Gadgetory

hmmm I have Gary Sims and this is Android authority now every year we expect smartphones that are more power efficient and have greater performance compared to the smartphones of the year before but to do that we need new CPU designs and new GPU designs and to that end arm has just released its details on its newest CPU designed the cortex a 76 so if you want to find out what is the court it's a 76 and what will it mean for smartphones of 2019 please let me explain okay so the way it works is this arm is a design company they don't actually physically make any chips they design CPU cores and GPU cores and then they pass those designs over to companies like Qualcomm or to Samsung or to Huawei or - mediatek and they use them in the SOC s that we actually find in our smartphones at last year's CPU was the cortex a 75 and this year's new design is the cortex a 76 now the corners a 76 is quite an interesting design there are some significant changes compared to the court it's a 75 so let's have a look at some of the main characteristics and the first thing to notice is the court it's a 76 is not an evolution of the court it's a 75 it's a brand new design with a brand new microarchitecture however of course it still uses dynamic which means there are high performance calls which will be coupled with a high energy efficient cause and here is the key figure that are telling us there's two times of performance boost for laptops and that's compared to the current performance and that really means compared to the Snapdragon 835 and the cortex a 73 and the reason that mentioning laptops of course isn't now we are in the era of Windows ARM based laptops although I'm a particularly underlining laptop performance whichever way you look at it a doubling in performance from the court is a 73 till the court it's a 76 is a major achievement now when you come to compare the court it's a 76 with the cortex a 75 we see some significant changes here so this is talking about a cortex a 76 clocked at 3 gigahertz with a 7 nanometer process mode compared to this year's cortex a 75 at 2 eight gigahertz at ten nanometers first of all the court is a 76 has a 35% performance increase it has a 40% efficiency increase and when it comes to AI and to machine learning which of course is the kind of the key word at the moment they're saying there is a four times performance increase and here we have some interesting comparisons using Geekbench for the court it's a 73 the court is a 75 and the cortex a 76 of course they are also clocked at different frequencies 2.45 2.8 and 3 gigahertz however we can see that the court is a 76 is again producing significant increases and there's a 35% increase we see there from the court is a 75 to the court is a 76 pretty interesting to note this 2.5 X increase in floating-point so arm have really put a lot of effort into improving the floating-point calculations that we get in this year's processor now we also mentioned increases in performance and in power efficiency on the left here you can see that if you are running the core a single a76 core at 750 millivolts then you can actually do 40% more work 40% more performance in exactly the same power envelope so that means it can do things fortunate and quicker and your battery doesn't go down any faster than it would compared to what the a 75 does and the other way you can look at that if you want to have the same level of performance as you do with today's cortex a 75 processors then actually the a 76 will use 50% less power so you can actually have the same kind of you know geek baby scores as you get today but your battery will drain much much less so that is a significant improvement so let's move on to look a little bit at the internals of the cortex a 76 now the first thing to note about the court is a 76 as I mentioned earlier is that this is a new micro architectural design and arm are confident that each year they're going to be able to take this initial design with this new architecture and tweak it some more to get better performance and even better power efficiency but starting with this first generation of this new architecture we're seeing couldn't changes compared to the cortex a 75 and you might be wondering how long does it take to do a process of design well arm of getting some information here they started the initial thoughts the initial scribblings on bits of paper about the court it's a 76 happened four years ago so that just shows you how complicated these CPUs are and how much effort arm put into tweaking every single tiny little bit of it to eke out as much power as they can so I'm going to show you a fairly complicated picture of the internals of the a76 and even this really is a simplified very simplified version but it just shows just how a CPU is put together and so we can see here there are three distinctive parts there's the part at the beginning which is called the front end which is how the instructions are fetched from memory and how they are ready to go down the pipeline so they can be executed and then you have this kind of decode part works out what the instruction meant to do is it's a floating point in structures it's an integer instruction does this instruction need to access the memory and all that gets sorted out in the decode part and then after that you have the execution part and so that we can get instruction level parallelism there are different parts to this execution because while you're executing for example a floating-point operation you can also be halfway through executing the next integer operation which generally is much simpler now the key takeaway here about the front end part is at the branch predictor and the instruction fetch are actually decoupled from each other and what I mean by d-cup of what it means is that the Brass predictor works independently to the instruction fetch what that means is when the branch predictor is actually working out where the program is going to next predicting the jump predicting the branches it can actually fix those instructions from memory work out what's going on and by time they get to the instruction fetch stage of the pipeline they're actually already in the cache which means it's the branch predictor that has done the hard work of actually fetching them from the memory and working out where they should be and what arm of done is they make the predictor work twice the bandwidth than the instruction fetch which means that while there's all these things going on with memory latencies and working out what goes on next the instruction fetch is always being fed by the branch predictor double the bandwidth and when the branch predictor is actually filled up it's kind of internal q is it hey I've got nothing more to do it just waits until the instruction fetch kind of goes through the instructions one at a time so this decoupling have been able to remove a lot of the latency that you can find in these early parts of the pipeline now it's a twit and a bit over your head a bit just know this they've made it quicker to get the right instructions out of the memory and into the CPU and of course that's vital for performance and then when we look at the execution core look at this there is a eight independent issue cues which are power optimized for the attached execution pipes some things take longer in a process of another so an integer calculation one plus one is maybe a lot simpler than kind of multiplying something by PI for example so when you have this instruction level parallelism it means that if you've got let's say a sequence of integer operations and before that you've got a mathematical operation you can kind of start the mathematical operation off then you can go ahead and kind of get working on those integer things and things start to work in parallel that's called instruction level parallelism so as a summary of what these micro architecture changes managed to get for us is there's a 25% more integer instructions per cycle than the court it's a 75 that is significant there's a 35% higher floating-point performance and here is the really important one my arm have been talking about laptops there is a 90 percent higher memory bandwidth now memory bandwidth is very important both in smartphones and in laptops and a 90 percent increase is a very very interesting so what does that mean for you and for me well basically it means that in 2019 we're gonna see flagship smartphones using the cortex a 76 and we're probably going to hear some announcements about the processors that power those smartphones towards the end of 2018 so maybe just six months from now so for 2019 we're looking at greater performance significantly greater performance better memory bandwidth we're looking at better power efficiency and then we're going to see a push into the win those are laptop area so I'm really looking forward to seeing what companies like Qualcomm and Samsung and Huawei can do with the cortex a 76 so my knife carry Sims and this is an authority I really hope you enjoyed this overview of the cortex a 76 if you did please do give it a thumbs up also you know we'd like to ask you please subscribe to our Channel please share this video on social media and I will be reading your comments below to see what you think about the cortex a 76 well one last thing to say don't forget to go over to Andrew authority calm because we are your source for all things Android