And then from here, you click on Launch Render. You select the Blender version that you have. The frames that you want to do, let's go crazy and let's do 250 frames. Even though we don't have an animation in them, that's fine. We can push a little bit the boundaries later and try what happened. So if I hit render, I go on Jobs Manager and I have my Beacon 2019 rendering. And if I click on View Details, I have the entire situation. So I have rendering here and I see that the first frames start coming. If I click on that, I have a preview of my file and I can download the single frame or I can download the zip file. As you can see, I immediately had just 170, almost 170 frames done. But as I said, it's a very small scene. Let me take an example or a more complex scene and it still keeps going. I think we just hit 250. So I think I need to log in with my personal account for a second so to show you other examples of things that we can do with a scene that is a little bit more complex. So just a midway between simple and complicated. I have a scene from the Blender files demo and it's here. It's called the Wasp Bot from Blender. And here in preview, you can see that it takes a little bit of time to open because it's a 2000 by 2000 image and it's 500 samples. And it's a fairly complex thing but it's not incredibly high. So we rendered this thing yesterday on the table upstairs and we watched what was the result. It was 250 frames rendered in three minutes. 250 of this, of the camera moving around in three minutes. I think this is something that can really change the way artists approach their way of working. You can see the timing here. It started at 1422, finished for 225, 250 frames rendered. And that's always the same. It depends really on the project that you run. So the reason why this we believe is very important is because if you think about it, the render time is something that affects you not at the end of the creative process. It's something that is with you when you start working on something. Even if you don't notice that, you're modeling something thinking about the render time in the back of your mind and you don't subdivide too much because of the render time. You don't go for a 6k texture because it will take too much time. You don't push forward probably even the frame rate because of the render time. If you consider the render time as not a problem since the beginning, because it's under your skin that the render time is solved, doesn't matter what render farm or what cloud render service you use. It's just simply you don't render on your computer because rendering with your computer is like thinking that you store all your file in your computer. It's going to be the same thing in the next years. The cloud computing is going to be normal. It's for everyone. It's going to be accessed by everyone. We want to do this for rendering. Imagine that you approach your creative process right off the bat without thinking about the rendering. You can go crazy. You can do whatever thing you want in your creative process. We have another example of things professional production done with CoreWeave. I directly go on YouTube to show you what we did recently with the director that decided to render the entire movie with us. This is the trailer. The movie is not yet out. The audio is from the computer, but I guess it's fine. It's all about images. The quality of the images that we just saw created in Blender with the water simulation with the resolution of the image, the fluid simulation cache, everything, every complexity that you can imagine was there. Of course, we can go beyond this and there are many other complexities in any Blender file, but this specific case took us to think of how many assets are in a project like a short movie. Yes, the textures, the objects, the cache simulation. What happened is that you try one frame to see how it's coming along, and then you see that that frame is not exactly what you were looking for. You try it again, and to try again, you have to reupload it. It doesn't matter what. You have to reupload everything. For Concierge, we adopted a technique so that you have a folder shared on your computer, and that folder is directly synced with your file manager in Concierge. Whatever is in that folder is in your file manager, so that when you open your file manager, which is the EasyClean page that I showed you before, this one, so these are your files. Whatever is in that folder on your computer is there. If you change something in that folder in even just a little thing in your Blender project in that folder and you save it, it's going to be immediately reflected here. So you technically never upload with Concierge. You can upload if you have a Blender file, you drag it inside, but you have 300 gigs of cache simulation. You want to keep it there and you want to leave it there forever and let your computer automatic sync with Concierge. So this is one of the things that came out after listening to Blender users who used to say, yes, but yeah, it rendered fast. So I can see immediately what's going on, but I don't want to upload all the texture again, all the things again, all the package again. Okay, perfect. So we act, we listen and then act. And the synchronized folder is something that we are really proud of because it's really cut a huge piece of the rendering work that it takes to get something rendered. Talking about is that our other application that is Cubic, it used to be a different application just a few months ago where we added, this is our website, where the application, the 2.4, had the two main bars, what is this? Where you can, we could select the GPU that you wanted to use and the number of GPUs that you want. What number is? That number is the actual number of GPUs that you use even on a single frame because we have, and we are the only one to have, the GPU visualization so we can combine the power of the GPUs together even if they are spread across different servers. Combine the power of GPUs means that you can use 22 GPUs of 48 or 150 on a single frame, on a single image. Not splitting the frames around, just firing all the power on that image. And then using the parallel rendering, using another 140 GPUs to the second frame and to the third and to all of them and deliver all to you 100 frame at the time. We're talking about seconds, processing, large scenes, 100 frames at the time. With Concierge, you can go up to 500 GPUs firing all at the same time and delivering you in seconds, even the highest and most complex scene. We decided that this button on this field of selecting the GPUs sometimes was confusing the users, which don't know exactly if it's by the 10, if it's by the 5, if it's by the 148, you don't pay more if you get more GPUs. You technically pay the same, but you have the product coming together faster. So we said, okay, why don't we take it off? Take off the GPU selection and let the algorithm decide after looking at the project that comes in, decide how many GPUs to use. So we are releasing Cubic 2.5, which is the new version of Cubic, which is going to look like this, like the one that I tested before. And it's just going to have one play button to get started. So you still have the model of GPU here, which is a P104. It's very similar to the 1080. And you can click this play button and you are connected. You are not paying when you are connected. You are just paying when you are rendering. And the price is 50 cents per hour per GPU. And to give you just an example, since basically they are the same price, the bot that we rendered earlier, the 250 frames, it's going to be something around $8 to get 250 frames altogether. So we think that these features can really help the users to, first of all, to be free from any boundaries created by the render time. The render time can be, as I said before, a limit to the creativity. And probably with all the new features coming together with Blender, we are able to keep the pace because Blender is going very fast and we are very happy about that. Adding every single new feature and add-on in the background. Just to give an example, the denoiser, both the denoiser, the node and the AI one, we need to implement it in the background to make sure that everything works. So if you use the plugin, we need to have it and it needs to be working. We have all the fluid simulation and the add-on, the flip fluids. We have the animation nodes. Every day there is something new and every day we are adding something to our workflow. I would love to have some questions from you. If you guys have some technical things, they're probably, yes, absolutely. So if you change something in the scene, I'll just repeat the question because of the camera and the audio. So if you change something while it's rendering, what happened, right? Okay, so explaining the way it works, you could probably, I hope I can also answer your question. So the way it works, we get the Blender file from you. It gets translated into data and goes to our facility and then basically that Blender file is kind of wrapped. So whatever you do on your computer, it's basically going to stay on the file in your computer unless you send it again. All with the file manager. Okay, unless you save it. So if the moment you render, let me give you an example. You are in your file manager and this beacon is the one that you have open now. If you launch this render right now, it's going to render that one that you have there. If you change it on your Blender, it's basically going to be changed in that one in this file here unless you click again launch render. So to give you a clear example, the moment you click on launch render, this thing is kind of closed and wrapped and sent to the computers. So the moment you click it again, yes, the ball is going to be blue or red or any other change. There is one portion of time to allow the system to sink. The moment you, to get it started, you just save it in your Blender file or you just technically re-upload it. But the best way to do this, if you have something that you want to change, you save the Blender file, it gets synced and then you relaunch the render from there. It's not going to happen. It's never going to happen that you have the first time frames red and then the rest blue just because you changed it over there. The click on render wraps the file and send it to the machines. Yes, yes, exactly. So for the dependency, yes, let me get a question. Sure. For the dependencies, I spoke today with a guru of dependencies. A guy only works with linked files and linked Blender. So we are launching this week a system that technically accepts any form of compressed file. This is mainly because we wanted to allow anyone with something that is not a blend file to put it in that file, to put it in a folder, wrap it and send it to us. And then while in file manager pointing at what is the blend file that you want to render and then automatically clips all the dependencies. So it's something that it cannot be done packing. You can also pack the blend file. But sometimes you don't want to pack on your desk unless you duplicate it, you pack it and you send it. But you have, if you have a 200 gigabyte scene, then you're occupying a lot of memory. So we said, okay, let's wrap everything into a zip file and send it over to the farm. So the P104 is, if you read the paper, it's four gigabytes, but it's actually eight. So, sorry, unless you unlock it. So if you unlock it, you have eight gigs. And so the most important thing when it comes to virtualization, not to parallel rendering, but the virtualization is to have all the same model of GPUs. Otherwise, it doesn't really work. So we have all the same model of GPUs. They are all 1080, which is the P104. And there is no other model available rather than the new generation that we have, which is the Titan V. So we have the P104 and the Titan Vs. We used to have also the 1070, the 1060s, but we just picked the most performing one, which is, in our case, as you can see, the speed of the rendering, the P104. The size, when it comes to the GPU, matters. So if you have something that is more than eight gigs, you cannot render with us. I'm talking about eight gigs, not the file of the blend file. I'm talking about when you open the file in Blender, the actual memory occupation of an Opal Blend file. If you have a ball, that ball is going to be like probably four mega, but if that ball is a million polygons, it's going to be a huge one. So in that case, we set the limit and we show on the website that the limit is eight gigs for Cubic. For Concierge, we can go up to 16 gigs because we have up to 16 gigabytes GPUs. We are acquiring the 32 gigs gigabytes GPU so that we can push forward and get up to 32 gigabytes. It's not a Hollywood standard. Now, Hollywood is moving on an average of 200 gigs per scene. We are not ready for that thing yet, but they're also not using Blender yet. So I hope that we all hope that Hollywood is going to start using Blender for big productions and we're going to be there, even though we have fantastic and amazing things released by Blender. If you think about NextGen and other beautiful things produced by Blender, those kind of scenes are way beyond GPU capabilities. You can mix them together so that the CPU gets the heavy load and the GPU accelerates just, I think it's called the BBH. So when there are very complicated angles, shadows, and that's when the GPU comes in, but most of the load is on the CPU. So to go straight to the question, the limit is 16 for Concierge. The limit is eight for Cubic. If a scene is bigger than that, we cannot afford it. We cannot get it. For the GPUs, we tend to have all the same model, which is originally used for mining, but they perform pretty good on rendering. So we decided to go for it. Sorry, yeah, the price. So the P104 is the only model we have on both platforms instead of the Titan V, and the price is the same. It's 50 cents per hour on both platforms. Concierge used to be 60 cents per hour because there are other processes going on and together with electricity consumption, but we managed to have both flat at 50 cents per hour without changing it. Technical questions, yes. The evaluation of the price. So breaking things is very, I mean, breaking things with hacking or making something really, well, that's, I mean, accessing somebody else's projects unless you know the password, you cannot do that. If you are a very, very bad person and try to hack the system, I mean, everybody is subject to that kind of hacking, but we have cybersecurity in place. If something goes wrong, we're insured for that. We take very seriously the privacy protection and the IP protection, because even if it's a freelancer trying to figure something out with Blender with the Simple Cube, we take care of everything in the same way, take care of everybody with the same attention. And so the servers physically are not accessible by anyone. We have security cameras and we have cybersecurity for attempt to get into the server because considering that, yes, we run renderings, but with 40,000 GPUs, if you get into that place, you can do really super bad things. I mean, with 45,000 GPUs, you can crack all the passwords of the world if you are a very, very bad person. So the security, it's our priority number one. And it's kind of hard to get into it. The first question I'm not sure about, I'm not sure about the, you mean the Reload Trusted? The old button Reload Trusted that run the Python script? That's a very, very tough question. I think if it works, it goes directly. If there is something that is not detected as default, it gives you an error. So sometimes happened in Cubic that there is a non downloaded from Blender or version of Blender and there is something tweaked inside. Cubic tells you. It also tells to us. So we know exactly what happened and we fix it or we reach out to you and we tell you what's going on. If you go on the script layer, the script level, it can probably happen that something goes wrong because there's a lot going on in the background. But I cannot tell you 100% because it depends on what kind of script you run. But usually it should be good. We run on Linux, old command line so it doesn't even look at the interface. So it technically should be all right. But thanks. One thing that I forgot to mention is that we have a real time accurate cost calculation on the file manager. So the cost calculation of whatever thing you want to calculate is exact. It's not something that, oh, let me see what happened on the file. Oh, probably on the jobs. Let me see. Okay. Probably because the page is small. Let me see. Okay. Right. So you get estimate here. So if you click on this, so this estimate is going to do mainly two things. So imagine that you have the camera pointing at something which is a cube and that's a frame. But on frame 10, the camera turns around and there is a Disney movie. And that's not an accurate estimate if I calculate that frame and that other frame it's left behind. So we take random samples along the entire work that we have. In this case, it's 250 frames. And we calculate the average of that, giving you the single frame estimate so that the only thing you need to do is to take that number, in this case 35 cents and multiply by the number of frames that you have. So this is simply estimate cost per frame. It's very straightforward and simple. There is no cost for this. There is absolutely no charge and it's accurate. It's not a random estimation or a very close estimation of what it's going to be. It's going to be mostly exact. Other things that you can do with concierge from the job manager. Let me see. The job manager. You could technically... Yeah. You can download the zip with everything. You can go, my friends, you can go to instead download the single frame. You can have the preview. We talked about it. Ah, yes. One cool thing about cubic that I forgot. There's many things. So you have this little eye over here. So this eye, it's not only gives you information, it gives you also the preview of what you are rendering. If you are rendering an animation, this is giving you the animation preview as well. Let me just go to where I render some animations. Even if the render is not completed, yeah, you can see, even if it's not completed, even the first 20 frames are done, you can already see what's going on in the animation so that you can check if all the movements are all right. The more it delivers frames, the more this animation is going to be basically complete in your preview. Well, this is CoreWeave. This is cubic. We have concierge for large and complex scene. The price is the same. We really hope that this can unlock whatever block the render time created to you during the rendering experience. For both platforms, we have free credits to start. We offer five credits for both platforms to start, which is nearly like 10 hours of render on one GPU. For the beacon, during the registration process, if you enter beacon 19, you're going to have three times the same. You're going to end up with $20 of free credits here. You log in and during the registration, which is really easy, you enter beacon 19 and you're going to have $20 of free credits only for the three days of the Blender conference. Thanks very much for being here. Thanks for the question. Thanks so much. I am Marco and you can find me upstairs together with the other friends of RenderStreet. We hope you guys are going to enjoy the rest of the beacon. Thank you so much.