WEBVTT - The Humble Origins of ARM Processors 0:00:04.400 --> 0:00:07.800 Welcome to Tech Stuff, a production from I Heart Radio. 0:00:12.160 --> 0:00:14.880 Hey there, and welcome to tech Stuff. I'm your host, 0:00:15.040 --> 0:00:18.200 Jonathan Strickland. I'm an executive producer with I Heart Radio, 0:00:18.239 --> 0:00:21.840 and I love all things tech. And when it comes 0:00:22.000 --> 0:00:27.360 to micro processors, there are a few names that tend 0:00:27.520 --> 0:00:31.639 to pop up. Intel is obviously a big one. A 0:00:31.840 --> 0:00:34.760 M D is another, and those are the two they 0:00:34.800 --> 0:00:38.280 get talked about when you're discussing stuff like dusktop computers, 0:00:38.280 --> 0:00:42.880 you know, PCs. But when it comes to more lightweight devices, 0:00:43.240 --> 0:00:47.640 you know, like mobile devices, there's another name, ARM a 0:00:48.040 --> 0:00:52.519 r M. Recently, news broke that the graphics card company 0:00:52.600 --> 0:00:57.200 in Vidio would be acquiring ARM for a staggering forty 0:00:57.240 --> 0:01:01.320 billion dollars a princely some so I thought it would 0:01:01.360 --> 0:01:04.040 be a good idea to kind of go on a 0:01:04.120 --> 0:01:08.160 full rundown on what ARM is, its history, and what 0:01:08.280 --> 0:01:12.759 this acquisition means for the industry and for people like 0:01:12.840 --> 0:01:15.800 you and me. And this is gonna be a two 0:01:15.880 --> 0:01:18.440 parter because ARM has been around for a while and 0:01:18.480 --> 0:01:21.480 its story is actually really interesting. Plus it gives me 0:01:21.520 --> 0:01:25.080 opportunities to go off on crazy tangents and tell you guys, 0:01:25.080 --> 0:01:29.039 how various stuff works, which you know is kind of 0:01:29.080 --> 0:01:33.720 my jam. As the kids say, you know, fifteen years ago. 0:01:34.560 --> 0:01:38.960 We'll start with some history lessons. Now, typically when I 0:01:39.000 --> 0:01:42.360 cover the history of a company or a technology, I 0:01:42.440 --> 0:01:46.560 run into a few cases where dates maybe a little confusing. 0:01:46.959 --> 0:01:50.120 Sometimes one source will have a specific date for an 0:01:50.120 --> 0:01:54.360 event that conflicts with a date that's found in another source, 0:01:54.400 --> 0:01:57.120 and so at that point I will usually say that 0:01:58.480 --> 0:02:02.320 I'm sorry, I apologize. I can't get too specific. And 0:02:02.360 --> 0:02:06.320 you would think that ARM wouldn't have this issue. The 0:02:06.360 --> 0:02:09.120 company isn't that old. We can measure its age in 0:02:09.160 --> 0:02:13.000 a few decades, but we only go back to the 0:02:13.080 --> 0:02:16.640 nineteen eighties or so to to look at origins really 0:02:16.680 --> 0:02:19.600 the late nineteen seventies, And yet when it comes to 0:02:19.760 --> 0:02:24.200 particulars such as which events really got things started for ARM, 0:02:24.240 --> 0:02:26.920 there's actually a lot of disagreements. So I'm going to 0:02:26.960 --> 0:02:31.200 give you a a version of ARMS history. But you know, 0:02:31.280 --> 0:02:34.960 don't think of this as the definitive version, because some 0:02:35.000 --> 0:02:37.679 people say, no, you shouldn't trace its history to that point. 0:02:37.760 --> 0:02:41.560 That silly go to this other point instead. Here in 0:02:41.600 --> 0:02:45.320 the United States, when we talk about the early days 0:02:45.639 --> 0:02:49.760 of personal computers, the names that typically pop up in 0:02:49.800 --> 0:02:56.480 those discussions are Apple, Texas Instruments, Commodore, maybe Tandy, and 0:02:56.520 --> 0:03:00.560 then IBM would follow not too long behind does. But 0:03:00.960 --> 0:03:04.480 across the pond in the UK there was another computer 0:03:04.600 --> 0:03:08.160 company that was trying to get an early part of 0:03:08.200 --> 0:03:11.799 the personal computer era, and this company was called Acorn 0:03:12.080 --> 0:03:16.200 Computers Limited. The three co founders of the company where 0:03:16.280 --> 0:03:21.000 Chris Curry, Hermann Hauser and Andy Hopper, whom I suppose 0:03:21.120 --> 0:03:23.360 was just not dedicated enough to go all in with 0:03:23.440 --> 0:03:26.920 the illiterate names of the other two founders, way to 0:03:26.960 --> 0:03:31.840 go and y. Chris Curry was born in nineteen forty 0:03:31.919 --> 0:03:35.720 six in Cambridge, England. He studied math and physics in 0:03:35.840 --> 0:03:39.240 school and went on to work for various technology companies, 0:03:39.280 --> 0:03:43.640 including Pie Limited that's a p y e. Not you 0:03:43.680 --> 0:03:46.240 know the kind of pie that I love, The Royal 0:03:46.360 --> 0:03:51.320 Radar Establishment and Sinclair Radionics. While his stints at Pie 0:03:51.360 --> 0:03:54.880 and Royal Radar were fairly short, he stuck around at 0:03:54.880 --> 0:03:59.480 Sinclair for several years. By the late nineteen seventies, Curry 0:03:59.800 --> 0:04:03.840 was interested in developing computers, but he was finding no 0:04:04.080 --> 0:04:08.119 real support at Sinclair. He had been working on some 0:04:08.160 --> 0:04:11.280 stuff he was trying to pitch the idea to Sinclair, 0:04:11.320 --> 0:04:13.920 but he wasn't finding them very receptive. So we're gonna 0:04:14.000 --> 0:04:17.680 leave off for now with with Curry and move on 0:04:17.880 --> 0:04:22.080 and we'll regroup in a second. Swapping over to Hermann Hauser, 0:04:22.200 --> 0:04:26.400 who was born in Austria in nineteen forty eight. He 0:04:26.520 --> 0:04:30.320 came to Cambridge as a teenager to attend school and 0:04:30.440 --> 0:04:34.360 learn English, but he really enjoyed it. He went on 0:04:34.480 --> 0:04:38.479 to pursue advanced studies in places like Vienna University, but 0:04:38.680 --> 0:04:43.040 he came back to England attending King's College in Cambridge 0:04:43.360 --> 0:04:46.320 and getting an advanced degree. They're smart, dude. His work 0:04:46.360 --> 0:04:49.279 in physics led him to become friends with Curry, and 0:04:49.320 --> 0:04:52.760 when Curry was ready to start a company to manufacture 0:04:52.800 --> 0:04:57.560 and market personal computers, Houser was on board. Andy Hopper 0:04:58.080 --> 0:05:01.479 was the youngest of the three co founders, having been 0:05:01.480 --> 0:05:06.080 born in nineteen fifty three in Warsaw, Poland. Hopper studied 0:05:06.200 --> 0:05:10.960 in London and later Swansea University before pursuing postgraduate studies 0:05:11.040 --> 0:05:14.760 at the University of Cambridge. He focused on computer science 0:05:15.160 --> 0:05:19.040 and he was researching networking technologies. He co founded a 0:05:19.080 --> 0:05:23.719 company called Orbis Limited, which focused mostly on Networking Tech, 0:05:23.960 --> 0:05:27.839 and this entity would end up merging with Curry and 0:05:27.920 --> 0:05:32.599 Houser's efforts to bring Acorn Computers Limited to life. Although 0:05:32.640 --> 0:05:36.840 the original name for the actual company was Cambridge Processing 0:05:37.040 --> 0:05:42.320 Unit Limited CPU cute right, but the co founders decided 0:05:42.360 --> 0:05:46.320 to use the name Acorn Computers as the trading name 0:05:46.560 --> 0:05:50.719 for the company, allegedly choosing the name Acorn so that 0:05:50.800 --> 0:05:55.960 their computers would appear ahead of rival Apple Computers whenever 0:05:56.000 --> 0:05:59.960 it was in an alphabetical listing. One of the earliest 0:06:00.160 --> 0:06:03.880 jobs that the group had was to develop micro controllers 0:06:04.160 --> 0:06:09.400 for fruit machines, and folks, I consider myself an Anglo 0:06:09.480 --> 0:06:12.680 file but when I first read that, I have to 0:06:12.760 --> 0:06:16.200 admit I had no idea what the heck it meant. 0:06:16.480 --> 0:06:19.880 In my imagination, it was some sort of harvesting device 0:06:19.960 --> 0:06:24.680 that depended on micro controllers to do something like pick apples. 0:06:24.760 --> 0:06:27.880 But no, that, of course, is not what a fruit 0:06:27.960 --> 0:06:30.640 machine is. And my guess is there's more than a 0:06:30.640 --> 0:06:33.400 few of you out there giggling at my ignorance right now. 0:06:34.120 --> 0:06:37.960 It's well warranted. So a fruit machine in this context 0:06:38.040 --> 0:06:41.680 is what Brits call a slot machine. I guess they 0:06:41.720 --> 0:06:45.600 do it because of the symbols of fruit that appear 0:06:46.200 --> 0:06:50.120 on the various parts of the slot machine. So the 0:06:50.120 --> 0:06:53.680 first big job that this new company had was designing 0:06:53.720 --> 0:06:57.520 micro controllers for these gambling machines the slot machines to 0:06:57.560 --> 0:06:59.960 make the more difficult to tamper with, as there were 0:07:00.040 --> 0:07:02.839 some clever hackers who are finding ways to rig big 0:07:02.880 --> 0:07:06.760 payouts from the machines. And while the machines are designed 0:07:06.760 --> 0:07:09.640 to take money from you, uh, they don't like the 0:07:09.680 --> 0:07:12.840 design to go the opposite direction. The house is not 0:07:12.960 --> 0:07:16.040 a big fan of that. A few years later, the 0:07:16.200 --> 0:07:20.720 UK launched an initiative to put a computer in every classroom. 0:07:21.200 --> 0:07:25.800 Acorn Computers secured the contract to provide these computers to 0:07:25.880 --> 0:07:30.520 produce them, and it was called the BBC Micro. The 0:07:30.640 --> 0:07:34.320 Micro used a processor called the six five O two. 0:07:34.640 --> 0:07:38.240 This was an eight bit processor from Rockwell h Though 0:07:38.280 --> 0:07:42.560 engineers from Most Technology originally developed the six five O 0:07:42.720 --> 0:07:45.600 two processor, the six five O two was a low 0:07:45.680 --> 0:07:49.640 cost processor that totally changed the processor market. It was 0:07:49.800 --> 0:07:53.760 much less expensive than Intel's d D processor at the time, 0:07:54.160 --> 0:07:56.800 and as such, the six five O two had found 0:07:56.840 --> 0:08:02.080 its way into numerous technologies in the including the Atrey 0:08:02.840 --> 0:08:06.360 video game console and the Apple two Computer Systems, among others. 0:08:06.880 --> 0:08:12.880 The micro contract gave Acorn Computers some momentum. In three 0:08:13.200 --> 0:08:17.160 the company wanted to free itself from dependence upon processors 0:08:17.280 --> 0:08:22.360 from other companies. To computer scientists from Cambridge University, Sophie 0:08:22.360 --> 0:08:26.400 Wilson and Steve Ferber became the head designers for the 0:08:26.600 --> 0:08:31.120 new thirty two bit processor. Ferber focused on the actual 0:08:31.200 --> 0:08:35.040 physical design of the chips architecture, while Wilson was focusing 0:08:35.200 --> 0:08:40.000 on the instruction set. The limited resources forced the pair 0:08:40.120 --> 0:08:43.280 to come up with a simplified approach to processors, and 0:08:43.320 --> 0:08:45.920 they chose to go with a specific approach to processor 0:08:46.000 --> 0:08:51.720 design in a category called reduced instruction set computing or 0:08:52.080 --> 0:08:56.400 risk r I s C. This is in contrast with 0:08:56.559 --> 0:09:01.360 complex instruction set computing or see I s C CISC. 0:09:01.960 --> 0:09:06.000 But what does that actually mean? Well, let's take a 0:09:06.080 --> 0:09:10.240 step back to understand this. A processor's job is to 0:09:10.280 --> 0:09:15.280 perform arithmetic and logic operations on data, and this includes 0:09:15.360 --> 0:09:18.560 basic stuff like you know, adding and subtracting, kind of 0:09:18.600 --> 0:09:22.599 like your basic calculator, and it can also involve transferring 0:09:22.679 --> 0:09:26.960 numbers and comparing different numbers to one another. The data 0:09:27.240 --> 0:09:31.200 comes in as binary or let's be zeros and ones. 0:09:32.000 --> 0:09:35.560 The processor follows instructions given to it by a program. 0:09:35.920 --> 0:09:39.760 So the processor gets its instructions like, you know, add 0:09:39.840 --> 0:09:42.839 the next two numbers together and then send it on, 0:09:43.520 --> 0:09:47.679 and then the processor executes that instruction on the supplied 0:09:47.920 --> 0:09:51.000 zeros and ones that come in. And that's basically what's 0:09:51.000 --> 0:09:54.400 going on with a processor at a very high level. 0:09:55.000 --> 0:09:59.119 In addition, we describe the number of operations a processor 0:09:59.200 --> 0:10:02.720 can complete in a second as its clock speed, which 0:10:02.760 --> 0:10:06.040 we talk about in hurts. A single pulse of the 0:10:06.080 --> 0:10:10.760 processor is a cycle. A one Hurts processor would only 0:10:10.800 --> 0:10:14.240 be able to complete a single operation every second. It 0:10:14.240 --> 0:10:19.720 would be unfathomably slow to us. A decent processor speed 0:10:19.720 --> 0:10:23.520 today is somewhere between two point five and three point 0:10:23.559 --> 0:10:26.920 five giga hurts. That's decent. I'm not talking about top 0:10:26.960 --> 0:10:29.319 of the line, but that would mean two point five 0:10:29.520 --> 0:10:33.760 to three point five billion cycles per second. So the 0:10:33.840 --> 0:10:39.560 processors in modern computers are pulsing billions of times every second, 0:10:39.559 --> 0:10:44.120 and each pulse can power and operation. Now, some instructions 0:10:44.160 --> 0:10:47.840 are pretty simple and they might only require one or 0:10:48.000 --> 0:10:52.720 two cycles to complete that instruction. Other instructions are more 0:10:52.760 --> 0:10:56.400 complicated and might have lots more steps involved, and this 0:10:56.480 --> 0:11:00.559 is where we get to the risk versus CISK approach. 0:11:00.640 --> 0:11:05.480 A risk based processor handles very simple instructions, so it 0:11:05.559 --> 0:11:10.239 handles each individual instruction very quickly, like within a cycle. 0:11:10.880 --> 0:11:15.880 CISC systems can handle much more complicated instructions. The flip 0:11:16.040 --> 0:11:19.320 side of that is that while a risk based processor 0:11:19.440 --> 0:11:23.439 can execute individual instructions very very quickly, you might need 0:11:23.480 --> 0:11:27.440 a lot more instructions to complete your overall task. The 0:11:27.559 --> 0:11:31.400 CISK approach might take longer to execute a single instruction, 0:11:31.679 --> 0:11:35.800 but you need fewer instructions overall to complete your task. 0:11:36.120 --> 0:11:39.880 Now that is a little confusing, So I'll use an analogy. 0:11:40.200 --> 0:11:42.160 If I told the typical person, I need you to 0:11:42.200 --> 0:11:46.000 go outside and check the weather, that's a deceptively complicated 0:11:46.040 --> 0:11:49.400 instruction because there's a lot of other stuff that's nested 0:11:49.559 --> 0:11:52.640 in that request. For example, if I were to try 0:11:52.679 --> 0:11:54.840 and tell this to a robot, I might have to 0:11:54.880 --> 0:11:58.079 include what direction the robot needs to go in, how 0:11:58.160 --> 0:12:01.720 fast it should move, where of the door is, whether 0:12:01.800 --> 0:12:05.840 that door opens inward or outward, the actual mechanism the 0:12:05.920 --> 0:12:08.920 robot would have to manipulate to open the door, and 0:12:08.960 --> 0:12:12.000 so on. So what appears to be a simple task 0:12:12.400 --> 0:12:16.760 is actually when you break it down into its individual components, 0:12:17.160 --> 0:12:20.800 much more complicated. So a program running on a risk 0:12:20.880 --> 0:12:24.160 based processor has to break down instructions kind of in 0:12:24.200 --> 0:12:28.200 that way in a longer series of simple tasks that 0:12:28.320 --> 0:12:31.960 add up to whatever you're end goal is. But risk 0:12:32.160 --> 0:12:36.400 chips are highly optimized, so for certain applications a risk 0:12:36.480 --> 0:12:40.160 based chip can be ideal. These days, we have a 0:12:40.200 --> 0:12:43.440 lot of risk chips and stuff like mobile devices, for example, 0:12:43.480 --> 0:12:48.040 because these devices are rarely running super complicated software and 0:12:48.080 --> 0:12:53.199 they need that low power, high efficiency output. One related 0:12:53.280 --> 0:12:56.520 thing I'd like to mention, though it doesn't tie directly 0:12:56.600 --> 0:13:00.080 into arms history or anything, is what is called a 0:13:00.160 --> 0:13:04.640 semantic gap. Now, remember when I said that processors taken 0:13:04.760 --> 0:13:08.320 data in the form of zeros and ones. This binary 0:13:08.400 --> 0:13:11.440 code is a type of machine language, or the kind 0:13:11.440 --> 0:13:15.920 of information a machine can actually process. Machines are not 0:13:16.080 --> 0:13:20.720 able to process information in other forms directly. The information 0:13:20.800 --> 0:13:24.720 must ultimately convert into machine language, in this case zeros 0:13:24.720 --> 0:13:29.800 and ones, and information that we can express pretty succinctly 0:13:29.960 --> 0:13:33.400 with language and numerals. Beyond just the ones and zeros, 0:13:33.960 --> 0:13:35.880 that ends up taking up a lot of space. You 0:13:35.920 --> 0:13:37.760 have to use a lot of ones and zeros to 0:13:37.880 --> 0:13:41.360 represent that kind of information. But computers are really good 0:13:41.360 --> 0:13:46.359 at processing machine code. It happens lightning fast. However, programming 0:13:46.400 --> 0:13:51.400 computers in machine code is really really hard. I mean, 0:13:51.440 --> 0:13:54.640 imagine having to type in a string of tens of 0:13:54.760 --> 0:13:58.520 thousands of zeros and ones while you're trying to program 0:13:58.520 --> 0:14:01.120 a machine, and you know that if you make just 0:14:01.280 --> 0:14:04.600 one mistake, you mess up the whole program because the 0:14:04.600 --> 0:14:07.360 whole chain is screwed up after that. Heck, if you 0:14:07.400 --> 0:14:10.080 did make a mistake, it would be really hard for 0:14:10.120 --> 0:14:13.040 you to track down where you made the mistake in 0:14:13.080 --> 0:14:16.240 the programming. You would have to compare two different, very 0:14:16.320 --> 0:14:19.480 long sheets of zeros and ones, and you'd probably lose 0:14:19.520 --> 0:14:22.960 your mind. That's one of the big reasons computer scientists 0:14:23.000 --> 0:14:26.880 have developed various programming languages. The idea is that the 0:14:26.960 --> 0:14:30.640 programming language is something that's easier for human beings to 0:14:30.720 --> 0:14:35.160 work with, but machines can't understand programming languages without the 0:14:35.280 --> 0:14:39.520 use of something called a compiler. The compiler's job is 0:14:39.640 --> 0:14:42.920 essentially that of a translator. It takes the program that's 0:14:42.920 --> 0:14:46.920 been written in whatever programming language and converts the instructions 0:14:46.920 --> 0:14:50.200 into machine code so that the computer can process it. 0:14:50.960 --> 0:14:55.120 The compiler is essentially a middleman between the program and 0:14:55.160 --> 0:14:59.560 the processor. We describe programming languages by calling them stuff 0:14:59.600 --> 0:15:03.480 like level or high level. This refers to how closely 0:15:03.560 --> 0:15:07.400 the language resembles the machine code. So a low level 0:15:07.480 --> 0:15:11.000 programming language is really only a couple of steps away 0:15:11.040 --> 0:15:14.840 from machine code itself. It's much easier for a compiler 0:15:14.880 --> 0:15:17.760 to handle that kind of language, but it's much harder 0:15:17.760 --> 0:15:20.840 to program in. You have to frame your programming closer 0:15:20.880 --> 0:15:25.000 to machine code, but it's still easier than programming instructions 0:15:25.000 --> 0:15:29.160 than just ones and zeros. A high level programming language 0:15:29.840 --> 0:15:33.320 is modeled closer to how we would think in terms 0:15:33.400 --> 0:15:36.640 of a typical language. There are still rules you have 0:15:36.680 --> 0:15:39.680 to follow, and if you're not familiar with that particular 0:15:39.720 --> 0:15:42.360 language and you're looking at an example of it, it's 0:15:42.360 --> 0:15:44.240 not likely going to make a whole lot of sense 0:15:44.280 --> 0:15:46.960 to you. But it's much easier for humans to work 0:15:47.000 --> 0:15:50.440 with these kind of languages. However, it's less efficient for 0:15:50.560 --> 0:15:54.640 compilers to handle that and compile that into machine language. 0:15:55.280 --> 0:15:59.960 We call this adding layers of abstraction. The programming lay. 0:16:00.120 --> 0:16:04.480 WHIGE provides an abstract platform that represents the various tasks 0:16:04.520 --> 0:16:08.000 that the processor will ultimately carry out, and the gap 0:16:08.240 --> 0:16:11.680 between what the programming language says and what the processor 0:16:11.840 --> 0:16:17.200 does is the semantic gap. CISK and risk designs deal 0:16:17.280 --> 0:16:21.280 with this gap in different ways. A CISK design includes 0:16:21.320 --> 0:16:25.040 a lot of addressing modes and lots of different instructions. 0:16:25.080 --> 0:16:28.800 A RISK design has a much more simplified instruction set 0:16:29.200 --> 0:16:32.400 that can meet the requirements of user programs. It's really 0:16:32.440 --> 0:16:36.640 just two different methods to achieve a similar result. Depending 0:16:36.720 --> 0:16:41.320 upon the history you read, Acorn Computer slash Cambridge Processing 0:16:41.400 --> 0:16:47.640 Unit called their risk based design Acorn Risk Machines, or 0:16:47.680 --> 0:16:51.800 they called it Advanced Risk Machines. Most histories say that 0:16:51.880 --> 0:16:55.120 originally it was Acorn Risk Machines and only later changed 0:16:55.160 --> 0:16:59.240 to Advanced Risk Machines. But either way, the initialism for 0:16:59.320 --> 0:17:04.080 this technolog g became a r M or ARM. When 0:17:04.080 --> 0:17:07.040 we come back, i'll talk about how this technology would 0:17:07.080 --> 0:17:11.600 ultimately transcend the company that spawned it, but first let's 0:17:11.640 --> 0:17:23.240 take a quick break. Steve Ferber, Sophie Wilson, and Robert 0:17:23.240 --> 0:17:27.600 Heaton program the initial instruction set for the ARM processor 0:17:27.720 --> 0:17:31.960 in Basic. That's a programming language that originated back in 0:17:32.119 --> 0:17:37.439 nineteen sixty four. Basic stands for Beginners all Purpose Symbolic 0:17:37.560 --> 0:17:41.680 instruction Code. It's a high level programming language that, as 0:17:41.720 --> 0:17:47.080 the name implies, simplifies things for beginner programmers. The Acorn 0:17:47.119 --> 0:17:50.600 team weren't beginners, but they wanted to keep instructions as 0:17:50.680 --> 0:17:54.760 simple as possible to optimize the processors. Their first effort 0:17:54.840 --> 0:18:00.520 yielded the ARM one processor That endeavor took two years 0:18:00.680 --> 0:18:05.119 of development, with the ARM one debuting in nine only 0:18:05.240 --> 0:18:13.119 debuting internally v l s I Technology, another company fabricating company, 0:18:13.160 --> 0:18:17.840 they actually produced the chip the working chips. The chip 0:18:17.920 --> 0:18:21.480 had fewer than twenty five thousand transistors on it and 0:18:21.560 --> 0:18:26.480 used a process with a resolution of three microns or micrometers. 0:18:26.520 --> 0:18:31.639 That's one millionth of a meter for comparisons sake, Today's 0:18:31.760 --> 0:18:35.959 Intel processors have more than a billion transistors and they 0:18:36.040 --> 0:18:38.520 use a fabrication process with a resolution of just a 0:18:38.560 --> 0:18:42.680 few nanometers, and a nanometer is one billionth of a meter, 0:18:42.840 --> 0:18:46.880 so we've definitely come a long way since the early eighties. 0:18:47.440 --> 0:18:50.080 The team learned a great deal through their experience of 0:18:50.119 --> 0:18:53.920 developing the ARM one, and rather than immediately go into production, 0:18:54.080 --> 0:18:57.120 the design team began to work on refining their product, 0:18:57.440 --> 0:19:01.120 creating the next generation of processors based on the architecture. 0:19:01.400 --> 0:19:05.040 They wanted to improve certain processes, and they added instructions 0:19:05.040 --> 0:19:10.200 for stuff like multiply and multiply and accumulate. They built 0:19:10.200 --> 0:19:13.560 in capabilities that would allow the processor to perform real 0:19:13.720 --> 0:19:18.240 time digital signal processing, a necessity if they wanted the 0:19:18.240 --> 0:19:21.720 processor to be able to handle processes meant to you know, 0:19:21.800 --> 0:19:25.280 generate sounds for example, which the company considered an important 0:19:25.359 --> 0:19:29.439 part of a computer's capabilities. They increased the number of 0:19:29.440 --> 0:19:33.040 transistors on the microprocessor from twenty five thousand from ARM 0:19:33.080 --> 0:19:37.440 one to thirty thousand for ARM two. The team also 0:19:37.520 --> 0:19:42.400 developed a coprocessor, which, as that name implies, is a 0:19:42.440 --> 0:19:46.800 processor that can work in concert with the primary processor. 0:19:47.200 --> 0:19:52.640 Coprocessors typically handle specific tasks. They're meant to kick in 0:19:52.680 --> 0:19:57.720 when something specific happens, and it offloads those tasks from 0:19:57.840 --> 0:20:01.120 the responsibility of the primary process sessor. So this would 0:20:01.119 --> 0:20:04.080 be kind of like having two people dividing up work 0:20:04.119 --> 0:20:08.160 among them, and one person handles a subset of chores 0:20:08.560 --> 0:20:10.359 and the other person has to do all the rest 0:20:10.359 --> 0:20:13.119 of the chores. Now, in this case, the coprocessor was 0:20:13.160 --> 0:20:18.280 powering a floating point accelerator. Ah, but that leads us 0:20:18.280 --> 0:20:23.399 to ask what is a floating point? Well, I'm sad 0:20:23.480 --> 0:20:25.600 to be the one to have to tell you this. 0:20:25.840 --> 0:20:31.280 Please set yourself down and and prepare yourself. Computers have 0:20:31.440 --> 0:20:36.400 a limited capacity. Computing memory is not infinite, and so 0:20:36.520 --> 0:20:39.760 we have to start making some concessions when we're working 0:20:40.040 --> 0:20:43.680 with numbers. Now, as you may be aware, some numbers 0:20:44.040 --> 0:20:49.239 can be really really big or really really small, and 0:20:49.280 --> 0:20:53.120 they might have a super long, perhaps even infinite number 0:20:53.359 --> 0:20:58.160 of numbers behind a decimal point. Computers can't cope with that. 0:20:58.680 --> 0:21:02.000 They have limitation on what they can handle. So we 0:21:02.080 --> 0:21:05.640 have to make some concessions, and floating points are one 0:21:05.640 --> 0:21:08.480 of the ways we make concessions. Now, at some point 0:21:08.800 --> 0:21:12.280 we have to cut off numbers, and when and where 0:21:12.320 --> 0:21:15.119 we cut off numbers depends upon what we're doing. So, 0:21:15.200 --> 0:21:18.760 for example, if we are making a tool like a rake, 0:21:19.760 --> 0:21:22.840 you know, just an old lawn rake, and you want 0:21:22.920 --> 0:21:25.120 the handle for this rake to be five ft long, 0:21:25.640 --> 0:21:28.920 you probably don't actually care if the handle comes out 0:21:28.960 --> 0:21:32.919 to be four ft eleven inches and some change, or 0:21:33.280 --> 0:21:36.200 five foot and a fraction of an inch that level 0:21:36.200 --> 0:21:39.639 of precision isn't really important to you. It needs to 0:21:39.640 --> 0:21:43.280 be five ft ish, but if it's not exactly at 0:21:43.280 --> 0:21:46.400 five ft it's not a deal breaker. But let's say 0:21:46.400 --> 0:21:50.200 you're building a transistor for a processor. Well, in that case, 0:21:50.240 --> 0:21:55.600 you're working in a very very very small frame of reference, 0:21:55.920 --> 0:21:58.120 and so the difference of a fraction of a meter 0:21:58.280 --> 0:22:02.560 represents a gargantu one difference. On the flip side, you're 0:22:02.600 --> 0:22:05.679 not likely to ever have to worry about distances of 0:22:05.880 --> 0:22:08.879 a centimeter. That would be way too big. So you 0:22:09.040 --> 0:22:12.280 just have to have a way to maintain accuracy relative 0:22:12.320 --> 0:22:15.280 to what you're doing. You have a different, you know, 0:22:15.600 --> 0:22:19.800 context for your work. This gets a bit more complicated 0:22:19.800 --> 0:22:22.239 when you need to work with both really big and 0:22:22.440 --> 0:22:25.960 really small numbers at the same time. For example, let's 0:22:25.960 --> 0:22:29.480 say you're a scientist and you're working with Newton's gravitational constant. 0:22:29.920 --> 0:22:32.600 That is a very small number that starts with a decimal. 0:22:33.040 --> 0:22:36.159 Then you have ten zeros before you get to the 0:22:36.160 --> 0:22:39.280 first non zero number, which is a six. By the way, 0:22:39.480 --> 0:22:41.760 you might also be working with the speed of light. 0:22:42.680 --> 0:22:46.720 That's a very big number, but the computer memory can't 0:22:46.760 --> 0:22:50.320 really handle number sizes that include that wide a spectrum 0:22:50.320 --> 0:22:54.000 of numbers, and that's why floating points are used, and 0:22:54.040 --> 0:22:58.080 they're sort of like using numbers and scientific notation. You've 0:22:58.080 --> 0:23:02.679 got a significant with contains the digit of the number 0:23:02.760 --> 0:23:05.479 or the digits of whatever number you're talking about, and 0:23:05.480 --> 0:23:08.560 you've got an exponent which tells you where the decimal 0:23:08.600 --> 0:23:11.760 point needs to be in relation to the first digit 0:23:11.880 --> 0:23:15.280 in the significant. So if I have a significant of 0:23:15.520 --> 0:23:19.320 one point seven and I have an exponent of six, 0:23:19.920 --> 0:23:22.040 it would be the same as if I wrote that 0:23:22.160 --> 0:23:25.280 number in the scientific notation as one point seven times 0:23:25.280 --> 0:23:27.920 ten to the sixth power, which is the same thing 0:23:27.960 --> 0:23:31.720 as one million, seven hundred thousand. These are all just 0:23:31.800 --> 0:23:34.800 different ways to represent the same value. So one point 0:23:34.840 --> 0:23:38.160 seven significant with an exponent of six is one million, 0:23:38.240 --> 0:23:41.960 seven hundred thousand. Likewise, if I had a significant of 0:23:42.000 --> 0:23:46.280 one point seven and an exponent of negative six, this 0:23:46.280 --> 0:23:48.520 would be the same as one point seven times ten 0:23:48.640 --> 0:23:54.160 to the negative sixth power or point zero zero zero 0:23:54.359 --> 0:23:59.440 zero zero one seven. By using floating points, we can 0:23:59.480 --> 0:24:04.159 simplify how we represent numbers without damaging the value of 0:24:04.200 --> 0:24:07.240 those numbers. And let's get around the limitations of computer 0:24:07.320 --> 0:24:10.600 memory and how many bits a processor can handle at 0:24:10.600 --> 0:24:15.120 a single time. We call the operations that processors perform 0:24:15.240 --> 0:24:19.199 on these types of numbers floating point operations, and we 0:24:19.280 --> 0:24:23.040 measure it in flops, which stands for floating point operations 0:24:23.080 --> 0:24:27.280 per second. A giga flop would be a billion floating 0:24:27.359 --> 0:24:33.679 point operations per second. The Japanese super computer Fugaku can 0:24:33.760 --> 0:24:38.200 reach more than four hundred fifteen peda flops. A pedal 0:24:38.240 --> 0:24:44.119 flop is a thousand million million floating point operations per second, 0:24:44.520 --> 0:24:47.440 So a pedal flop would be a one followed by 0:24:47.600 --> 0:24:55.000 fifteen zeros yauza. So the ARM to architecture included a 0:24:55.000 --> 0:24:59.920 coprocessor for floating point acceleration, not a full floating point process, 0:25:00.240 --> 0:25:04.720 but to accelerate floating point operation calculations, as well as 0:25:04.720 --> 0:25:09.480 the possibility of adding other coprocessors with the basic ARM architecture. 0:25:09.520 --> 0:25:12.200 It was kind of a sort of a modular design. 0:25:12.720 --> 0:25:16.040 This generation was called, fittingly enough, ARMED two, and the 0:25:16.119 --> 0:25:19.880 first product to market that featured the ARM two wasn't 0:25:19.960 --> 0:25:23.800 a fully fledged computer, but rather the ARM development system, 0:25:23.840 --> 0:25:27.880 which included the ARM processor, four megabytes of RAM, three 0:25:27.920 --> 0:25:31.240 support chips, and some development tools. So essentially this was 0:25:31.280 --> 0:25:34.320 a product meant for programmers. It wasn't like it was 0:25:34.400 --> 0:25:38.800 meant for your average end consumer. Meanwhile, at the company 0:25:38.840 --> 0:25:43.200 at large, things were not going so super well. Acorn 0:25:43.240 --> 0:25:46.520 Computers was in a bit of a financial crisis and 0:25:46.680 --> 0:25:50.760 an Italian company known for computer systems and office equipment 0:25:50.760 --> 0:25:54.719 in Europe called Olivetti ing s c. And I know 0:25:54.840 --> 0:25:58.280 I've butchered it, but Olivetti is what's best known as 0:25:58.680 --> 0:26:02.480 it swept in and it acquired the English computer company. 0:26:02.840 --> 0:26:07.480 At the time, Olivetti was reportedly unaware that within Acorn 0:26:07.560 --> 0:26:10.840 Computers there were engineers who are working on new processors 0:26:10.880 --> 0:26:14.879 because the original Acorn computers we're using processors made from 0:26:14.920 --> 0:26:18.480 other companies, so the acquisition would slow things down a 0:26:18.480 --> 0:26:20.480 little bit. That's one of the reasons why there was 0:26:20.520 --> 0:26:24.000 a delay between the development of the original ARM one 0:26:24.119 --> 0:26:28.720 processor and an actual Acorn computer system running on an 0:26:28.840 --> 0:26:32.080 ARMED two processor. However, the day did eventually come around, 0:26:32.320 --> 0:26:35.639 and that day arrived in n seven, and that is 0:26:35.720 --> 0:26:40.040 when Acorn Computers launched the Archimedes. It was a home 0:26:40.080 --> 0:26:43.840