Gadgetory

hello my name is Gary Sims from Android or authority now one of the most important components on a system on a chip besides the CPU is the GPU the graphics processing unit now for some people the GPU is shrouded in mystery how does it work what does it do well let me explain okay at the simplest lowest level a GPU is a clever piece of hardware that's very good at certain type of mathematical operations in 3d graphics everything is made up of triangles lots and lots of small triangles that are built together to make complicated models now as you know a triangle has three corners three points and each one of those points is called a vertex and it has a coordinate in the X and the y and the Z dimension so one of the things that happens in 3d graphics is those vertexes need to be processed for example they might need to be moved from one place to another place that's called a translation maybe it needs to be scaled to be bigger or to be smaller and of course it can be rotated now these three basic operations are called a transformation it turns out in 3d graphics the best way to express this XYZ coordinates is using a vector and if you want to perform a transformation on a vector you can multiply it by a special matrix a special four by four matrix now these matrices all have different properties you can do scaling and transfer translation and rotation just using a particular predefined matrix therefore a GPU needs to be really good at matrix operations if it needs to be good at vector operations floating-point operations and matrix multiplication now in the old days a GPU had a fairly fixed pipeline a fairly fixed process so you 3d graphics would go in or your vertex is all your triangles go at the beginning it would go through certain processes and at the other end you would get the display and the how that pipeline worked you could tweak some flags you could do certain things but basically it was a fixed pipeline and then what happened is that you came up with the idea of creating a programmable pipeline where the game designer could actually write the code food that handles the vertexes and something that handles the pixels so what came out of this was a the idea of a programmable vertex shader and a programmable fragment shader and we'll talk about both of those now now a vertex shader was programmable the game designer could write some code that would handle each of those vertex operations now a vertex shader runs once per vertex on a 3d model now after it's been down the pipeline at some point it's getting near to being pixels getting near to being something that we displayed on the screen and therefore you have a thing with the pixel shader or the fragment shader now basically the fragment shader decides the color the texture the lighting and the shadows that are being thrown by the different objects in the 3d world so at one end you have the 3d models with the vertex is the vertex shader let the other end just before we're about to put things onto the screen you have a fragment shader which controls the color of each individual pixel another fragment shader runs once per pixel on the screen now 3d pure is amongst you will say yes Gary it's a bit more complicated that and yes it is but I'm just trying to give here a basic overview of what's happening now where CPUs and GPUs differ significantly is that the vertex shaders are pretty autonomous they can run on each vertex independently and in fact it turns out the fragment shaders are fairly autonomous now of course there are some interlinking between these things but basically GPUs are very good at doing things in parallel and therefore you find that GPUs actually have multiple execution engines in fact what they call them is multiple shader calls a shader that can run vertex work or fragment work and if you look at an are Mali GPU it's often the end of the name like the are Mali t88 0 will have MP something in mp4 and p12 and that number at the end tells you how many shader calls it has and the more shader cores it has the more work you can do in parallel now because GPUs can do things in a parallel manner it actually opens them up to a whole other area of technology called GPU computing GPU computing is when you have a task that needs to be and highly in parallel using 12 shader calls for example that actually turns out that certain types of image processing certain types of computer vision and actually certain types of machine learning work very well on GPUs and so all modern GPUs both on the desktop and on the mobile phone support a kind of GPU computing renderscript maybe or OpenCL which allows them to be used for these kinds of tasks now there's one other thing I want to mention that is of course there is a difference between 3d graphics on a desktop and 3d graphics on a mobile phone of course if you've seen a 3d graphics card from Nvidia or from ati/amd they are they are big they've got big fans on them they've got huge coolers on them they need their own separate power supply sometimes they go occupy two slots you can have two of them in your car on your PC and so on now of course a mobile phone can't do that I mean the Stratis cars themselves are bigger than most phones so of course power efficiency is key when it comes to mobile GPUs and that's why it's interesting that companies like Qualcomm and companies like arm have actually seemed to be doing very well in the GPU market but maybe Nvidia isn't doing so well because it hasn't been able to translate its desktop GPU technology down into the mobile where as specifically designed mobile GPUs from say arm are doing a much better job now of course the amazing thing is is that we as consumers we expect to see the greatest and best 3d graphics we can on a mobile phone because we're used to seeing things on a console or on a desktop and the amazing thing is the GPU designs are maging to bring high levels of quality to a smartphone now of course you've always sensed to yourself when you've played a 3d game the back of the phone does start to heat up a bit and the reason for that is because of the parallel nature of the GPU with a CPU lit if you're doing something some of the cause might shut down it might be only running at a few megahertz now while there are power saving things that go on inside a GPU at that time when you're actually rendering those 3d frames it wants to get them out there as fast it can so all the shaders are working all of the stuff is going as quickly can so that it can produce those graphics pinnacles is doing that 60 frames a second that means every 16 milliseconds is processing a whole frame including the word from the CPU and the GPU and it's amazing that we can still actually get such great graphics from such a tiny piece of silicon and from such with such great battery life so hats off to all mobile GPU designers you've definitely got my respect now in a minute I want to talk to you about the unified shader model and about the Vulcan API but before I do that please just if you like this video do give it a thumbs up also don't forget to subscribe to Android authorities YouTube channel now I said I'd mention the unified shader model now previously the vertex shader and the fragment shader were two specific different types of shader if you look at say the Mali 470 GPU you can see it has a vertex shader and up to 4 fragment shaders what happened with OpenGL ES 3.0 and the equivalent in the DirectX world was that the shaders became unified so any shade it could become a vertex shader or a fragment shader it could do both operations and that's useful because if a shader was sitting idle because had nothing to do because there wasn't any more vertex work to be done it was a waste but now if it can be a universal if it gets unified it you can do any type of shading work then your shaders can be employed to do all kinds of work and nothing ever has to stay idle and of course where it's not sitting idle that means you get greater performance now the latest thing that's happened in the 3d graphics world is the release of the Vulcan API now basically the Vulkan API comes after OpenGL it's this new level with being supported by just about everybody across the 3d graphics industry and it's got two important improvements over the previous api's the first is that there's all the overheads for checking for errors has been reduced significantly because the idea is if you're writing again need to find the errors we're developing it they don't need to be running while you're running the program itself so they've been reduce and that means that there's actually greater performance because the error checking has been reduced the other thing is it plays very nicely with multi-core C P use which means that if two CPUs are trying to do two things to the graphic taper at the same time they can now do it at the same time previously they'd have to wait one for the other now it's multi-core friendly so you'll see that the Galaxy S seven for example actually with the first mobile phone Android mobile phone release that supported the Vulcan API okay to sum up 3d graphics are done using triangles triangles have three corners three vertexes and those three vertexes need to be transformed in size and in position into the 3d graphic world and to do that you use vectors floating point and 4x4 matrices and a GPU is very good at doing those kinds of operation and very fast now in modern day GPUs you can program the vertex shader that runs once per vertex every point on those triangles and you can program the fragment shader every pixel that appears on the screen to set the color and what happens is the programmer writes the shaders he writes the code that runs on the GPU to produce the final image and as the 3d designers come up with more and more interesting ways of handling lights and shadows and textures and all these amazing things you get better and better graphics so the next time you play a 3d game on your mobile phone or even on your console or on your desktop just give a thought to the GPU and all that hard work is doing just that you can enjoy your gaming experience I have to go Sims from Android authority and I hope you enjoyed this video if you did as I said earlier please do give it a thumbs up don't forget to subscribe to handle Authorities youtube channel also please come over to the android aura tea forums there's a special section in there where you can talk to me about GPUs and CPUs and other gary explained videos that I have produce I look forward to connecting with you there also don't forget to download our authority app so you can get access to all of our content directly on your mobile phone and last but not least don't forget to stay tuned to Andrew Authority calm because we are your source for all things Android