0 00:00:00,000 --> 00:00:30,000 Dear viewer, these subtitles were generated by a machine via the service Trint and therefore are (very) buggy. If you are capable, please help us to create good quality subtitles: https://c3subtitles.de/talk/938 Thanks! 1 00:00:14,870 --> 00:00:17,219 With me on stage, it's a wonderful person 2 00:00:17,220 --> 00:00:19,339 who will talk to us about CPU 3 00:00:19,340 --> 00:00:21,619 development, power, 4 00:00:21,620 --> 00:00:23,869 open source and 5 00:00:23,870 --> 00:00:26,749 how this all fits together and how this 6 00:00:26,750 --> 00:00:28,909 will hopefully influence of future 7 00:00:28,910 --> 00:00:31,549 CPU architecture and development. 8 00:00:31,550 --> 00:00:33,679 So please give a very warm 9 00:00:33,680 --> 00:00:35,899 round of applause to Matthew Michel. 10 00:00:41,370 --> 00:00:42,539 OK. 11 00:00:42,540 --> 00:00:44,369 Thanks for having me here today. 12 00:00:44,370 --> 00:00:46,589 And I apologize at the beginning for my 13 00:00:46,590 --> 00:00:49,259 strong German accent mixture of German 14 00:00:49,260 --> 00:00:52,049 and English. So sorry for that 15 00:00:52,050 --> 00:00:55,259 because I'm coming from 16 00:00:55,260 --> 00:00:57,419 second year actually living in Kemet, 17 00:00:57,420 --> 00:00:59,129 still born in chemist's, grew up in 18 00:00:59,130 --> 00:01:01,319 Kemet, so starts whoever 19 00:01:01,320 --> 00:01:02,549 knows about that. 20 00:01:02,550 --> 00:01:04,708 Um, that's why the accent is still a 21 00:01:04,709 --> 00:01:06,959 mixture of sex and German, but still 22 00:01:06,960 --> 00:01:09,149 want to show that are still smart people 23 00:01:09,150 --> 00:01:11,399 living in Saxony, not just 24 00:01:11,400 --> 00:01:13,469 people in Western waving 25 00:01:13,470 --> 00:01:15,689 some strange flecks all the time. 26 00:01:15,690 --> 00:01:16,739 Um, I want to say 27 00:01:17,970 --> 00:01:18,970 thanks. 28 00:01:19,980 --> 00:01:22,049 I just want to give a preproduction 29 00:01:22,050 --> 00:01:24,629 today about how I got inspired. 30 00:01:24,630 --> 00:01:26,909 I think two years ago by a talk by David 31 00:01:26,910 --> 00:01:29,609 Caplan, who was talking about Sebou 32 00:01:29,610 --> 00:01:31,799 development altogether from 33 00:01:31,800 --> 00:01:33,839 scratch through the final product and all 34 00:01:33,840 --> 00:01:35,369 these steps for it. 35 00:01:35,370 --> 00:01:37,499 I also want to touch that briefly 36 00:01:37,500 --> 00:01:39,749 today, but I want to go more 37 00:01:39,750 --> 00:01:42,359 into open power and 38 00:01:42,360 --> 00:01:43,919 just tell you something about this 39 00:01:43,920 --> 00:01:45,689 because I'm not really sure if everybody 40 00:01:45,690 --> 00:01:46,649 knows about that. 41 00:01:46,650 --> 00:01:49,049 Uh, yeah, foundation. 42 00:01:49,050 --> 00:01:51,149 And um, might also 43 00:01:51,150 --> 00:01:53,309 be a little off topic in this conference 44 00:01:53,310 --> 00:01:54,809 here. But uh, maybe it's still of 45 00:01:54,810 --> 00:01:55,810 interest. 46 00:01:56,430 --> 00:01:58,499 Um, this is 47 00:01:58,500 --> 00:01:59,409 my agenda for the day. 48 00:01:59,410 --> 00:02:01,769 So first of all, uh, like to 49 00:02:01,770 --> 00:02:03,899 just talk a little bit of 50 00:02:03,900 --> 00:02:06,329 Barbie then uh, as mentioned 51 00:02:06,330 --> 00:02:08,429 about this, if you would open flow at all 52 00:02:08,430 --> 00:02:11,699 and y or power, uh, 53 00:02:11,700 --> 00:02:13,889 myself, as I mentioned, um, 54 00:02:13,890 --> 00:02:14,849 still living in cabinets. 55 00:02:14,850 --> 00:02:17,069 But I'm um, my desk is located 56 00:02:17,070 --> 00:02:18,899 here. This is Steve bobbling of IBM 57 00:02:18,900 --> 00:02:21,059 Germany. Um, I'm there 58 00:02:21,060 --> 00:02:23,219 in the verification department of 59 00:02:23,220 --> 00:02:24,419 CPU development. 60 00:02:24,420 --> 00:02:26,759 Um, it's not just located there, 61 00:02:26,760 --> 00:02:29,009 it's uh, built by a distributed team, 62 00:02:29,010 --> 00:02:30,019 virtual teams. 63 00:02:30,020 --> 00:02:32,129 Um, I'm looking to get to with people in 64 00:02:32,130 --> 00:02:34,799 Bangalore, US, uh, France 65 00:02:34,800 --> 00:02:36,869 and so on. So will you? 66 00:02:36,870 --> 00:02:38,639 Uh, yeah, all the candidates are 67 00:02:38,640 --> 00:02:39,640 included. 68 00:02:40,290 --> 00:02:42,839 Um, my main focus is verification. 69 00:02:42,840 --> 00:02:44,969 Um, I'm also not an expert of a 70 00:02:44,970 --> 00:02:47,369 power. Um, but I just still 71 00:02:47,370 --> 00:02:49,289 involved in some topics of it. 72 00:02:49,290 --> 00:02:51,569 Uh, that's why I think I can at least 73 00:02:51,570 --> 00:02:53,789 talk about some steps to you, 74 00:02:53,790 --> 00:02:55,799 and that's all about me. 75 00:02:55,800 --> 00:02:58,559 Let's go into the technical 76 00:02:58,560 --> 00:02:59,699 agenda. 77 00:02:59,700 --> 00:03:01,829 Um, this is the CPU development 78 00:03:01,830 --> 00:03:04,319 for s we are still looking at today 79 00:03:04,320 --> 00:03:05,219 at IBM. 80 00:03:05,220 --> 00:03:07,619 Um, so we basically it's um, 81 00:03:07,620 --> 00:03:09,689 you know, a simple overview. 82 00:03:09,690 --> 00:03:11,849 You start at the left, 83 00:03:11,850 --> 00:03:14,039 uh, by either coding or 84 00:03:14,040 --> 00:03:16,349 generating or somehow specifying 85 00:03:16,350 --> 00:03:17,639 some hardware description. 86 00:03:17,640 --> 00:03:19,709 Um, maybe it should 87 00:03:19,710 --> 00:03:22,139 maybe be a log or whatever you think 88 00:03:22,140 --> 00:03:23,129 might work. 89 00:03:23,130 --> 00:03:25,739 Um, this is also, um, 90 00:03:25,740 --> 00:03:28,079 this, uh uh, large 91 00:03:28,080 --> 00:03:29,249 arrow pointing to it. 92 00:03:29,250 --> 00:03:31,589 Um, this also I'm using usually for 93 00:03:31,590 --> 00:03:34,499 verification. Um, there are also multiple 94 00:03:34,500 --> 00:03:36,569 different versions of each year that are 95 00:03:36,570 --> 00:03:37,589 possible. 96 00:03:37,590 --> 00:03:40,019 But just assume it's basically one 97 00:03:40,020 --> 00:03:42,509 thing that's busy specifying 98 00:03:42,510 --> 00:03:44,009 the hardware how it should work. 99 00:03:44,010 --> 00:03:46,199 And uh, yeah, there are 100 00:03:46,200 --> 00:03:48,029 some kind of two or three that body 101 00:03:48,030 --> 00:03:50,159 converts. Um, all the different 102 00:03:50,160 --> 00:03:52,469 things to the final product, um. 103 00:03:52,470 --> 00:03:54,569 Right? Um, so out 104 00:03:54,570 --> 00:03:56,879 of the HL, uh, you will get the 105 00:03:56,880 --> 00:03:58,499 list with all those logic components 106 00:03:58,500 --> 00:03:59,399 being generated. 107 00:03:59,400 --> 00:04:01,709 Uh, that also some other specific 108 00:04:01,710 --> 00:04:03,809 application teams working on that one. 109 00:04:03,810 --> 00:04:05,969 Um, then you can generate the 110 00:04:05,970 --> 00:04:06,929 schematics out of it. 111 00:04:06,930 --> 00:04:08,999 Uh, yeah, you will see 112 00:04:09,000 --> 00:04:11,249 some resistors in there. 113 00:04:11,250 --> 00:04:13,049 Um, yeah, maybe. 114 00:04:13,050 --> 00:04:15,629 Uh, forming those transistors into layout 115 00:04:15,630 --> 00:04:17,699 and layout, you basically get to 116 00:04:17,700 --> 00:04:20,039 define a product at the end of 117 00:04:20,040 --> 00:04:21,539 the application process. 118 00:04:21,540 --> 00:04:23,759 Um, I will also have some numbers 119 00:04:23,760 --> 00:04:26,099 later today, but uh, assume this process. 120 00:04:26,100 --> 00:04:28,259 Uh, you can go through 121 00:04:28,260 --> 00:04:30,959 hell on a daily basis, but uh, 122 00:04:30,960 --> 00:04:33,119 from a first line of uh, 123 00:04:33,120 --> 00:04:35,369 yes, code until the final product. 124 00:04:35,370 --> 00:04:37,829 Uh, I just assume like two or four years 125 00:04:37,830 --> 00:04:39,299 until you have a real working product. 126 00:04:41,560 --> 00:04:43,869 Um, now there's some 127 00:04:43,870 --> 00:04:46,389 nasty, uh, uh, flowchart 128 00:04:46,390 --> 00:04:48,499 from some of my colleagues, but somehow 129 00:04:48,500 --> 00:04:50,739 it just should illustrate how the 130 00:04:50,740 --> 00:04:52,659 overall flow is somehow working. 131 00:04:52,660 --> 00:04:55,239 So you just saw the diversion 132 00:04:55,240 --> 00:04:57,249 coming from, uh, hot weather scripting 133 00:04:57,250 --> 00:04:58,659 language into the product. 134 00:04:58,660 --> 00:05:00,729 Now this is the workflow inside 135 00:05:00,730 --> 00:05:02,409 the company itself. You have basically 136 00:05:02,410 --> 00:05:03,369 some kind of product. 137 00:05:03,370 --> 00:05:05,619 Um, you believe CEO 138 00:05:05,620 --> 00:05:08,049 targets or the previous version. 139 00:05:08,050 --> 00:05:10,269 You have some customer that somehow 140 00:05:10,270 --> 00:05:12,369 has some need for certain things to 141 00:05:12,370 --> 00:05:14,589 run on that and typically gives some 142 00:05:14,590 --> 00:05:15,789 kind of specification. 143 00:05:15,790 --> 00:05:17,529 You have some teams working on high level 144 00:05:17,530 --> 00:05:20,019 design and then this is basically your 145 00:05:20,020 --> 00:05:22,179 input source. Uh, sometimes very 146 00:05:22,180 --> 00:05:24,279 specific, sometimes very loose. 147 00:05:24,280 --> 00:05:26,019 Um, and then you have different teams 148 00:05:26,020 --> 00:05:28,149 working in parallel, uh, trying to to get 149 00:05:28,150 --> 00:05:29,829 this product put together. 150 00:05:29,830 --> 00:05:31,809 So we have some kind of system component 151 00:05:31,810 --> 00:05:34,089 teams that are looking at the boards 152 00:05:34,090 --> 00:05:36,669 and all the electrical components on it. 153 00:05:36,670 --> 00:05:38,049 You have a team that's looking at this 154 00:05:38,050 --> 00:05:39,639 black box. Uh, we gotta to put it at the 155 00:05:39,640 --> 00:05:41,769 end, you have premier teams, you have 156 00:05:41,770 --> 00:05:43,539 basically a design team that writing that 157 00:05:43,540 --> 00:05:45,639 visual code or very low code and then you 158 00:05:45,640 --> 00:05:47,499 have basically at the end, uh, 159 00:05:47,500 --> 00:05:49,359 verification team that is busy making 160 00:05:49,360 --> 00:05:51,429 sure that this also 161 00:05:51,430 --> 00:05:53,509 works at the end before you go through 162 00:05:53,510 --> 00:05:55,599 it and get it produced. 163 00:05:55,600 --> 00:05:57,729 And then you have also other teams 164 00:05:57,730 --> 00:06:00,039 looking at timing and looking at circuit 165 00:06:00,040 --> 00:06:01,149 design integration. 166 00:06:01,150 --> 00:06:03,549 So that's all going into physics, 167 00:06:03,550 --> 00:06:04,929 which is really important that the 168 00:06:04,930 --> 00:06:06,489 process design, because you want to have 169 00:06:06,490 --> 00:06:08,169 the maximum performance coming out of 170 00:06:08,170 --> 00:06:09,170 your product. 171 00:06:10,440 --> 00:06:12,629 Um, yeah, when you're done Besley, 172 00:06:12,630 --> 00:06:14,789 you push the button, get it fabricated. 173 00:06:14,790 --> 00:06:16,829 Then there are some tests going on. 174 00:06:16,830 --> 00:06:19,139 And if that goes well, then you have 175 00:06:19,140 --> 00:06:20,559 a good machine. 176 00:06:20,560 --> 00:06:22,709 Um, sometimes you 177 00:06:22,710 --> 00:06:23,939 find bugs. 178 00:06:23,940 --> 00:06:26,159 That's why this small red 179 00:06:26,160 --> 00:06:27,719 bellows is coming in. 180 00:06:27,720 --> 00:06:29,819 Um, if you find bugs, you 181 00:06:29,820 --> 00:06:31,349 basically want to find them as early as 182 00:06:31,350 --> 00:06:33,929 possible to basically 183 00:06:33,930 --> 00:06:35,939 turn and turn around your logic until it 184 00:06:35,940 --> 00:06:37,939 really works before you get to the FAP. 185 00:06:37,940 --> 00:06:40,019 Um, if you missed that point, um, do 186 00:06:40,020 --> 00:06:41,549 you have to spend much more dollars on 187 00:06:41,550 --> 00:06:44,009 it? And then you, yeah, um, 188 00:06:44,010 --> 00:06:45,899 have to spend multiple cycles on that. 189 00:06:45,900 --> 00:06:47,999 Um, just be sure you don't want to be 190 00:06:48,000 --> 00:06:50,219 that person being responsible for having 191 00:06:50,220 --> 00:06:52,379 to respond, uh, through the FAP, because 192 00:06:52,380 --> 00:06:54,779 that process going to that is 193 00:06:54,780 --> 00:06:56,249 like two months, 12 months. 194 00:06:56,250 --> 00:06:58,349 Uh, and you don't want to be 195 00:06:58,350 --> 00:07:00,149 that person, first of all, spending the 196 00:07:00,150 --> 00:07:01,889 money and also causing the delay for the 197 00:07:01,890 --> 00:07:03,599 customers at the end. 198 00:07:03,600 --> 00:07:05,549 And if you are a really bad person, then 199 00:07:05,550 --> 00:07:07,619 um, at the end you get into 200 00:07:07,620 --> 00:07:08,519 manufacturing. 201 00:07:08,520 --> 00:07:09,869 All these machines are being put to the 202 00:07:09,870 --> 00:07:11,939 customer sites, they push a button 203 00:07:11,940 --> 00:07:14,459 and nothing works. And um, yeah, 204 00:07:14,460 --> 00:07:15,809 let's don't talk about that. 205 00:07:15,810 --> 00:07:17,909 Um, but I didn't see that in my life so 206 00:07:17,910 --> 00:07:19,079 far. 207 00:07:19,080 --> 00:07:21,749 Um, this is not the picture 208 00:07:21,750 --> 00:07:23,459 somehow trying to illustrate the same 209 00:07:23,460 --> 00:07:24,910 thing again, but now 210 00:07:25,920 --> 00:07:28,199 trying to decide to the left more 211 00:07:28,200 --> 00:07:30,449 to design verification on the right side, 212 00:07:30,450 --> 00:07:33,419 more to, uh um, prevent 213 00:07:33,420 --> 00:07:34,979 physical stuff that is being done. 214 00:07:34,980 --> 00:07:36,629 So you see a physical science into this 215 00:07:36,630 --> 00:07:39,419 placement routing stuff that you 216 00:07:39,420 --> 00:07:40,559 know needs to be done. 217 00:07:40,560 --> 00:07:42,809 But I personally don't want to be 218 00:07:42,810 --> 00:07:43,709 that we are. 219 00:07:43,710 --> 00:07:45,989 It's always critical and you'll always 220 00:07:45,990 --> 00:07:47,749 be at the end of the process. 221 00:07:47,750 --> 00:07:49,589 Um, everybody's waiting for you to push 222 00:07:49,590 --> 00:07:50,789 the button. 223 00:07:50,790 --> 00:07:52,949 That's why I chose the left side of being 224 00:07:52,950 --> 00:07:54,489 willing to do verification front. 225 00:07:54,490 --> 00:07:56,709 Um, also dealing very close 226 00:07:56,710 --> 00:07:58,109 50 design teams. 227 00:07:58,110 --> 00:08:00,749 Um, and you see, it's it's two circles 228 00:08:00,750 --> 00:08:03,149 circling because you are always 229 00:08:03,150 --> 00:08:05,339 some hole in the same source, different 230 00:08:05,340 --> 00:08:07,739 times and that's basically what you do 231 00:08:07,740 --> 00:08:08,979 all the time. 232 00:08:08,980 --> 00:08:10,919 You are going into a kind of continuous 233 00:08:10,920 --> 00:08:11,920 integration. 234 00:08:13,590 --> 00:08:15,449 OK. And here are the numbers that I 235 00:08:15,450 --> 00:08:17,789 promised um, some would try to compare 236 00:08:17,790 --> 00:08:20,549 the numbers from 237 00:08:20,550 --> 00:08:22,799 building a larger ship 238 00:08:22,800 --> 00:08:24,689 compared to a processor. 239 00:08:24,690 --> 00:08:26,399 And that's quite interesting because 240 00:08:26,400 --> 00:08:28,799 somehow at the end, when comparing 241 00:08:28,800 --> 00:08:31,079 you for both things you 242 00:08:31,080 --> 00:08:33,509 need for the whole system, about 5000 243 00:08:33,510 --> 00:08:35,609 people. Um, also 244 00:08:35,610 --> 00:08:37,109 about the construction type. 245 00:08:37,110 --> 00:08:38,579 It's similar. 246 00:08:38,580 --> 00:08:40,288 So we're like three years with the ship, 247 00:08:40,289 --> 00:08:42,298 like a four to five years for the whole 248 00:08:42,299 --> 00:08:43,979 protozoan system. 249 00:08:43,980 --> 00:08:45,929 It's not four to five years for just one 250 00:08:45,930 --> 00:08:48,149 shot. It's basically a multiple turns 251 00:08:48,150 --> 00:08:49,709 of the processor until the machine is 252 00:08:49,710 --> 00:08:51,690 basically ready to ship to the customer. 253 00:08:52,920 --> 00:08:54,979 And also, you see it 254 00:08:54,980 --> 00:08:57,089 because at the end, so it's 255 00:08:57,090 --> 00:08:58,739 more than one billion dollars that needs 256 00:08:58,740 --> 00:09:00,929 to be spent for one generation 257 00:09:00,930 --> 00:09:01,930 of those systems. 258 00:09:02,970 --> 00:09:04,979 Just to mention, there are two 259 00:09:04,980 --> 00:09:06,779 architectures currently still being 260 00:09:06,780 --> 00:09:08,609 worked on at IBM. 261 00:09:08,610 --> 00:09:09,959 It's the power architecture and the sea 262 00:09:09,960 --> 00:09:11,999 architecture. Those numbers are for this 263 00:09:12,000 --> 00:09:13,049 processor. 264 00:09:13,050 --> 00:09:15,569 But you can compare those 265 00:09:15,570 --> 00:09:17,309 pretty similar from a cost point of view. 266 00:09:19,050 --> 00:09:21,329 So that's already it 267 00:09:21,330 --> 00:09:23,729 from the um, CPU 268 00:09:23,730 --> 00:09:25,409 development side. 269 00:09:25,410 --> 00:09:28,259 So let's jump into Apollo. 270 00:09:28,260 --> 00:09:31,079 Um, and 271 00:09:31,080 --> 00:09:32,169 you are looking at that. 272 00:09:32,170 --> 00:09:34,319 Um, that's my 273 00:09:34,320 --> 00:09:36,569 view of white power because, 274 00:09:36,570 --> 00:09:38,539 um, sometimes it looks strange. 275 00:09:38,540 --> 00:09:40,649 Um, why should something 276 00:09:40,650 --> 00:09:42,599 change in that kind of popular stuff? 277 00:09:42,600 --> 00:09:43,739 It's always be there. 278 00:09:43,740 --> 00:09:45,449 Uh, so you see you look on your PC, you 279 00:09:45,450 --> 00:09:47,099 see somehow getting smaller, you have 280 00:09:47,100 --> 00:09:49,469 your phones, but somehow a 281 00:09:49,470 --> 00:09:51,599 similar kind of stuff 282 00:09:51,600 --> 00:09:53,759 in there, also from the server part 283 00:09:53,760 --> 00:09:55,979 of it. Um, it's somehow in some 284 00:09:55,980 --> 00:09:58,199 building in the cellar and uh, it's just 285 00:09:58,200 --> 00:10:00,359 work. Um, but there is some 286 00:10:00,360 --> 00:10:01,499 commercial component to it. 287 00:10:01,500 --> 00:10:03,719 Um, over years, this 288 00:10:03,720 --> 00:10:06,179 kind of process worked very well for 289 00:10:06,180 --> 00:10:08,189 players like IBM or others. 290 00:10:08,190 --> 00:10:10,589 Um, but all the time 291 00:10:10,590 --> 00:10:12,459 the market shifted drastically. 292 00:10:12,460 --> 00:10:14,699 Um, so you have a Moore's law, 293 00:10:14,700 --> 00:10:17,159 which basically, uh, provided over years, 294 00:10:17,160 --> 00:10:19,649 uh, a very constant flow of, uh, 295 00:10:19,650 --> 00:10:20,999 performance improvements. 296 00:10:21,000 --> 00:10:22,139 So we didn't have to do much. 297 00:10:22,140 --> 00:10:24,329 Uh, you just had to move to the next 298 00:10:24,330 --> 00:10:26,039 generation of your technology. 299 00:10:26,040 --> 00:10:28,169 And then surprisingly, you got, uh, ten 300 00:10:28,170 --> 00:10:30,059 twenty whatever percent of new 301 00:10:30,060 --> 00:10:31,649 performance or changing much in your 302 00:10:31,650 --> 00:10:32,829 design. 303 00:10:32,830 --> 00:10:34,229 Nevertheless, you added more features to 304 00:10:34,230 --> 00:10:36,659 it, uh, to satisfy customer needs, 305 00:10:36,660 --> 00:10:38,879 but that it's, uh, more more, 306 00:10:38,880 --> 00:10:40,889 uh, never true anymore. 307 00:10:40,890 --> 00:10:43,019 Um, also on the other side, you have, uh, 308 00:10:43,020 --> 00:10:44,729 other trends like, uh, new kind of 309 00:10:44,730 --> 00:10:45,719 workloads are coming up. 310 00:10:45,720 --> 00:10:48,059 Uh, just think about, uh, some 311 00:10:48,060 --> 00:10:50,129 kind of bitcoin mining, which is, uh, 312 00:10:50,130 --> 00:10:51,479 somehow trending today. 313 00:10:51,480 --> 00:10:53,609 Uh, or these days, but also a 314 00:10:53,610 --> 00:10:56,069 um, kind of, uh, consumption 315 00:10:56,070 --> 00:10:58,409 of, um, those server farms 316 00:10:58,410 --> 00:10:59,639 are changing. 317 00:10:59,640 --> 00:11:01,829 In previous years, it was like, uh, 318 00:11:01,830 --> 00:11:03,929 somebody from some company called the 319 00:11:03,930 --> 00:11:06,329 I need new machines and the customer. 320 00:11:06,330 --> 00:11:08,529 They moved there and, uh, yeah, uh, 321 00:11:08,530 --> 00:11:10,349 sold something of which was staying there 322 00:11:10,350 --> 00:11:11,789 for the next couple of years. 323 00:11:11,790 --> 00:11:13,889 Uh, but these days, um, cloud 324 00:11:13,890 --> 00:11:16,079 is something, uh, very 325 00:11:16,080 --> 00:11:18,329 drastically changing, uh, the operation 326 00:11:18,330 --> 00:11:21,479 model of our sellers at IBM. 327 00:11:21,480 --> 00:11:23,579 And that also was something 328 00:11:23,580 --> 00:11:25,649 that we felt because at the 329 00:11:25,650 --> 00:11:27,959 end, um, just less and less products 330 00:11:27,960 --> 00:11:30,029 were sold. Um, the cloud 331 00:11:30,030 --> 00:11:32,249 infrastructure was not up and running 332 00:11:32,250 --> 00:11:33,779 from our point of view because, uh, we 333 00:11:33,780 --> 00:11:35,669 didn't bet on that. 334 00:11:35,670 --> 00:11:37,799 Other companies like Amazon were very 335 00:11:37,800 --> 00:11:39,849 fast on bringing it up. 336 00:11:39,850 --> 00:11:42,119 Um, so as you can see today, uh, most 337 00:11:42,120 --> 00:11:44,159 of the cloud infrastructure is x86 based. 338 00:11:44,160 --> 00:11:46,259 And uh, I think it's hard to 339 00:11:46,260 --> 00:11:48,869 find, uh, power servers 340 00:11:48,870 --> 00:11:51,059 for the open public cloud 341 00:11:51,060 --> 00:11:53,189 providers or others 342 00:11:53,190 --> 00:11:54,359 that you have seen. 343 00:11:54,360 --> 00:11:56,789 Also, the software ecosystem changed 344 00:11:56,790 --> 00:11:59,069 a little so the UM two more 345 00:11:59,070 --> 00:12:01,169 open and mature open 346 00:12:01,170 --> 00:12:03,269 providers, um, that's 347 00:12:03,270 --> 00:12:04,270 one part of it. 348 00:12:05,220 --> 00:12:07,499 And on the other side, you have some kind 349 00:12:07,500 --> 00:12:10,469 of strategy changes that you 350 00:12:10,470 --> 00:12:12,959 need to do now. Um, but more to that 351 00:12:12,960 --> 00:12:13,960 later on. 352 00:12:15,090 --> 00:12:17,399 Um, just back to the, uh, to Moore's 353 00:12:17,400 --> 00:12:19,589 law. Um, on that 354 00:12:19,590 --> 00:12:21,419 chart, you see, uh, it's basically 355 00:12:21,420 --> 00:12:23,249 complete performance per dollar. 356 00:12:23,250 --> 00:12:25,439 That's something um, the customers 357 00:12:25,440 --> 00:12:27,569 look at, um, they they look at 358 00:12:27,570 --> 00:12:29,009 how much money they need to spend for 359 00:12:29,010 --> 00:12:29,909 which kind of performance. 360 00:12:29,910 --> 00:12:31,469 And that's the number they compare 361 00:12:31,470 --> 00:12:32,729 between the different providers. 362 00:12:32,730 --> 00:12:35,069 And then only if you can, uh, 363 00:12:35,070 --> 00:12:37,519 match or be below the 364 00:12:37,520 --> 00:12:39,569 the other companies that you are able to 365 00:12:39,570 --> 00:12:41,219 to sell stuff to them. 366 00:12:41,220 --> 00:12:43,349 Um, and as I mentioned before, that 367 00:12:43,350 --> 00:12:45,329 was working quite well, uh, for the past 368 00:12:45,330 --> 00:12:46,529 couple of years. But now obviously, 369 00:12:46,530 --> 00:12:48,239 Moore's law is slowing down and doesn't 370 00:12:48,240 --> 00:12:49,319 help and more. 371 00:12:49,320 --> 00:12:51,389 So what you are now trying to 372 00:12:51,390 --> 00:12:53,459 do is, um, you need to 373 00:12:53,460 --> 00:12:55,469 somewhat enhance the performance by other 374 00:12:55,470 --> 00:12:57,709 means. So what could it be? 375 00:12:57,710 --> 00:12:59,669 Um, so you could plug something to your 376 00:12:59,670 --> 00:13:01,679 system so you can't move everything onto 377 00:13:01,680 --> 00:13:03,689 the processor because you just can't 378 00:13:03,690 --> 00:13:05,849 fabricated so you you need 379 00:13:05,850 --> 00:13:07,979 to do something close to the post so you 380 00:13:07,980 --> 00:13:10,559 can plug some extra radios on it. 381 00:13:10,560 --> 00:13:11,819 You can somehow. 382 00:13:11,820 --> 00:13:13,919 On the storage systems to be faster, you 383 00:13:13,920 --> 00:13:15,659 can look at the I o. 384 00:13:15,660 --> 00:13:17,999 If you can move to a new generation there 385 00:13:18,000 --> 00:13:19,919 or you can look at the memories. 386 00:13:19,920 --> 00:13:22,239 If they can provide you something. 387 00:13:22,240 --> 00:13:24,509 Um, but yet everything 388 00:13:24,510 --> 00:13:26,699 comes together at a data point. 389 00:13:26,700 --> 00:13:29,039 So data is that thing that somehow 390 00:13:29,040 --> 00:13:31,169 is supposed to be the issue over 391 00:13:31,170 --> 00:13:33,509 the next couple of years because of 392 00:13:33,510 --> 00:13:35,429 different trends you have, in other 393 00:13:35,430 --> 00:13:37,499 things, providing 394 00:13:37,500 --> 00:13:39,569 much or sometimes too much data 395 00:13:39,570 --> 00:13:41,219 and you just somehow need to handle it. 396 00:13:41,220 --> 00:13:43,379 So every device is 397 00:13:43,380 --> 00:13:44,549 somehow creating data. 398 00:13:44,550 --> 00:13:46,859 At the end, you need to have some 399 00:13:46,860 --> 00:13:49,129 backbone working on it and basically 400 00:13:49,130 --> 00:13:51,179 use some computer centers that can deal 401 00:13:51,180 --> 00:13:53,729 with all those data and to see 402 00:13:53,730 --> 00:13:56,459 what kind of data the 403 00:13:56,460 --> 00:13:58,169 greatest of those charts basically 404 00:13:58,170 --> 00:13:59,970 envisioned coming from different fields. 405 00:14:01,750 --> 00:14:03,819 Um, coming to 406 00:14:03,820 --> 00:14:05,889 power, um, 407 00:14:05,890 --> 00:14:08,169 and typically, um, 408 00:14:08,170 --> 00:14:10,269 the Moore's law is something I've already 409 00:14:10,270 --> 00:14:12,189 talked about. But there will be two other 410 00:14:12,190 --> 00:14:14,679 things that forced this kind of creation 411 00:14:14,680 --> 00:14:17,649 of this open power community. 412 00:14:17,650 --> 00:14:19,089 The second thing was Google. 413 00:14:19,090 --> 00:14:21,219 Um, Google was basically known to be 414 00:14:21,220 --> 00:14:23,799 a large provider of something 415 00:14:23,800 --> 00:14:25,059 like search or other stuff. 416 00:14:25,060 --> 00:14:27,399 Um, and somehow 417 00:14:27,400 --> 00:14:30,069 they were not really happy with their x86 418 00:14:30,070 --> 00:14:32,319 infrastructure. So some of the big talks, 419 00:14:32,320 --> 00:14:34,479 um, in other companies 420 00:14:34,480 --> 00:14:36,909 and also with IBM, and they 421 00:14:36,910 --> 00:14:39,819 basically ran some stuff on the 422 00:14:39,820 --> 00:14:41,979 back then power aids machines and 423 00:14:41,980 --> 00:14:44,079 uh, surprised, uh, form 424 00:14:44,080 --> 00:14:45,999 different patterns that were surprisingly 425 00:14:46,000 --> 00:14:48,399 good. And um, on the other side, 426 00:14:48,400 --> 00:14:50,529 they also, um, basically were looking 427 00:14:50,530 --> 00:14:52,359 right in the eye of a monopoly of intel. 428 00:14:52,360 --> 00:14:54,279 Um, so that was the second thing. 429 00:14:54,280 --> 00:14:56,619 Uh, Google was not being afraid, 430 00:14:56,620 --> 00:14:58,809 was afraid of uh, and 431 00:14:58,810 --> 00:15:00,999 basically, um, seeing that this 432 00:15:01,000 --> 00:15:02,559 was creating a very costly 433 00:15:02,560 --> 00:15:04,869 infrastructure, basically being bound 434 00:15:04,870 --> 00:15:07,389 to a monopolist um and then 435 00:15:07,390 --> 00:15:09,609 not open to any movement, something else. 436 00:15:09,610 --> 00:15:11,799 That's why the better you felt to 437 00:15:11,800 --> 00:15:14,319 to do something else and also, um, 438 00:15:14,320 --> 00:15:16,629 at the end, um, I 439 00:15:16,630 --> 00:15:19,179 don't really know, but at the end they 440 00:15:19,180 --> 00:15:21,159 want to build their own, uh, computers, 441 00:15:21,160 --> 00:15:23,109 and that's why they did the search for 442 00:15:23,110 --> 00:15:24,159 some partnering. 443 00:15:24,160 --> 00:15:26,589 Um, and they didn't came to IBM and 444 00:15:26,590 --> 00:15:28,749 uh, yeah, looked at us if 445 00:15:28,750 --> 00:15:30,939 we can do something together as well as 446 00:15:30,940 --> 00:15:33,039 the third point is extend the 447 00:15:33,040 --> 00:15:34,509 reach of power architecture. 448 00:15:34,510 --> 00:15:36,639 Um, so as mentioned before, in 449 00:15:36,640 --> 00:15:39,009 the past there were PowerPC years 450 00:15:39,010 --> 00:15:40,959 architecture was somehow a little, uh, 451 00:15:40,960 --> 00:15:43,209 stronger before, um, 452 00:15:43,210 --> 00:15:45,369 how architecture is, uh, 453 00:15:45,370 --> 00:15:46,839 getting market share or gaining market 454 00:15:46,840 --> 00:15:48,309 share in the last couple of years. 455 00:15:48,310 --> 00:15:50,439 But the overall Unix market 456 00:15:50,440 --> 00:15:52,719 is decreasing that fast, that 457 00:15:52,720 --> 00:15:54,689 it's basically not really working out 458 00:15:54,690 --> 00:15:57,549 well for keeping that architecture 459 00:15:57,550 --> 00:15:58,599 just how it is. 460 00:15:58,600 --> 00:16:00,699 So you need to find other means to 461 00:16:00,700 --> 00:16:03,039 to strengthen your whole 462 00:16:03,040 --> 00:16:05,289 infrastructure and the architecture of 463 00:16:05,290 --> 00:16:07,689 those boundaries and see how 464 00:16:07,690 --> 00:16:09,979 you can work with other parties 465 00:16:09,980 --> 00:16:12,369 to get into other different markets. 466 00:16:14,570 --> 00:16:17,389 And that was basically the root cause 467 00:16:17,390 --> 00:16:19,549 to basically get this 468 00:16:19,550 --> 00:16:21,919 open power initiative being 469 00:16:21,920 --> 00:16:24,109 done in 2013. 470 00:16:24,110 --> 00:16:26,329 Um, if you look at the other part 471 00:16:26,330 --> 00:16:28,249 of this chart, it back then it looked 472 00:16:28,250 --> 00:16:30,319 like all those companies somehow losing 473 00:16:30,320 --> 00:16:31,340 market share of all 474 00:16:32,540 --> 00:16:34,309 right to come together and found 475 00:16:34,310 --> 00:16:35,609 something. 476 00:16:35,610 --> 00:16:38,029 But basically looking at this, 477 00:16:38,030 --> 00:16:40,219 you base, it somehow looks 478 00:16:40,220 --> 00:16:41,299 like a good move. 479 00:16:41,300 --> 00:16:43,429 And let me show you how that worked 480 00:16:43,430 --> 00:16:44,430 out. 481 00:16:45,630 --> 00:16:47,759 Um, the first steps that were done, 482 00:16:47,760 --> 00:16:49,979 um, were basically done 483 00:16:49,980 --> 00:16:52,349 by ABM, um, first of all, 484 00:16:52,350 --> 00:16:54,179 basically, uh, investing a lot of money 485 00:16:54,180 --> 00:16:56,399 into a Linux, uh, Linux, being able 486 00:16:56,400 --> 00:16:57,659 to run on power. 487 00:16:57,660 --> 00:16:59,729 Um, that was a complaint even done 488 00:16:59,730 --> 00:17:01,899 by engineers internally over 489 00:17:01,900 --> 00:17:04,259 the last decades, basically that 490 00:17:04,260 --> 00:17:05,939 even our own applications were not 491 00:17:05,940 --> 00:17:08,969 finding really good on our own machines 492 00:17:08,970 --> 00:17:11,098 because we built our own computer pool 493 00:17:11,099 --> 00:17:13,139 was, uh, mainly, uh, based on x86 494 00:17:13,140 --> 00:17:15,328 machines because we were 495 00:17:15,329 --> 00:17:16,499 selling them all before. 496 00:17:16,500 --> 00:17:18,809 Um, that was a strong complaint, 497 00:17:18,810 --> 00:17:20,699 but it didn't help, uh, to complain 498 00:17:20,700 --> 00:17:22,979 internally about, uh, the customers, 499 00:17:22,980 --> 00:17:25,049 uh, basically, uh, showed us that 500 00:17:25,050 --> 00:17:27,389 the machines we weren't really, uh, able 501 00:17:27,390 --> 00:17:29,069 to, uh, to beat the market. 502 00:17:29,070 --> 00:17:30,359 So there was some kind of mind shift 503 00:17:30,360 --> 00:17:32,789 going on. And uh, basically, uh, resulted 504 00:17:32,790 --> 00:17:34,499 in investing a lot of money in, 505 00:17:35,670 --> 00:17:37,859 uh, structuring the code around Linux 506 00:17:37,860 --> 00:17:40,019 and basically even the little engine big 507 00:17:40,020 --> 00:17:42,209 engine complications to sort 508 00:17:42,210 --> 00:17:44,189 them out and get them resolved at the 509 00:17:44,190 --> 00:17:45,449 end. 510 00:17:45,450 --> 00:17:47,519 The second thing is this kind of 511 00:17:47,520 --> 00:17:49,299 off source model for hardware. 512 00:17:49,300 --> 00:17:51,179 Um, I think I will talk about that in a 513 00:17:51,180 --> 00:17:52,379 minute. 514 00:17:52,380 --> 00:17:54,689 And basically, it's do you feel step 515 00:17:54,690 --> 00:17:56,879 to release, uh, lots of firmware 516 00:17:56,880 --> 00:17:59,529 code to really make it available? 517 00:17:59,530 --> 00:18:01,529 I myself was even surprised. 518 00:18:01,530 --> 00:18:03,989 Uh, sometimes I just, uh, googled 519 00:18:03,990 --> 00:18:06,269 my name and, uh, some kind of, uh, 520 00:18:06,270 --> 00:18:07,169 word. 521 00:18:07,170 --> 00:18:09,449 Um, and surprisingly, 522 00:18:09,450 --> 00:18:11,549 I, uh, one of the first five 523 00:18:11,550 --> 00:18:13,619 matches was, uh, just something that I 524 00:18:13,620 --> 00:18:15,489 wrote myself with a colleague. 525 00:18:15,490 --> 00:18:17,579 Um, so just try to to search for 526 00:18:17,580 --> 00:18:19,799 it. Uh, you should find it very easily 527 00:18:19,800 --> 00:18:20,880 on the net. No. 528 00:18:22,500 --> 00:18:24,629 Um, this is the mission statement, uh, 529 00:18:24,630 --> 00:18:26,819 I think nothing to really read through 530 00:18:26,820 --> 00:18:29,189 completely, but um, the 531 00:18:29,190 --> 00:18:31,259 idea behind this is it's not a 532 00:18:31,260 --> 00:18:33,539 new company to fire phone, but 533 00:18:33,540 --> 00:18:35,819 um, it's some kind of partnership 534 00:18:35,820 --> 00:18:37,769 or some some open community 535 00:18:38,880 --> 00:18:41,039 that, um, is to be installed 536 00:18:41,040 --> 00:18:43,289 to get people together that want 537 00:18:43,290 --> 00:18:45,299 to improve the power architecture, as 538 00:18:45,300 --> 00:18:47,219 well as to to work on the whole stack. 539 00:18:47,220 --> 00:18:49,469 So that includes basically the basics of 540 00:18:49,470 --> 00:18:51,509 the development, as well as the the 541 00:18:51,510 --> 00:18:53,129 software development flow and basically 542 00:18:53,130 --> 00:18:55,079 use, uh, same tooling use, same 543 00:18:55,080 --> 00:18:57,269 infrastructure, same 544 00:18:57,270 --> 00:18:58,270 boundaries. 545 00:18:59,930 --> 00:19:02,239 That's just a picture to illustrate that 546 00:19:02,240 --> 00:19:04,399 on the left. Um, so you have 547 00:19:04,400 --> 00:19:06,529 to have the component, um, that 548 00:19:06,530 --> 00:19:08,509 basically consists basically of, first of 549 00:19:08,510 --> 00:19:10,759 all, the process itself and uh, 550 00:19:10,760 --> 00:19:12,319 still the process that doesn't hurt much. 551 00:19:12,320 --> 00:19:14,569 You need to have, uh, some connection 552 00:19:14,570 --> 00:19:17,239 points and which is interfaces to 553 00:19:17,240 --> 00:19:18,979 accelerators and not just those 554 00:19:18,980 --> 00:19:20,959 interfaces, but have them rather open 555 00:19:20,960 --> 00:19:23,689 that other companies can also use them 556 00:19:23,690 --> 00:19:25,339 also roadmaps for both of it. 557 00:19:25,340 --> 00:19:26,389 Um, that's what I learned. 558 00:19:26,390 --> 00:19:28,699 Also, uh, very early in 559 00:19:28,700 --> 00:19:30,859 my work, um, that 560 00:19:30,860 --> 00:19:32,629 customers really want to see a roadmap 561 00:19:32,630 --> 00:19:34,339 over the next couple of years that they 562 00:19:34,340 --> 00:19:36,619 are, uh, we can be sure to invest 563 00:19:36,620 --> 00:19:38,779 in your product. And, uh, yeah, 564 00:19:38,780 --> 00:19:40,319 can be sure that this continuous over the 565 00:19:40,320 --> 00:19:42,589 next, uh, technology that technology 566 00:19:42,590 --> 00:19:44,659 cycles, uh, as well as, um, 567 00:19:44,660 --> 00:19:46,849 it is not just a of it to make 568 00:19:46,850 --> 00:19:48,889 sure that this is basically, uh, common 569 00:19:48,890 --> 00:19:51,079 and uh, being used by many, 570 00:19:51,080 --> 00:19:53,209 uh, different parties, um, and that it 571 00:19:53,210 --> 00:19:54,499 basically gets debunked. 572 00:19:54,500 --> 00:19:55,500 The. 573 00:19:56,800 --> 00:19:59,019 Um, and that basically gets 574 00:19:59,020 --> 00:20:01,779 us to the ecosystem as of today. 575 00:20:01,780 --> 00:20:03,579 So the top before there showed you have 576 00:20:03,580 --> 00:20:06,009 seen there were five parties, um, funding 577 00:20:06,010 --> 00:20:07,029 this community. 578 00:20:07,030 --> 00:20:09,099 I know you have a whole stack that is 579 00:20:09,100 --> 00:20:11,409 basically, uh, starting with the basics 580 00:20:11,410 --> 00:20:12,549 at the bottom. 581 00:20:12,550 --> 00:20:14,939 Uh, now with Infinium as well as 582 00:20:14,940 --> 00:20:17,229 included with the chips, the bolt 583 00:20:17,230 --> 00:20:18,309 manufacturers. 584 00:20:18,310 --> 00:20:20,559 I o storage companies. 585 00:20:20,560 --> 00:20:22,089 System integration companies. 586 00:20:22,090 --> 00:20:24,219 Also software, but 587 00:20:24,220 --> 00:20:26,409 as well. You have, uh, many research 588 00:20:26,410 --> 00:20:27,309 teams included. 589 00:20:27,310 --> 00:20:29,709 Um, and that's basically, uh, something 590 00:20:29,710 --> 00:20:31,809 that is, um, also a key 591 00:20:31,810 --> 00:20:33,999 interest of that group to not just do 592 00:20:34,000 --> 00:20:36,099 something with companies and, uh, try 593 00:20:36,100 --> 00:20:38,619 to sell stuff, but also to to keep 594 00:20:38,620 --> 00:20:41,139 people involved in research and 595 00:20:41,140 --> 00:20:43,479 trying to get this, uh, 596 00:20:43,480 --> 00:20:44,560 whole method enhanced 597 00:20:46,150 --> 00:20:48,309 to trust them up. Uh, like, 598 00:20:48,310 --> 00:20:50,679 uh, twenty members, uh, as of today, 599 00:20:50,680 --> 00:20:53,199 uh, in thirty two countries, um, 600 00:20:53,200 --> 00:20:55,299 you know, these days, I think 601 00:20:55,300 --> 00:20:57,179 most of that is still centered around the 602 00:20:57,180 --> 00:20:58,809 US because they are basically the 603 00:20:58,810 --> 00:21:00,849 headquarters of IBM and the main, uh, 604 00:21:00,850 --> 00:21:02,649 labs still located. 605 00:21:02,650 --> 00:21:04,389 And that's why also many of the 606 00:21:04,390 --> 00:21:06,639 universities, they're joined 607 00:21:06,640 --> 00:21:07,809 in those areas. 608 00:21:07,810 --> 00:21:10,089 Um, but still basically, um, 609 00:21:10,090 --> 00:21:12,219 try to get members 610 00:21:12,220 --> 00:21:13,449 over the whole stack. 611 00:21:13,450 --> 00:21:15,969 And even those partners, 612 00:21:15,970 --> 00:21:18,249 um, also started to 613 00:21:18,250 --> 00:21:20,199 bring out first products. 614 00:21:20,200 --> 00:21:22,479 Um, but I think 615 00:21:22,480 --> 00:21:24,609 I have one chart for that 616 00:21:24,610 --> 00:21:25,739 later as well. 617 00:21:27,700 --> 00:21:29,919 That's basically, um, 618 00:21:29,920 --> 00:21:32,079 just to illustrate also, um, 619 00:21:32,080 --> 00:21:34,179 how Opal is somehow Mitrice or the 620 00:21:34,180 --> 00:21:36,889 other initiative initiatives. 621 00:21:36,890 --> 00:21:39,039 So we have this open compute group that 622 00:21:39,040 --> 00:21:41,169 is, I think, more concentrating 623 00:21:41,170 --> 00:21:44,019 on mechanical electrical synchronization 624 00:21:44,020 --> 00:21:45,909 and specifications. 625 00:21:45,910 --> 00:21:48,009 Um, the Opal club somehow 626 00:21:48,010 --> 00:21:50,199 is trying to, uh, to fill that gap 627 00:21:50,200 --> 00:21:52,449 between whatever is coming 628 00:21:52,450 --> 00:21:54,819 from top from, uh, software. 629 00:21:54,820 --> 00:21:57,309 Linux, OpenStack, um 630 00:21:57,310 --> 00:21:59,529 and open compute at the bottom. 631 00:21:59,530 --> 00:22:01,749 Um, open capital is, uh, 632 00:22:01,750 --> 00:22:03,759 some a part of that open power 633 00:22:03,760 --> 00:22:06,039 initiative, but it's basically, uh, just 634 00:22:06,040 --> 00:22:07,479 one fact of it. 635 00:22:07,480 --> 00:22:09,489 And as you can see at the screen box in 636 00:22:09,490 --> 00:22:11,649 there, uh, limits itself to 637 00:22:11,650 --> 00:22:13,779 a protocols and uh, I own the 638 00:22:13,780 --> 00:22:16,209 connections internally. 639 00:22:16,210 --> 00:22:18,279 Um, so that's how it would fit 640 00:22:18,280 --> 00:22:20,349 um or how it's intended to be. 641 00:22:22,470 --> 00:22:24,739 Um, and now what's so special about it? 642 00:22:24,740 --> 00:22:26,819 Um, just to show you that's the the 643 00:22:26,820 --> 00:22:28,289 current road map. 644 00:22:28,290 --> 00:22:29,969 I'm not sure if this is even still 645 00:22:29,970 --> 00:22:31,979 accurate, but at least it gives you an 646 00:22:31,980 --> 00:22:34,499 overview of how that, uh, it's, uh, 647 00:22:34,500 --> 00:22:35,999 what's been worked out over the last 648 00:22:36,000 --> 00:22:38,369 years. Um, so as a process 649 00:22:38,370 --> 00:22:40,649 of manufacture, you try to to have a real 650 00:22:40,650 --> 00:22:42,749 good, uh, uh, sense of, 651 00:22:42,750 --> 00:22:44,939 uh, regular push of 652 00:22:44,940 --> 00:22:47,099 products out into the markets so 653 00:22:47,100 --> 00:22:48,039 that different needs. 654 00:22:48,040 --> 00:22:50,369 Um, if you look at power, um, 655 00:22:50,370 --> 00:22:51,809 there are different circumstances. 656 00:22:51,810 --> 00:22:53,939 Uh, because you have a different, um 657 00:22:53,940 --> 00:22:55,679 uh, there are markets that you want to 658 00:22:55,680 --> 00:22:57,059 attack if you look at, uh, see, 659 00:22:57,060 --> 00:22:58,709 architecture. This will look much 660 00:22:58,710 --> 00:23:00,749 different. Uh, a C processor will come 661 00:23:00,750 --> 00:23:02,909 out each and every time at 662 00:23:02,910 --> 00:23:04,049 the same point in time so that it's a 663 00:23:04,050 --> 00:23:06,569 very regular pattern by the site 664 00:23:06,570 --> 00:23:09,209 that needs to be much more flexibility 665 00:23:09,210 --> 00:23:11,399 due to the customer changes or market 666 00:23:11,400 --> 00:23:12,779 changes at the end. 667 00:23:12,780 --> 00:23:15,329 But as you can see in the next year, um, 668 00:23:15,330 --> 00:23:17,369 10 seven nanometer quality products that 669 00:23:17,370 --> 00:23:19,979 need to come out and basically you 670 00:23:19,980 --> 00:23:22,139 will be, uh, specified and 671 00:23:22,140 --> 00:23:23,140 pushed 50 partners. 672 00:23:24,360 --> 00:23:26,429 Um, but that's the really interesting 673 00:23:26,430 --> 00:23:28,919 thing about the initiative, um, 674 00:23:28,920 --> 00:23:31,169 that you somehow try 675 00:23:31,170 --> 00:23:32,759 to to connect at the end. 676 00:23:32,760 --> 00:23:34,989 Um uh, as I mentioned before. 677 00:23:34,990 --> 00:23:36,989 Um, external accelerators into the 678 00:23:36,990 --> 00:23:39,659 system. It's not just to plug them into, 679 00:23:39,660 --> 00:23:41,609 but to make them part of the system, to 680 00:23:41,610 --> 00:23:43,499 make them part of the the transmission of 681 00:23:43,500 --> 00:23:45,509 the processor. So it's, uh, not just 682 00:23:45,510 --> 00:23:47,609 enabling, uh, some PCI also, 683 00:23:47,610 --> 00:23:49,439 but it's a protocol layer that makes it 684 00:23:49,440 --> 00:23:51,509 really fast, uh, compared to 685 00:23:51,510 --> 00:23:52,499 other customers. 686 00:23:52,500 --> 00:23:54,899 Um, also, uh, with 687 00:23:54,900 --> 00:23:56,969 this power eight enhanced product, 688 00:23:56,970 --> 00:23:59,369 um, built together with Nvidia 689 00:23:59,370 --> 00:24:01,619 um, the building was introduced as 690 00:24:01,620 --> 00:24:03,929 in, uh, its own link, uh, 691 00:24:03,930 --> 00:24:06,059 built into the CPU, um, to 692 00:24:06,060 --> 00:24:08,159 make that, uh, point with, 693 00:24:08,160 --> 00:24:10,409 uh, much faster into those 694 00:24:10,410 --> 00:24:11,519 chip use. 695 00:24:11,520 --> 00:24:13,379 Um, I'm not sure if you heard about this 696 00:24:13,380 --> 00:24:15,269 announcement quite some weeks ago. 697 00:24:15,270 --> 00:24:17,339 Um, there are those labs 698 00:24:17,340 --> 00:24:19,559 in the US which are basically distributed 699 00:24:19,560 --> 00:24:20,999 with those new systems. 700 00:24:21,000 --> 00:24:22,739 Um, there was some kind of government 701 00:24:22,740 --> 00:24:24,899 contract, um one by 702 00:24:24,900 --> 00:24:26,789 Nvidia together with IBM to distribute 703 00:24:26,790 --> 00:24:28,049 those systems. And basically, they 704 00:24:28,050 --> 00:24:30,269 consist with the latest hardware, um 705 00:24:30,270 --> 00:24:32,429 of, uh, power processors connected 706 00:24:32,430 --> 00:24:34,619 to the latest chip use and basically, 707 00:24:34,620 --> 00:24:36,869 uh, built a room full of those servers. 708 00:24:36,870 --> 00:24:39,089 Uh, and basically make possible 709 00:24:39,090 --> 00:24:41,819 some um computations 710 00:24:41,820 --> 00:24:44,159 and use this year to use generation 711 00:24:44,160 --> 00:24:46,319 of the willing and uh, 712 00:24:46,320 --> 00:24:48,419 that provides, um bentworth that you 713 00:24:48,420 --> 00:24:50,130 are not able to get with other CPUs. 714 00:24:51,420 --> 00:24:53,489 Um, that's the 715 00:24:53,490 --> 00:24:55,739 same thing. Um, but 716 00:24:55,740 --> 00:24:57,959 you see that you have um, the interface 717 00:24:57,960 --> 00:24:59,459 is not just over Ivy League, you still 718 00:24:59,460 --> 00:25:02,069 have PCI generations available 719 00:25:02,070 --> 00:25:04,579 to get that faster, just to decouple 720 00:25:04,580 --> 00:25:06,869 you extensions that you have. 721 00:25:06,870 --> 00:25:09,089 And um, here you can 722 00:25:09,090 --> 00:25:11,279 get, uh, Xilinx products attached. 723 00:25:11,280 --> 00:25:13,679 Uh, also with, uh, Cappy open copy. 724 00:25:13,680 --> 00:25:16,289 Um, which is basically this 725 00:25:16,290 --> 00:25:18,779 new protocol and, uh, hardware 726 00:25:18,780 --> 00:25:19,829 built into the processor. 727 00:25:22,160 --> 00:25:24,319 This is just for numbers, not for 728 00:25:24,320 --> 00:25:25,379 marketing. 729 00:25:25,380 --> 00:25:27,679 Um, basically, they were able to, 730 00:25:27,680 --> 00:25:29,869 um uh, Exxon Mobile was able 731 00:25:29,870 --> 00:25:32,119 to basically reduce, uh, 732 00:25:32,120 --> 00:25:33,979 thousand servers into 30 open power 733 00:25:33,980 --> 00:25:36,049 servers just to put together an idea what 734 00:25:36,050 --> 00:25:39,049 that makes. At the end, um, 735 00:25:39,050 --> 00:25:41,209 just for numbers that's regarding to the 736 00:25:41,210 --> 00:25:42,169 community itself. 737 00:25:42,170 --> 00:25:44,569 Um, the idea is basically to get uh, 738 00:25:44,570 --> 00:25:46,249 or do you have the possibility to get 739 00:25:46,250 --> 00:25:47,559 everybody involved? 740 00:25:47,560 --> 00:25:49,789 That means people that, uh, 741 00:25:49,790 --> 00:25:52,099 write software, people that use 742 00:25:52,100 --> 00:25:54,639 software or people at the end that, uh, 743 00:25:54,640 --> 00:25:56,869 uh, use the whole systems and uh, try to 744 00:25:56,870 --> 00:25:59,449 basically, uh, get those different 745 00:25:59,450 --> 00:26:01,609 interest groups together and have a place 746 00:26:01,610 --> 00:26:04,399 for them in that overall community. 747 00:26:04,400 --> 00:26:06,649 And that's why, um, they created, uh, 748 00:26:06,650 --> 00:26:08,869 different groups, um, to get more 749 00:26:08,870 --> 00:26:10,939 into technical discussions and basically 750 00:26:10,940 --> 00:26:12,709 technical solutions at the end. 751 00:26:12,710 --> 00:26:14,149 And that's basically what we currently 752 00:26:14,150 --> 00:26:15,069 have today. 753 00:26:15,070 --> 00:26:17,509 Um, different different, uh, 15 754 00:26:17,510 --> 00:26:19,219 different work groups available. 755 00:26:19,220 --> 00:26:21,589 Um, there are some that, um, I'm 756 00:26:21,590 --> 00:26:23,059 even involved with some colleagues. 757 00:26:23,060 --> 00:26:24,409 Um? 758 00:26:24,410 --> 00:26:26,599 Um, that's basically we're talking 759 00:26:26,600 --> 00:26:28,729 about real problems and 760 00:26:28,730 --> 00:26:29,809 solutions at the end. 761 00:26:29,810 --> 00:26:31,999 Uh, they re trying to solve them 762 00:26:32,000 --> 00:26:33,319 at different points of the stack. 763 00:26:35,270 --> 00:26:37,609 And that's basically should be the last 764 00:26:37,610 --> 00:26:39,859 light regarding what 765 00:26:39,860 --> 00:26:42,259 is at the end, um, it's still some 766 00:26:42,260 --> 00:26:44,239 something that costs money. 767 00:26:44,240 --> 00:26:46,689 That's why companies need to do to 768 00:26:46,690 --> 00:26:49,429 to spend money to be part of open power. 769 00:26:49,430 --> 00:26:52,219 But if you are coming from academia, 770 00:26:52,220 --> 00:26:54,349 then you are able to spend zero dollars 771 00:26:54,350 --> 00:26:56,119 and still be part of the community. 772 00:26:56,120 --> 00:26:57,949 And there's different levels between 773 00:26:57,950 --> 00:27:00,229 them. Um, so it 774 00:27:00,230 --> 00:27:02,329 needs to be somehow a little 775 00:27:02,330 --> 00:27:04,309 bureaucratic that way, obviously. 776 00:27:04,310 --> 00:27:06,829 Um, but you are able to get involved, 777 00:27:06,830 --> 00:27:08,659 especially if you are coming from 778 00:27:08,660 --> 00:27:10,999 research. Um, I have even 779 00:27:11,000 --> 00:27:13,069 checked today, um uh, 780 00:27:13,070 --> 00:27:14,959 to you mentioned should be involved. 781 00:27:14,960 --> 00:27:17,299 Um, the, uh, 782 00:27:17,300 --> 00:27:18,859 Paderborn University. 783 00:27:18,860 --> 00:27:20,749 I have seen the chart, uh, as well today 784 00:27:20,750 --> 00:27:22,399 as there are already some German 785 00:27:22,400 --> 00:27:24,859 universities joining that, uh uh, 786 00:27:24,860 --> 00:27:26,869 community. Uh, but most of them are still 787 00:27:26,870 --> 00:27:27,870 located in the U.S.. 788 00:27:29,270 --> 00:27:31,999 Um, that was all 789 00:27:32,000 --> 00:27:34,129 from my side. Just a brief overview, so 790 00:27:34,130 --> 00:27:36,559 it was not intended to be in that, 791 00:27:36,560 --> 00:27:38,659 uh, also not complete if 792 00:27:38,660 --> 00:27:40,459 you have more questions, uh, just come to 793 00:27:40,460 --> 00:27:42,709 me if you also contacts to 794 00:27:42,710 --> 00:27:44,149 other people, if you want to talk more 795 00:27:44,150 --> 00:27:46,489 details regarding an interface 796 00:27:46,490 --> 00:27:49,489 or any contacts in those groups, um, 797 00:27:49,490 --> 00:27:51,639 but uh, at first you should check or 798 00:27:51,640 --> 00:27:53,389 dot org. Um, that's the website. 799 00:27:53,390 --> 00:27:55,429 I bet you can get more information about 800 00:27:55,430 --> 00:27:57,979 that. And um, I hope that 801 00:27:57,980 --> 00:28:00,229 gave you some inspiration and some 802 00:28:00,230 --> 00:28:02,419 information about what is going on 803 00:28:02,420 --> 00:28:03,420 in the hardware community. 804 00:28:05,130 --> 00:28:06,130 People. 805 00:28:12,470 --> 00:28:13,819 Thank you very much, Matthew. 806 00:28:13,820 --> 00:28:15,829 We do have time for a few questions and 807 00:28:15,830 --> 00:28:17,959 someone already lined up, might one 808 00:28:17,960 --> 00:28:20,599 please? Hi, thanks for your talk. 809 00:28:20,600 --> 00:28:22,729 As a hacker, how can I get a open 810 00:28:22,730 --> 00:28:25,159 power development board of a not like a 811 00:28:25,160 --> 00:28:26,839 big data center like Google or someone? 812 00:28:27,920 --> 00:28:30,049 Um, I think if you have 813 00:28:30,050 --> 00:28:32,119 contacts to universities, 814 00:28:32,120 --> 00:28:33,319 it's pretty positive. Look at something 815 00:28:33,320 --> 00:28:34,609 there. 816 00:28:34,610 --> 00:28:37,099 Um yeah, 817 00:28:37,100 --> 00:28:38,100 that would be my advice. 818 00:28:39,550 --> 00:28:40,789 I'm not a university, 819 00:28:42,080 --> 00:28:44,599 so but I'm sure if you are 820 00:28:44,600 --> 00:28:47,419 an hacking activist, 821 00:28:47,420 --> 00:28:49,469 you should have contacts in that group. 822 00:28:49,470 --> 00:28:51,979 But I think you can even 823 00:28:51,980 --> 00:28:53,359 maybe get some some of the stuff 824 00:28:54,550 --> 00:28:56,869 in there. Are you all the thoughts 825 00:28:56,870 --> 00:28:58,459 coming up yet? 826 00:28:58,460 --> 00:28:59,460 It is already 827 00:29:00,560 --> 00:29:02,869 the old version. No previous version 2.0 828 00:29:02,870 --> 00:29:03,870 and I just don't work out. 829 00:29:04,850 --> 00:29:06,319 Maybe there are some possibilities to get 830 00:29:06,320 --> 00:29:07,369 some older machines available. 831 00:29:08,720 --> 00:29:11,179 The internet woke up and has a question. 832 00:29:11,180 --> 00:29:11,989 Yes. 833 00:29:11,990 --> 00:29:14,449 So you talked about 834 00:29:14,450 --> 00:29:16,639 companies and academia joining 835 00:29:16,640 --> 00:29:17,629 open power. 836 00:29:17,630 --> 00:29:19,729 And one user wants to know, are there 837 00:29:19,730 --> 00:29:22,129 plans from IBM site to offer affordable 838 00:29:22,130 --> 00:29:24,229 systems to consumers so they 839 00:29:24,230 --> 00:29:25,230 can join? 840 00:29:25,880 --> 00:29:27,379 I think that's the at the end. 841 00:29:27,380 --> 00:29:30,149 It's at least an idea for 842 00:29:30,150 --> 00:29:32,449 for companies to to get cheap systems 843 00:29:32,450 --> 00:29:34,759 available, um, for the 844 00:29:34,760 --> 00:29:36,439 end user. So I think it's not the current 845 00:29:36,440 --> 00:29:38,659 target since it's also not the market 846 00:29:38,660 --> 00:29:40,219 IBM is looking into. 847 00:29:40,220 --> 00:29:42,469 So it's more business related issue 848 00:29:42,470 --> 00:29:43,470 at the moment. 849 00:29:44,210 --> 00:29:45,210 Mike, for 850 00:29:45,890 --> 00:29:46,890 Hello? 851 00:29:47,270 --> 00:29:49,459 One question about the 852 00:29:49,460 --> 00:29:51,619 firmware, you said it's 400000 853 00:29:51,620 --> 00:29:52,849 lines of code. 854 00:29:52,850 --> 00:29:54,999 Why is that big and why couldn't you just 855 00:29:55,000 --> 00:29:57,049 reuse Corbould, which already supports 856 00:29:57,050 --> 00:29:59,200 open power the older generations? 857 00:30:00,770 --> 00:30:03,169 That is a specific question. 858 00:30:03,170 --> 00:30:04,909 I think it's just this tech that is 859 00:30:04,910 --> 00:30:07,309 coming out with the machines already, 860 00:30:07,310 --> 00:30:09,469 at least from an IBM site. 861 00:30:09,470 --> 00:30:11,479 And I think the same is coming from 862 00:30:11,480 --> 00:30:13,459 Google. Did you have additional lines of 863 00:30:13,460 --> 00:30:16,159 code written for their stuff 864 00:30:16,160 --> 00:30:18,319 and just released what they 865 00:30:18,320 --> 00:30:19,320 already have? 866 00:30:20,060 --> 00:30:21,169 So it would be question, would you feel 867 00:30:21,170 --> 00:30:23,329 about that the teams might be going 868 00:30:23,330 --> 00:30:25,520 to reuse, as you suspected 869 00:30:26,990 --> 00:30:29,779 Mike to a question regarding 870 00:30:29,780 --> 00:30:31,369 Moore's law. It's never been about 871 00:30:31,370 --> 00:30:33,139 performance. It's been about number of 872 00:30:33,140 --> 00:30:35,329 transistors. So do we have some numbers 873 00:30:35,330 --> 00:30:36,619 for pin power to give them? 874 00:30:38,210 --> 00:30:39,919 Um, I don't have numbers. 875 00:30:39,920 --> 00:30:42,049 You need to look 876 00:30:42,050 --> 00:30:43,249 that up somewhere, I guess. 877 00:30:43,250 --> 00:30:45,469 Um, but basically at the end performance 878 00:30:45,470 --> 00:30:47,159 is that what counts for the customer? 879 00:30:47,160 --> 00:30:49,399 So they look at performance 880 00:30:49,400 --> 00:30:50,749 and that performance is that what you 881 00:30:50,750 --> 00:30:51,750 can? 882 00:30:52,130 --> 00:30:54,439 And Moore's law was this the thing to 883 00:30:54,440 --> 00:30:55,699 improve performance over time? 884 00:30:55,700 --> 00:30:57,919 And now maybe it's not that easy anymore. 885 00:30:57,920 --> 00:30:59,410 That's why you need to trust other means. 886 00:31:00,500 --> 00:31:02,359 Unfortunately, quite a few questions 887 00:31:02,360 --> 00:31:04,309 lining up. Unfortunately, we are out of 888 00:31:04,310 --> 00:31:05,929 time. I'm very sorry for that, but I 889 00:31:05,930 --> 00:31:07,699 think my tell you is still around, so 890 00:31:07,700 --> 00:31:09,889 grab him and ask him in person. 891 00:31:09,890 --> 00:31:11,509 So please give another round of applause 892 00:31:11,510 --> 00:31:13,069 for Matthew. Thank you very much. 893 00:31:34,830 --> 00:31:35,830 The.