Gadgetory

hello I'm Gary Sims from Andhra Thoris and I've been locked exciting announcements from armed this week we've got more details on its new SOC architecture called dynamic we've got details about the new high-performance core the core tech a 75 we also have learned more about the new power efficiency core the core is a 55 and then of course it's the new GPU the mali g72 so my plan is that they said I'm going to do four different videos for you they're going to be shortened to the point in each one is going to concentrate on each of these four announcements so the first day is going to be about dynamic so the question before us is is what is dynamic and how will it affect SOC design in the future well let me explain so first let's start with a bit of history now one of the biggest changes we've seen in mobile processors over the last recent years is the use of heterogeneous multi-processing now that means that in a heterogeneous environment not all the cores inside the processor are the same now typically nowadays in an octa-core processor those before high-performance cause that's a core tech a 73 or 48 72 and then they're before power efficient core that say the cortex a53 now these groups of cores were put into different clusters so you had a cluster or four high a performance course and in a separate cluster or for energy-efficient cause and originally the design work you could actually just switch between them so it was either running for one type or for the other type and that's actually what our app will do in the 18 fusion today but on the other processors from Samsung on Qualcomm and Huawei things have developed quite significantly we're now in a stage we can run all eight cores or six cores simultaneously and the prep or all of them are being used at the same time and jobs are moved from one core to the other according to the energy requirements needed now the promise designs these classes are quite independent so for example across a cluster you might have a level two cache and across the other class you might have a level two cache but they didn't share a cache and so therefore the indicate had to be really clever to make sure that the information in the caches was kind of spread amongst both sets of classism and basically because there was kind of semi-autonomous when it that created barriers whenever there's a barrier between something that reduces performance and so armor been thinking now for about three years how could you make a SOC where there were different types of cause but inside the same cluster which means they could share a lot of the same resources they could share the same cache it would make the interconnect less complicated and of course it would increase performance and that's what dynamic is they can't the solution that allows SOC manufacturers to produce anything up to an optical processor mixing and matching different types of high performance and energy efficient cause inside one cluster now the ramifications of this are some really quite significant first of all it's now much easier for an SOC maker to produce an octa-core processor before they had to have the two clusters are complicated interconnect but now inside one cluster you can just pick and match which of the calls you want to put into that cluster and in fact arm say there are now up to three thousand different configurations that you can make inside a processor using dynamic and that of course includes what cores to use also what levels of caching to you what components to include in terms of floating-point and neon and all that inside of a different cause there really a whole bunch of different configurations that's great for us as consumers because it means actually there will be a whole bunch of different different shading types of CPU so if maybe one company take Alcorn want to go one direction maybe while we want to go to another direction maybe some some will take it in another direction there are just so many different ways and each company will have its unique proposition that why its processor design is maybe better which because as a configuration is better than a different one now up until now and we've really seen HMP processes they've really been two plus four or four plus four so all of the current flagship SOC they're using for high-end cause that's cortex a seventy three for example and then for power efficient cause that's a the cortex a53 and we've had some hexa-core process before where there were two high performance cause and or power efficiency cause but now in with this new system in dynamic we can see that whole thing mixed up for example you might see a 4 and a 1 or you might see a 3 in a 1 or you might see a 7 and a 1 I mean the combinations are anything you like inside of that one cluster up to a processor calls now why they important it means that SOC makers particularly those that are making for the mid and low range can now mix and match much easier what types of cause they want to put inside their process and because it's all about silicon size when it comes to pricing when it comes to profit or about how much silicon is used inside of each of these processors now maybe a mid-range processor might just have one how high performance core would you do this great single trading performance great scores on Geekbench on the single threaded performance and a great score user experience because some types of tasks really do heavily use one single core but then they were still before power efficiency calls for all of those other like work that needs to go on and yet that would be maybe less silicon size and maybe an octa-core processor that was just using cortex a53 across the board so that actually means we can see quite a lot of differentiation occurring now in the mid-range now I'm pretty sure that at the high end we're still going to see four plus four as the norm but there is room for experimentation maybe we'll see five plus three or maybe we'll see six plus two I mean these processor makers now can really start to think about different combinations they want to make and see what comes out better in terms of performance and in terms of energy efficiency and we might see quite some diverse different solutions coming from the big SOC makers and the other big difference is for the first time now in an are mobile based processor we have l3 cache and if you don't know what else recache is I really recommend you go and watch my video what is cache memory gary explains and there you'll see all about l1 l2 and l3 cache now up until now the l2 cache has been shared across the cluster the l1 cache was on a per CPU core basis and there was no l3 cache now if I now make you've got an l1 cache that is per CPU core and l2 cache which is per CPU core and then an l3 cache which is across the entire cluster now if you're thinking that would actually raise the silicon cost because now you would have an l1 l2 and l3 cache it actually turns out that actually the l1 the l2 cache are very much smaller than they were before because now the l3 cache is up to 4 megabytes now in implementing this l3 cache are more than a really clever thing the SOC maker is allowed to partition the cache into different sections and that's good for two reasons one it stops what they call cache thrashing where one particular part of the SOC is dominating the cache because it's shared across all of the CPU cores but also because of this partitioning they can shut down parts of the cache and therefore save power so when a particular core powers down if it's the only core using a particular part of the cache then that part of the cache can power down in fact there are even scenarios where you might not even our up the cache for certain operations only later on you might power that were you when you need it and actually that's all controlled in the software a one last thing worth means because all the CPU cores are now inside of the same cluster powering down those CPU cores so they are switched off and therefore not using any power not draining your battery at all is much easier than it was before in fact the number of steps required internally to do that has been halved and before it was mainly handled by the software inside of it saved within its kernel but now actually it's all handled in Hardware which means of course it is super efficient sort of this means for us as consumers but I think in 2018 we're going to see phones using processes based on the dynamic architecture I think we're going to see them from Qualcomm and Samsung and Huawei and from mediatek now how they choose to do that of course we don't yet know I have a feeling that certainly in the mid-range we're going to see Qualcomm processors using dynamic we're certainly see processors from media tech using dynamic maybe at Qualcomm we'll continue with their cryo series using the built on court a license so they'll take this dynamic they'll take the a 7500 take the 855 and they will modify it to make a custom semi custom processor it's unknown what Samson and I do because they've got the m1 and the m2 cause and I don't know how they going to get to integrate that with dynamic so it can be interesting to see what they do in the future and I think while we are going to basically do what they've done for their last current generation they're going to take four cortex 875 calls and four cortex 855 calls and they're going to put them into a big beefy queueing processor with a Mali g72 GPUs this is my speculation we'll see what happens so to recap dynamic is a new way of building mobile proteins based on the ARM architecture it's coming in the with the cortex a 75 it's coming with the court it's a 55 it'll be mine next videos it brings all of these core up to an optical into one cluster you've now got a level three cache which boosts performance and they and makes it more energy efficient and of course those are the things that we want i'm garrison for my under authority i hope you enjoyed this video if you did please do give it a thumbs up there are three more videos coming in this series so please don't forget to subscribe to our Channel hit that Bell icon to get a notification when we publish these new videos and last but not least do go over to Andrew authority calm because we are your source for all things Android