 So here we're checking out an open-source GPU here at the Lunara Connect and who are you? I'm Fabricio, from the University of Sao Paulo, I think it's... Cool, so you're working on an open-source GPU, how does that work? Well, it works basically using an SOC that has an arm inside, then we run an entire software part of the graph pipeline over the arm, then the hardware part that is implemented we get on the FPGA. So the hardware part, when you say that, it means the GPU hardware? What do you mean? The GPU hardware, the part that was implemented on you. And so this is an Altera, a Cyclone, SOC Cyclone 5, this is arm Cortex-A9 or something? Yes, yes, it's A9, if I remember well. And then is this special? Nobody has done this before? Is this the first time somebody do this, like this? Like this, I think the approach of using SOC I think is the first, but there are some GPU implementations. Maybe not the entire graph pipeline, but I think this project is new about this. So what GPU design do you have? Do you make open-source? Yes, it is open-source, you can get it on the link in the GitLab, it is on the PUSH license, PUSH 2, it is open to everyone. Did you design it? Yes. You designed the GPU? Yes, the part that was implemented in hardware, I designed it by myself. So it's not like Mali, it's not like PowerVR, it's not like the other ones? No, no, no, because it's much more simple, but the main question is not that. The main point is the approach, the methodology. You can just take this entire graph pipeline on software, then choose some part of this pipeline and implement it on the hardware, using the FPGA. So you can have a development board and a development methodology to GPU and to GPGPU. GPGPU 2? But you say it's a simple GPU, what do you have? It is simple because it implements just on the Hasterizer. The what laser? Hasterizer. Hasterizer is that point that takes the vertices of triangles and chooses when it is filled with pixels. So inside the triangle must have pixels, outside don't have pixels. So based on this open source project you have right here, it's possible to do much more. What's going to happen in the future? Well, we can do a better memory management, we can do shaders for using the shader processing, we can do the GPGPU approach, we can do video codification and it's hardly like that. But what's the future end goal for this? Is it right now running on the FPGA, is it going to be for everyone on the FPGA? No, I think if we have a very good and very well developed hardware, we can take this and put on an easy key. There's a lot of work to do this, but that is the first step. So there's a lot of work still? Who is going to do this? Yes, maybe me, maybe the community. Today I'm working almost alone, but if anyone wants to help, it will be great. When did you start this project? Well, I did this in half of the 2016 and it took me about eight months to achieve this stage. And you had a presentation here at the LeonardoConnect? You present this? Yes, that's a presentation. I killed the LeonardoConnect. Many people show up? Yes, people get interested in it because well, it's a very good project. It is great to have an open GPU. It was hard to develop it, but it's great. Some guys around here have been working on the open source Adreno GPU, free Adreno, and there's a Mali open source Lima, right? Has anything to do with what you do? No, nothing to do. I think it's much more the hardware part, not the driver part. They're just doing open source driver only? I think, yes, I don't know deeply they work, but I think it is more the driver part. Here we are talking about the hardware, the RTL and everything like that. So, what's an RTL? RTL is the registered transfer level. It is the part that you use to make a NASIQI, so it's the intermediary part to make a NASIQI. And what kind of considerations have been made when Mali designed Mali hardware, or when PowerVR designed PowerVR? Did you study all that, what the differences are, or are they all secret, nobody knows what the hardware is? Well, the point is the hardware is a thing that is very closed. Today the GPUs are more or less equivalent with more or less the same approach, but they differ on the details. So, that's the part we want here. We want a really open GPU with the community giving a great step to the knowledge, to the free knowledge. So, what do you think about the Linaro Connect and what kind of questions did you get at your sessions? What did the people come up and say ask some stuff? Well, they get very excited to know about it. They asked me about the source code, it's very important to have everything open. They asked me about if I could use other platforms, if it is... I don't know, it's things like that. People get really excited about it. What do you think about the Linaro Connect? How about the Linaro guys? Well, the Linaro is very great, it's a central point of the technology. People say about technology even at lunch. People say about technology all the time, that's very good. And that are people that knows very deeply about this, about their subject. It's very interesting. And how is it that the university in Sao Paulo, how is it over there? Well, we have a lot of opportunities in this hardware part. We have some approach, we have some initiatives, but it's not so easy to find the correct people, the right people to talk about, to think about these things. But I think it's good enough to start a development. Anybody working on new chipsets over there, in the division where you are? Some hardware solutions and chips? Yes, people try to do that, but the chip is a very complex thing. And we don't have the structure to do that in large scale. But we have some initiatives like that. Today we are working more on the IoT and the higher level development of technologies. Not so deep in this part that we are talking about here. It's an international network in the community, it has to be global to get these kinds of things done, right? Yes, yes. Alright, so looking forward to the future, open source GPU, what is the name for it? What is the name of your GPU? OpenGPU, it's simple.