WEBVTT 00:00.000 --> 00:15.600 Okay, hello everybody, I'm Yuling and thank you, one more minute, all right, so, getting 00:15.600 --> 00:22.240 me love with, you know, let me do a talk in front of such a big crowd, you know, one 00:22.240 --> 00:29.360 of the best conference in the world for open source, okay, I'm Yuling and the founder 00:29.360 --> 00:43.960 of Deep Computing, I demoted my last career for this fight, so, and I love Chris for 25 years, 00:43.960 --> 00:52.160 yeah, back in zero, so, and now, at that time, we're doing mips and then now, now, we start 00:52.160 --> 00:58.360 doing this fight again, so, and I'm my owner to do the presentation, do the talk, and 00:58.360 --> 01:05.640 asking for help here, and for some, today, we talk about why do we deal with Android, 01:05.640 --> 01:13.440 and how I'm going to do it, so, my name, I'm a software guy, and I do electronic 01:13.440 --> 01:20.120 sadly, and then I do usually do the whole VM potting back in the light since 94, so, putting 01:20.120 --> 01:26.120 all the VMs, I will VM, all the from interpreter, to jit, to whatever, to all the mobile device, 01:26.120 --> 01:33.480 so, Android is not new thing for me, that time, so, and, and I run the NAS in Nokia, run the 01:33.480 --> 01:40.720 future from Java platform, so, and I found the deep computing, and doing this fight during 01:40.720 --> 01:46.480 the COVID, and then, because I'm a software guy, I did the electronic, so, I'm serious 01:46.480 --> 01:53.360 about software, so, it decided to my hardware, so, I'm losing all the money, sadly, very 01:53.360 --> 02:06.480 sad, so, and back in 2023, we make the first laptops with the 4 core, and 2024, we make 02:06.480 --> 02:15.280 the 2 set of a product, and same laptop and tablet, that's the first time, I was going 02:15.280 --> 02:22.600 to put the whole Android, until I found so many mobile devices, so, and this, this tablet, 02:22.600 --> 02:31.280 with the external modem, can make the core, using Linux, and then, the funder of Wisfy, 02:31.280 --> 02:39.520 JvPeterson, use this tablet, call his wife, and, because anonymous call, and then, his wife 02:39.520 --> 02:51.280 rejected the core, so, and it's a post pedolar, and, and Ubuntu, first ever at the time, 02:51.280 --> 02:56.160 so, and then, later on, let's start, and realize that, I'm losing too much money, and too 02:56.160 --> 03:01.840 much time, to make it the hardware, and then, I say, let me focus, do one thing, then, it 03:01.840 --> 03:09.360 decided to part out with the most modular, start up framework, so, they have a book there, 03:09.360 --> 03:16.880 basically, the idea of being open, open up the hardware, so, Apple, go in vertical optimise, 03:16.880 --> 03:21.600 and then, framework, go all the way open, you can fix it, you can upgrade it, you can do 03:21.600 --> 03:25.440 whatever you want to do it, and they open up all the cat, you can make any module around 03:25.440 --> 03:32.480 it, so, I decided to make a motherboard for it, so, and then, it's support, all the architecture, 03:32.480 --> 03:36.520 one laptop, you can support all the architecture with all different motherboard, you can go 03:36.520 --> 03:43.080 come up to our booth, and the framework to, to see it back in the A, W, buildings, 03:43.080 --> 03:51.720 it's a perfectly fit for the European cultures, environmental, DIY, why, same as IKEA, 03:51.720 --> 03:56.920 and then, this is the first motherboard, and we used the four core, then, I made a spin 03:56.920 --> 04:03.560 the ball in three months, so, make the motherboard in three months, fit into their laptops, 04:03.560 --> 04:10.280 and then, I do another generation this year, last year, sorry, and took a while, because 04:10.280 --> 04:16.840 the AI, and then, now, it's coming in the watch, it's a 16-core 24-hour gig, and, because 04:16.840 --> 04:22.680 one on the framework has the fabric versions of 12-inch, you're going to see the 12-inch 04:22.680 --> 04:31.080 motherboard with the fabric, and, and laptop all together, so, today, we talk about Android 04:31.080 --> 04:39.560 on Res5, so, anyone knows how we're doing the chipsets, anyone here, just now, people 04:39.560 --> 04:47.880 talk about mobile, and, and, and when we do in mobile, so, usually now, the meat and 04:47.880 --> 04:53.240 nowhere and already 12 nanometers, for, for especially for mobile phone, for phone 04:53.240 --> 04:59.200 tablet, you might still find the 20 nanometers, so 12 nanometer SOC, they need to sell 04:59.200 --> 05:05.880 one million, so they have to break even, so they're very expensive, and then, if a 20 nanometer 05:05.880 --> 05:13.720 is easier, so, that's why the chip topic is the easiest device, but now, once the DDR 05:13.720 --> 05:21.560 is this price, now the price I got, for 4 giga DDR, is $50 already, so, the chip 05:21.560 --> 05:28.200 topic time is gone, so, now, back to the 16 nanometer, we need to sell two from 05:28.280 --> 05:35.320 my name, SOC, to make a volume to break even, so, you can see, when we look at that, 05:35.320 --> 05:43.320 do we see, what OS can support the hardware guide to survive, otherwise they go bankrupt, 05:43.320 --> 05:56.400 so, windows, they're curing now, right, only a 86-to-wife, I'm doesn't, Qualcomm spend 05:56.480 --> 06:05.760 billions on it, can't even make money out of it, and then,liness, rate of, especially the meat 06:05.760 --> 06:12.720 high end rate of, so, industry, BC, that's that man, he's a very, very romantic, we will not survive 06:12.720 --> 06:20.400 a SOC manufacturers, so, the only thing, people can survive and make their SOC, they have to 06:20.720 --> 06:30.640 do Android, so, far, the whole with fly, and, and, that's actually, I did three ports for, 06:34.960 --> 06:41.200 the T-head, Android 10, we do the initial port for two chips, a test chip, two core, 06:41.200 --> 06:47.360 and then, when they have, a T-H-1, V-20, there's four core, so, then, put it again, Android 12, 06:48.000 --> 06:54.960 and then, and then, Android 14, then, we do the RVV 0.1 special optimization for the VMs, 06:56.720 --> 07:09.600 so, Android 16, fucking kick-up, so, people in West, the chipset guide in West, nearly $1 million, 07:10.560 --> 07:18.320 for three times, couldn't meant to produce it, the reality is set, it's a extremely 07:18.320 --> 07:24.960 lack of resources for this five, where we set, so, you look at the arm side, right, we know that 07:24.960 --> 07:29.760 we're going to arm side, how many, how many software engineers they put them into support Android, 07:30.320 --> 07:38.560 just arm, right, thousand guys, and then, SOC guide, core comes, media tech, 07:40.480 --> 07:52.160 okay, how many guys, hundreds, and then put that side, Xiaomi OPPO Vivo, blah, blah, blah, blah, 07:52.160 --> 08:00.720 all those guys making a thousand armies, right, so, for this one, the problem we lack of time, 08:00.720 --> 08:09.120 sci-fi sci-fi, all most of them now, the SOC I use, they use sci-fi sci-fi, so, sci-fi is not enough 08:09.200 --> 08:16.640 resources, and stuff I, I run the first four course from stuff I use in sci-fi sci-fi, so, all 08:16.640 --> 08:23.600 the stuff I was running time, and the yeast wing, my AIPC, same, they were all now, and the space 08:23.600 --> 08:31.760 misses the youngest, that's the 16 core coming up, right, so, that's already four years, so, 08:32.720 --> 08:40.400 if you don't do Android, not picking even, puto, right, you can hide about night to 08:40.400 --> 08:45.440 the investor for three years, five years, one day you don't break even, you know where you 08:45.440 --> 08:52.400 you can do it, so, it's where it's at, right, so what's the challenges for the hardware side, 08:52.400 --> 08:57.120 people going through the arm usually go through the arm side course, I'm done a lot of things, 08:57.760 --> 09:05.280 hit them thing you never know, right, hidden, but you will sometimes, you will find arms disappear, 09:05.280 --> 09:12.480 I bet most of them are SOC guys, just go take some, right, the IPs, I GPUs, I, all the, the whole 09:12.480 --> 09:19.040 software stack, potting, verification, it's most eaten by arm, but this fight, there's not even one I 09:19.040 --> 09:28.240 P guy on GPU guy match up the resources, so, just not enough to verify, and then the SOC is 09:28.240 --> 09:35.840 compared to SOC, before all the multimedia, told that, you know, display and all those 09:35.840 --> 09:42.400 compensator, and then press all the, what it do, be, bra, bra, bra, bra, bra, bra, 09:42.480 --> 09:49.840 wow, we're complicated, right, so past V5 rule, 2015, the endis, like the whole stack, and so 09:49.840 --> 09:57.600 far as I, Google that does claim, it's supposed it, but because it's chicken and net, we're 09:57.600 --> 10:05.280 reaching into chicken and net, so it's supposed in catfish, but it's just, it's very similar 10:05.280 --> 10:13.720 it or not. Nothing's rare for the two-chain, the kernel, the listing. No one knows the quality 10:13.720 --> 10:23.520 of it. No one. And then, same. This side kernel as well. So this side kernel will be 10:23.520 --> 10:29.600 the SOC, put it there, and then, because Google detaches the Android kernel versions. So 10:29.600 --> 10:36.120 we had to do all the part-porting, blah, blah, blah, blah, blah, blah, nothing. It's verified 10:36.120 --> 10:43.520 the quality. And then, never the mess. The whole GPU stack. We use imagination, because 10:43.520 --> 10:50.120 it's in no way you can use money. So we use imagination. Imagine this, do have some 10:50.120 --> 10:55.640 put up on the empty case side or on the other Chinese side, the U.D. stocks, or this, they 10:55.640 --> 11:01.400 have a certain IP that can do mass production of Android devices. But still, for research, 11:01.400 --> 11:08.680 no one rare for the whole stack. And most of the SOC guy, they usually, they don't ask for the 11:08.680 --> 11:15.280 Android DDK, or whatever DDK from me, imaginations. So usually, we had to, if we want to do, 11:15.280 --> 11:22.000 after that, we want to do the Android port. We had to do a lot of reverse engineer. So that 11:22.080 --> 11:30.560 is the, and nevertheless, the whole capacity of the multimedia and Wi-Fi Bluetooth, just 11:30.560 --> 11:41.240 shit not so work. Right. And attitude side, even worse, right. We have, my low windows, 11:41.240 --> 11:49.480 the vis-ficed circle. They do their port on Android, but they never share. With nimbitip 11:49.560 --> 11:56.920 resources, you do some work, but you don't share. So everyone is building a toy. It's 11:56.920 --> 12:03.440 not, it's not a main for match production. And then my solution to it, I can be fed up, 12:03.440 --> 12:12.120 so. And then I say that we need a, I could not solve the GPU issues as a hardware manufacturer. 12:12.120 --> 12:19.680 So, but I can solve it with an external GPU. So, and that's why I start with it. Now, 12:19.680 --> 12:25.920 I put in all the devices, for example, and then with all everything open source and reference, 12:25.920 --> 12:31.160 reference, and putting guy, net people putting their device, but with external GPU support. 12:31.160 --> 12:37.640 Just same as Intel. Same as Intel, you have an Intel laptop, you can run Android. That's 12:37.640 --> 12:44.640 the same idea. So, let me get the vis-fi AOS piece, it's kept to a quality first. So, 12:44.640 --> 12:51.000 that it moved the kernel. And then after that, then I was trying to solve the SOC-specific 12:51.000 --> 13:02.840 multimedia, right. And the multimedia, and the IO Wi-Fi pull to all this making, as a complete 13:02.840 --> 13:09.200 baseline, have a reference, portable, or whatever further complete baseline. And then after 13:09.200 --> 13:18.000 that, then I can verify the kernel, whatever we do is cool enough. Because the SOC-Guy is 13:18.000 --> 13:26.280 so, so tie up with the Linux kernel, LTS. Why usually Android kernel, you need to either 13:26.280 --> 13:34.840 back port to their specific versions. So, from the latest LTS. So, that is the thing that 13:34.840 --> 13:42.520 we need to, after step-by-step, then we can stabilize the whole thing. And then, one 13:42.520 --> 13:47.480 last thing we think about the whole integrated GPU, how we're going to do it. So, basically, 13:47.480 --> 13:53.560 at that time, I will, after we got the first two thing done, then I will negotiate with 13:53.560 --> 14:00.560 the SOC-Guy to ask imagination. See whether you can support the whole graphics step-by-step. 14:00.560 --> 14:05.440 Without the first two, even though imagination gives you a DDK, we, we see in that one 14:05.440 --> 14:10.240 chip on, you just crush everywhere. You, you know idea where to crush. You don't know 14:10.240 --> 14:16.440 where to crush it in the, at the baseline or the kernel and the GPU. It's just not a mystery, 14:17.440 --> 14:26.280 right. So, and then, last but not least, that, we, we decided to open source it out and 14:26.280 --> 14:32.640 then I forced all the guys in response. So, I open source, they have nothing to hide, right? 14:32.640 --> 14:38.440 They have nowhere to anymore, they have to keep my stuff. Because they start there, 14:38.440 --> 14:43.520 there, there, there, there, the stuff can explain to the boss. You need to do your own 14:43.520 --> 14:48.880 support again, right? You don't open it, I open it. So, you, you know, your boss can 14:48.880 --> 14:54.480 dip someone, open it, why you do it yourself, right? That's the perfect example of open 14:54.480 --> 15:00.960 source. And the, but the, the one thing AOSB, one thing is, 15:00.960 --> 15:06.000 Bradbury that I think Chris knows about it, right? When you, when you, when you do 15:06.080 --> 15:15.000 or a AOSB feature pot, where the Google accepted it or not, right? It's hard, right? 15:15.000 --> 15:23.320 So, then then I had to fork it and have to sing it if you want to keep up with it. 15:23.320 --> 15:31.440 And that's, that's our, the whole Android port there. So, if we specify, you guys want 15:31.440 --> 15:34.840 to have a port, learn to port the whole thing, 15:34.840 --> 15:40.840 then keep looking on it, and then I warn you guys to port. 15:40.840 --> 15:45.080 And I've been saying that I come down here for the reason 15:45.080 --> 15:48.120 is I warn you guys to support me. 15:48.120 --> 15:51.920 Do a help me to do the walk, and then I can give the hardware 15:51.920 --> 15:53.000 for free. 15:53.000 --> 15:55.520 There's no problem. 15:55.520 --> 16:00.760 And we have all the sponsorship program, and one 16:00.760 --> 16:05.080 basically, if we slide continue like that, 16:05.080 --> 16:07.760 they don't scale up the SOC ties at the end. 16:07.760 --> 16:12.200 I don't have the chip, so I can't make the advice anymore. 16:12.200 --> 16:15.360 So we need to solve Android. 16:15.360 --> 16:18.960 And I do not know Google when they're going to close the source up. 16:18.960 --> 16:20.600 I don't know. 16:20.600 --> 16:28.120 But one day, but hopefully we could enough to carry on from there. 16:28.120 --> 16:32.640 And that's all for me, and there's a lot of work. 16:32.640 --> 16:34.720 Really, I know more. 16:34.720 --> 16:41.920 And now I can, if you guys want to know more about WISFI, 16:41.920 --> 16:45.120 come down to our WISFI booth in the National booth 16:45.120 --> 16:48.020 and the AWS building, and also, 16:48.020 --> 16:54.360 free free to connect me on LinkedIn or whatever, hands up. 16:54.360 --> 16:57.200 I want to support you, give me the hardware, 16:57.200 --> 17:00.680 just run the email, subscribe to the Support Support Program, 17:00.680 --> 17:01.760 and help us. 17:01.760 --> 17:07.360 And then, I'm sure that once we get over this hurdle, 17:07.360 --> 17:12.880 and this FI, just exactly the same as NINES, 17:12.880 --> 17:17.040 we'll be launched to everyone, and then make the open source 17:17.040 --> 17:18.960 for the launch of everyone. 17:18.960 --> 17:24.160 Not just our country, not just the continent, right? 17:24.160 --> 17:32.920 That list of my NASCARIA on WISFI, something really open. 17:32.920 --> 17:36.640 No one can shut any body up. 17:36.640 --> 17:38.640 Thank you. 17:38.640 --> 17:49.920 So we have, like, five minutes for questions. 17:49.920 --> 17:53.800 If anybody has any questions, they're feeling anything 17:53.800 --> 17:55.600 is, since it's only one microphone. 17:55.600 --> 17:56.800 Can you repeat this question? 17:56.800 --> 17:58.040 So it gets recorded. 17:58.040 --> 17:59.640 OK. 17:59.640 --> 18:06.640 So the question you're asking is that, because we call it, 18:06.640 --> 18:09.640 we have a lot of envelope on the app side. 18:09.640 --> 18:15.640 So, and you will see that we do have, 18:15.640 --> 18:19.640 um, to RISFI emulator. 18:19.640 --> 18:23.640 That is part of the deeper. 18:23.640 --> 18:26.640 I will say it's medium-term application side. 18:26.640 --> 18:29.640 For us, if we do the Android app, 18:29.640 --> 18:31.640 we will have a lot of data on the app side. 18:31.640 --> 18:35.640 So the question you're asking is that, because you have, 18:35.640 --> 18:40.640 we call it, we have a lot of envelope on the app side. 18:40.640 --> 18:43.640 I will say it's medium-term application side. 18:43.640 --> 18:46.640 For us, if we do the Android, we will probably focus 18:46.640 --> 18:50.640 on certain market that do not have those kind of apps. 18:50.640 --> 18:55.640 And for example, most of the apps on frutter, 18:55.640 --> 18:59.640 dot, those are when, probably when, perfectly fine. 18:59.640 --> 19:02.640 And for those apps that, you know, lie in, 19:02.640 --> 19:05.640 in public in the Far East, or probably, I think, 19:05.640 --> 19:09.640 even in the West, they have lot of so-called encrypted app. 19:10.640 --> 19:12.640 And those are when the hard. 19:12.640 --> 19:15.640 And, and, and, and I can see, the Android is very 19:15.640 --> 19:18.640 recommended compared to Apple. 19:18.640 --> 19:21.640 Apple's Apple's Pickle Base. 19:21.640 --> 19:23.640 So, when you download the app, 19:23.640 --> 19:25.640 the server will be compiled to, 19:25.640 --> 19:27.640 whatever the target, or you over your app 19:27.640 --> 19:29.640 onto your, your mobile phone. 19:29.640 --> 19:31.640 That's, that's how Apple does it. 19:31.640 --> 19:35.640 But, you know, and, and Android, 19:35.640 --> 19:37.640 probably Google, don't do that, 19:37.640 --> 19:40.640 because the, or the market is so recommended. 19:40.640 --> 19:42.640 So, we, we do have, um, 19:42.640 --> 19:44.640 um, to this file translator. 19:44.640 --> 19:46.640 At the moment, we can see the, 19:46.640 --> 19:48.640 the binary translation is around, 19:48.640 --> 19:50.640 um, uh, two types of the performance. 19:50.640 --> 19:53.640 We lose around 30%. 19:53.640 --> 19:56.640 I think there's no, no technology barrier, 19:56.640 --> 19:59.640 because we'll set up, done it already. 19:59.640 --> 20:01.640 On their upside. 20:01.640 --> 20:03.640 Thank you. 20:04.640 --> 20:06.640 Uh, it's up to a lot of, uh, 20:06.640 --> 20:08.640 I'm very fired, uh, 20:08.640 --> 20:09.640 uh, I'm tested basically. 20:09.640 --> 20:10.640 Yep. 20:10.640 --> 20:16.640 Yeah, basically, is that, uh, 20:16.640 --> 20:18.640 what is and verify means, 20:18.640 --> 20:21.640 yep. 20:21.640 --> 20:23.640 Software side basically. 20:23.640 --> 20:26.640 Now, Google makes an Amazon, uh, 20:26.640 --> 20:29.640 AOSP runs on emulators. 20:29.640 --> 20:32.640 But, so, just repeat the question, 20:32.640 --> 20:35.640 basically, how, uh, what is and verify base 9. 20:35.640 --> 20:37.640 And what, what, and verify base 9, 20:37.640 --> 20:40.640 they never go through the whole hardware cycles. 20:40.640 --> 20:44.640 So, it's a purely software emulation on the, 20:44.640 --> 20:48.640 the cumul, or Android Studio. 20:48.640 --> 20:50.640 That's it. 20:50.640 --> 20:53.640 So, they never go through, uh, 20:53.640 --> 20:57.640 any of the, uh, hardware, 20:57.640 --> 21:00.640 at all, even the death ball. 21:00.640 --> 21:03.640 So, the device, which passes CTS on this side. 21:03.640 --> 21:06.640 No, no, so far. 21:06.640 --> 21:09.640 None hardware. 21:09.640 --> 21:12.640 The cumul, my, the Android Studio, my pass, 21:12.640 --> 21:14.640 because underlying the upstream, 21:14.640 --> 21:16.640 the upstream is X86 stuff. 21:16.640 --> 21:17.640 Right. 21:17.640 --> 21:19.640 So, they, they, they got it. 21:19.640 --> 21:20.640 And that's why, 21:20.640 --> 21:23.640 yeah, basically, it's unknown to many. 21:23.640 --> 21:25.640 That's why, uh, 21:25.640 --> 21:27.640 you haven't gone through the market, 21:27.640 --> 21:29.640 and, uh, whole mass production experience, 21:29.640 --> 21:31.640 is none of them have a quality. 21:31.640 --> 21:33.640 All we're doing is just, for example, 21:33.640 --> 21:35.640 for the past five years, the port I've done is, 21:35.640 --> 21:37.640 I would say that it's just, uh, 21:37.640 --> 21:39.640 a port of time. 21:39.640 --> 21:41.640 A demo. 21:41.640 --> 21:45.640 We'll ever reach to our mass scale productions. 21:45.640 --> 21:47.640 Thank you for your questions. 21:48.640 --> 21:51.640 What else? 21:51.640 --> 21:59.640 Okay. So, thank you all for joining for that presentation. 21:59.640 --> 22:05.640 And we'll be in short force while we switch over to the next speaker. 22:05.640 --> 22:06.640 Thank you. 22:06.640 --> 22:08.640 Thank you.