Gadgetory

resinize garrison from henry authority we're here at Mobile World Congress 2017 I'm here with Keith Preston from Kolkata Keith great to see you great to see you he looks like having a great show me and you've got a quite a great booth going on here and you've got awfully lots to talk about particularly the Snapdragon 835 tell me a bit about that yes we're really excited about the 835 we announced it at CES first products in terms of handsets are being announced here at the show at Mobile World Congress but we're really said about the 835 we expect a lot of success so we were just talking earlier about the unique aspects of the of the eight three five let's start with the application presence ten nanometers that's leading age JUnit tell me a bit about that yes so it's the first product I'm ten nanometer so when I mean first product I don't just mean mobile I mean any device in the world okay on ten nanometer so 10 nanometer provides what you'd expect in terms of new crosses technology better performance better power better area and the things that are of increasing importance in the handset space are visible area and power so 10 nanometer gave us a power advantage also allow us to shrink the package of our chip by about 35% which allows OMS to keep the same visible parameters of their phone we had put in more functionality or larger batteries that was the talk to somebody out it with the set of his area into the silicon cost fabrication because it's actually a large part of the overpricing of these things so reducing that is very important in terms of the economics yes certainly got it yes we couldn't produce the same chip if we were in the previous generation process so it's important that for us to get that smaller package get that lower power while at the same time functionality substantially well that's a great claim you couldn't produce the same chip on it on a grater so even at 40 nanometers you have to go to tend to make the a 3-5-2 chipper that's going to be that's absolutely right yeah all of the cores that are integrated in our Qualcomm design course and we designed the course in parallel with the process so we designed them all specifically for 10 animator so let's just talk about the courtroom and I know these you've got to try of 280 calls in them and this is the first time you're doing the built on cortex technology with a little bit about that yeah that's right so so there's there's different CPU types one you can just use the exact same CPU from arm one you can take an architectural license and design of McGraw ronzo this is a different program you've partnered with arm so it's built on cortex technology and it's a it's similar to what you get from arm but there's been some modifications including the implementations that are designed specifically for the Qualcomm chip I wouldn't ask about the secret sauce but we're all kids we'd love to know but I know you're not gonna tell us but that's that's great that you you've got is it is a custom core in the sense that you've had a major input into how that's been customized isn't it well yeah especially on the implementation side so from from a core design there's the fundamentals are mostly from arm but in terms of implementation it's ours and the key to is willing to make sure we don't look at the IPS discreetly we want to make look at things from an SOC system perspective that works well with our chosen buses works well with our cash structures interfaces well to our TPU DSP because as you know especially for mobile the length is a lot more than the CPU and more and more of the applications are actually moving from the CPU to the other course because customers want to get the best power performance they can so as time progresses they're moving to more specialty course I must have noticed that in your presentations you deliver the 8:35 as an overall solution and from camera DSP graphic you're not just saying hey we've got this call with you it's the overall solution they gives you that great experience on mobile as a whole package yes for example for your other customers obviously their connectivity is key on camera is key camera pre and post processing audio processing video processing machine learning many of these things are going on on whether it be the GPU or our industry-leading DSP and that really also reflects the changing face of the mobile market maybe worth talking to me is about music about AI and VR they are here we are today saying hey off chips are built to take into account those new experiences that happen yeah that's right in many new experiences I mean you know what the algorithms are you know their specific scale or a vector processing you know for example some of these machine learning algorithms provide certain levels of multiply accumulating functions those functions are best done in the DSP that's one of the reasons our DSP we added vector extensions so even perform a lot of advanced mathematics with in low power as an envelope as possible doing it in the holoband nor than the self to it yeah not in the generic processor so the generic processor is great for you know application loading or some of the physics and some of the games obviously the GPU is important for running the games and graphics but then the DSP is where a lot of the algorithms come in for camera pre and post-processing for audio processing or sensor processing and for machine learning for example if you compare the CPU Indian 35 which is a great CPU but you compare the CPU versus the DSP it's about 25 times more efficient on running many of these algorithms panicles DSP is really just nobody the right name nowadays it's so much more complicated maybe it's already there so it's really yeah that's really a good point sir so DSP is kind of a historical name you'll see more and more some of the others are calling things you know vision processors are a multitude of different names but the fact is that the DSP is found the underlying functionality we've added extensions to it and with those extensions specifically designed to manipulate pixels for images or or sound or the machine learning and artificial intelligence so for example with the 8:35 is the first time we support tensorflow so someone designs a machine living algorithm a tensor flow we have a software stack that efficiently distributes that load across our course and if the decent DSP is free that then is the most efficient core to process that tensor flow algorithm and it's really important so the days of true heterogeneous computing upcoming or the republic's yeah they're absolutely here nothing - no doubt about it in you know we have about a 25% power advantage on the a priori fibers 5820 part of that's the node but part of that also is the header gene processing and the software that's all involved so if someone's trying to run image recognition or facial recognition or advanced always listening processing for audio the death can be done in a low-power envelope and increase our days of use so it's really hard to make things much more capable and improve battery life but we've done that with the 8:35 and a lot of that is because of the energy is computing the emission now you mentioned the the GPU for specially good consumers that's often a big ticket item do want time to be about the GPU and yeah so so we advanced the GPU by about 30% in terms of capability from the previous generation it's the same family but every we typically promote the new family about every two years the adrenal 3 adrenal 1405 this is in the adrenal 5 family but then the second year that we come up with the same architecture we find out ways to optimize it and so we've improved it roughly 30% we're really pleased with our GPU because not only is the performance great but the power is great so if you go on you run a manhattan benchmark or something or pick pick your own graphics benchmark you can measure the frame rate we have a you know either on a par or leading framerate across the best in the industry but then when you measure the power it significantly better than anyone else and we have a pretty strong graphics team internally that's one of the things that makes the a35 great a really power performance all this is about its amped isn't it between these two things you've got battery but yet we want great framerate we want great processing speed that's right we actually are even though a peak performance matters I'd say it matters a lot more historically much more now that focuses on sustain and that's because of the use case for example let's say you slide in this for a dreamer here we are etc for the VR experience you want you don't want to lose frame rates after you've played the game for 10 minutes or something you don't want it to overeat and so peak performance going idle is a use case that maybe mattered more than past that it does in the president in the president what cost any customers want is to be able to run their game whether you're using the handset or they're using some like pokemon go for AR or they have a virtual reality mode to have equal sustained performance was trying to first use it till they're done and because of that it's performance at power is the critical benchmark absolutely well case message great speaking to you my name's Gary seems row and row Authority we're here at Mobile World Congress stay tuned for more news and interviews during this week and thank you