It's so weird to hear these guys talk about the "good old days" of chip design in a 30-year old TV show. It's even weirder to hear them talk about CMOS as an emerging technology.
This is one of my favorites episodes! So much of this still applies today, just on smaller scales. It is sad to know a TV show like this today would never make it. :(
David Maiolo All kinds of shows like this on youtube. Maybe not so informative but definitely similar shows. Linus tech tips being one for example. Not the greatest but its something. Problem of today is technology is getting so complex that normal people can't understand as easily anymore. Your right about a TV show though, TV itself is dying and will be gone (or at least reduced to news or bulletins) in a decade or less.
@@decimat777 "Your right about a TV show though, TV itself is dying and will be gone (or at least reduced to news or bulletins) in a decade or less." Just like newspapers died when the cinema was invented and just like cinema died when TVs became affordable? To this day, no medium completely replaced another. The internet also didn't replace any of those.
Poor Herb. They called him out of retirement to fill in for Kildall and he had to drag all his outdated knowledge out of the closet and blow the dust off of it. (17:46)
Where has this show been all my life? And why on earth was ever canceled? This show should be required watching for anyone with a computer science degree. I don’t know of anything of this high of quality explaining computers today.
As a European, I have discovered these shows a few years ago, and I find myself re-watching some of them. Truly fascinating. I've written it before but some years ago I've been reading books, printed in the early 1980s, about the 4004, 8008 and 8080. In them, the path to the present future has been well clarified. The top notch in the IT industry understood the scalability of the technology, its full potential, hence what was laying ahead. The books already explained VLSI, 16, 32 and 64 bit personal computing, and that the step to 128 bit would be a very hesitate one; costs and gains going towards negligible. The top brass knew the overall path of the innovations, and that to get there successfully would be a matter of cunning capitalisation, hard pushing R&D, clever commercialisation, standardisation, muscling the industry and pushing ahead as fast as they possibly could to get to the next stage first. Of course, details evolve as they go (smart phones, tablets etc), but they knew what they were at and after. Some didn't understand too well, perhaps had other short-term motives, and/or couldn't get the capital and/or talents, made wrong turns leading them to failure. But yeah, these guys like AMD here at the table..., they understood all potential very well profoundly.
I think if you added together every single "micro" computer manufactured in 1982-83, you might come up with something about as powerful as a single mid-range Android or iPhone from 2017-today. It would take many years to open and run a single app though.
Just from watching this series though, I'm quite impressed that in this era they had the concepts of things like SIMD and parallelism which laid the foundations for today's stuff.
What's really fascinating about the discussion with Downing is how you can zoom in on and examine--in slow motion, so to speak--the relationship between the technological developments and the market, the way each drives the other. Stewart asks for Downing's take on the claim that the technology is not only ahead of ability to design for it but (he seems to be implying) of the more general ability to apply the technology (around 11:00), i.e. in the market. But Downing can only answer as an engineer. What's of note is how, as their other episodes from that first season show, the PC industry was on the eve, almost literally, of the GUI-revolution. The market was inventing new needs for the new technology and the new technology would end up creating more and more needs for these more powerful chips and this mutual push-pull would continue throughout the decade. If Kildall had been present, maybe he could have pinpointed this future trend more precisely.
and now nvidia gpu has over 78 Billion. funny in this show he was saying 100,000 transistors coming (in their soon future which was a few years away from air date) was consider SUPER. lmao. if you told him how about 78 billion, theyd all probably say, IMPOSSIBLE, no way, why would you need that.. well ill tell you why, as transistor count increased, programmers got so lazy, no software is run efficiently, they just write crap code and let the powerful chips deal with it. if every os and software was 90-100% fully efficient, we would have computers run 10-100 times faster than they do today.
@@davidt8087It's not always due to laziness. You also have tradeoffs for going to market faster, making the code more maintainable, making it easier for many people to work together, etc
I used to watch this show when it aired. I confess I didn't catch every episode, but it is a delight to watch them again. I get to see the ones I messed (though it's been so long every episode is new to me) and it is much more educational to me now than it was then!
I remember people commenting during the 386 and 486 era how hardware progress was outpacing software as wait times for OSes to properly use the new chips increased. And again with multicore as devs took their time getting comfy with multithreading.
Megabits. So 512kb, which is a crazy amount in 1983, to have all that on one chip. The Commodore 64 came out in 1982 and it's 64kb of memory was spread over 8 8kb chips.
This is like interviewing one of Gutenberg's employees. "Well, we think that page stamping technology is really going to take off and continue to develop over the next decade..."
Thank you for posting these. I recall watching this show back in the day. Fascinating watching it today, after all that has happened. And to see these pioneers working it through in the early days.
Here are some numbers for perspective: 1983 - Intel 8088 had 29,000 total transistors(NMOS or CMOS), 3,000 nanometers. 2017: Intel Xeon Broadwell-E5 22 core has 7,200,000,000 transistors (3D trigate), 14 nanometers. This amounts to a 214X increase in transistor densities.
@@DJKinney Due to the llack of any real competitors for the last decade. Now with AMD back in the game they have no choice but to start innovating again.
and now nvidia gpu has over 78 Billion. funny in this show he was saying 100,000 transistors coming (in their soon future which was a few years away from air date) was consider SUPER. lmao. if you told him how about 78 billion, theyd all probably say, IMPOSSIBLE, no way, why would you need that.. well ill tell you why, as transistor count increased, programmers got so lazy, no software is run efficiently, they just write crap code and let the powerful chips deal with it. if every os and software was 90-100% fully efficient, we would have computers run 10-100 times faster than they do today.
I love this content and have watched quite a few of these episodes. It seems, anecdotally, that executives reluctant to change represent companies that no longer exist. I'm writing this comment on a modern computer using an AMD CPU and GPU, so they clearly did something right.
We are so many iterations ahead of this, literally every new nanometer process that we were into now has a new refined process for how we get there. They couldn't even imagine that at the time
I was 2 at the time this was filmed, seeing how technology has progressed though the 80's, 90's, 00's to now, it's interesting looking back and seeing what people in the tech industry thought about the then current technology and where they thought were things going in the future
Four megabit DRAM memory was achieved by three companies just three years after this program aired. Back in the 1980s, even optimistic projections turned out to be less than what would be achieved.
Would've been great if the question of what would be seen in 5 years could've been added to 10 years, 15 years, 20 years, and so on. I recall one episode a little later predicting when the 100MHz processor would come out and they were dead wrong saying something like 1999.
I wonder if nanotechnology today is at roughly the same level as macrocomputing was in the 80's. Imagine a microscopic nanoprocessor with a scaled down 5.5" 256KB disk lol. You could play a microscopic game of King's Quest!
That AMD guy dodged that question that was really about x86, instruction sets, form factor, and other kinds of 'standards' by instead answering the example about network standards.
I’ve come from the future to say, 40 years in the future, we’ve moved from 4mb memory chips to NAND flash chips at 256gb and beyond. Herb would be amazed, since it’s a whole lot more than 100k switches. Lol
@@tookitogo Not in every section of CPU's however, certain areas, Cache being the main example here, have struggled to shrink at the same rate as other sections of a CPU's makeup.
My first PC was an AMD 486 DX2 , 8 mb ram, trident gpu and a stacked 250mb hd. Upgraded it in 1995 with an AWE 32 and a Toshiba 4x speed cd rom. 3 years later it was out dated. The Pentium came, the rest is history. This pc I have now is number 9 since 94. A PC last on average between 3-5 years. Then an upgrade is needed. Although half of the pcs were upgraded in some form or another. Better GPUs or processors. Its been a ride, and now PC graphics are very real. I think in 10 years we wont see the difference between data and real. We have reached the treshold. Next is VR machines, to be in the game. And A.I
What a sweet memory! In the same year I wrote my thesis using wordstar, chi writer, and lotus 123 in my IBM PC compatible 80286 with 20mb hard drive and my cga screen.
At 13:30 The 2901C. In 1977 AMD used the 2900 bit slice family to build an Intel 8080 emulator. Signetics did the same with it's 3000 series bit slice family. Signetics made it's design available as a kit and based it on the Intel SBC 80 bus. The CPU board had a 6-3/4" x 12" form factor. I built a few of these kits and made an adaptor to interface with the S100 bus. Running on CP/M I ran several performance tests which showed the emulator as equivalent to an 8Mhz 8080. That was great in 1977 as the Intel 8080 was clocked at 2 Mhz (4 times the speed). However my kit build was in 1982, at which time I was already running Z80 systems at 8 Mhz.
With them talking about "sub-micron" chips, it's important to remember that it is just another step on the prefix ladder. So one micron is 1000 nanometers; 40 years on we're approaching 1 nanometer.
And I'm from the future, I live in 2019 lol. The iPhone Xs has a 7nm processor! We're approaching the limit of this technology. Can you imagine living in a world without yearly hardware improvements? (Assuming they don't find a revolutionary way of making computers)
@@svirrsvarr Transistor gates are now a mere 100 atoms wide. You can only get so small before quantum tunneling of electrons becomes significant to where classical electric theory doesn't apply.
1 millicent per bit in 1983 = 8 millicents per byte in 1983 = $2,560,000 per 32GB in 1983 = $7,884,235 per 32GB in 2014 (accounting for inflation) (£6,240,984, €7,295,677) Actual price in 2014: £60 (€70, $75) lol
As an electrical engineering student in the early 90's Gallium-Arsenide was always the semiconductor used in any class where the material mattered. Silicon was never used in any examples. My trademark for Gallium-Arsenide Valley never panned out.
@@ph2869 Zilog Inc., Campbell, Calif., which designs and manufactures integrated circuts, named J. Philip Downing senior vice president for technology 1992
Did some searching and his last activities were in a computer history museum seminar in 2009. That was quite awhile ago in itself by now so not sure what has transpired since. No reports of an obituary so I am guessing he is still alive. It would be great if someone could interview him to capture his historical knowledge of events. Edit: Apparently he earned his BS in Electrical Engineering in 1965. So if he was a typical age of 22 at that time, that would place his birthdate around 1943; about 79/80 years old which really isn't that old.
The various microchip manufacturers (e.g. Advanced Micro Devices, Cyrex, Intel, MOS Technologies & others) each develop newer and newer processor chips every year, however the CPU chips are the central component of any computer system.
Back in the early 80’s they used a 10.000nm process for processors,in the 90’s that was 800nm, in the early 2000’s that was 150nm, anno 2018 we are now at 7nm, in 2023 that will be 3nm and in 2030 it will be just 1nm. Just mind blowing how they could fit sooo much transistors on such small chip.
Zod yes they will get past 5nm GARATEED, that’s because scientists have already making a 1nm transistor possible by using carbon rather then silicon, this solves that tunneling effect problem, i believe that in the late 2020’s they will gonna use 1nm, but 5 nm will be used by 2020 followed by 3nm in 2023.
@@LetoxxIant WRONG. carbon will be more expensive, and too difficult to make chips with. ffs learn wht your talking about. carbon nanotubes are no joke. go look up how expensive and time consuming it is to make a gram of nanotubes. carbon nanotubes have been around a long time and its still just as lengthy and difficult to make it today as it was decades ago
This man J. Philip Downing clearly knew what he was talking about, an educated foreseer and one of a few pioneers whose work shaped human life as we know it today and for many decades to come; yet, you google the man and you cannot find a single Wikipedia entry, no biography, to the internet, he is no one, the same internet which was formed on the work of people like himself. If you google the name of any serial killer, any rapest, drug lord, porn star and many others, criminals or not, people who contributed pure evil or nothing at all, all are known to google, to the internet. We humans are pathetic.
Ah, the 2901. If a 16bit home micro had been designed around this time using four of those chips, it could have owned the world. Not only would it have been blazing fast, it would also have been able to precisely emulate just about any of the then popular architectures, making it compatible to everything.
anonUK they weren’t *that* expensive. You needed at least four of them to get some advantage over an eight bit. You needed some control logic because the AMDs where essentially the ALU. And you needed some fast RAM. A fast small piece of SRAM would have sufficed as a software managed scratchpad. So yes it would have been a bit more expensive but not ridiculously much so. You could leave out a lot of other specialized hardware, because it could be emulated by the fast bit slice processor for the instances where it was needed. Classic examples are disc controllers, advanced sound and decompression. Many arcade boards had a “mathbox”. Essentially a geometry transform unit.
I thought that's exactly what Steve Jobs wanted to do when he looked at the Xerox Star, which used the 2900 series. He balked when he realized there was no chance to make the system remotely affordable and was full of mass production manufacturing issues. He settled for a standardized microprocessor and replicated the same environment instead.
@@oldtwinsna8347 and made the only slightly less insanely expensively priced Lisa. Also the famously underpowered Lisa. Star was ridiculously over-designed in many, many other ways. The bitslice architecture was not the clinching factor. An Alto redesigned and cost reduced to production with 1980s VLSI processes would have been the sane choice. We would have been in a much different and quite possible much better world had that happened. Remember that Alto was actually sold and used outside PARC in limited quantities. CMOS (or something like it) would probably have won out eventually. But in the early 80s when things got locked in, a bit more power to allow for rudimentary OoP and microprogramming would have allowed a far better starting point. A starting point is not something you can catch up to later.
A hundred thousands transistors on one chip. Golly! Hand-drawn chip designs. Wow! Those were the days. Today, you need a computer to design a CPU chip. All the fun has gone out of it.
Imagine someone 30 years into our future dumping a unit no larger than an iPhone on our desk, with a 4096 core 60 TeraHertz quantum CPU, 512 Petabytes of memory and a 3D crystal storage with a 800 Exabytes capacity...
@@quadgon relax. It's not happening. We are approaching the limit of computing. Unless we find a totally new, revolutionary way of making computers, we are staying at 7 or 5 nm
@@quadgon I still doubt quantum computing is going to be mainstream, since they require extremely low temperatures to operate. As for crystal computing, well... I've never heard of it. I should research it more :) I think the 2020s are gonna be the decade of AI development and deep learning, and how we input data to the machines in quicker and intuitive ways, more than raw power or transistor count increases.
I guess they were able to handle the heat generation problem for many years after this, because CPU heatsinks didn't really become a thing until the (mid-)1990s, at least on desktop PCs... A typical 386 or 486 CPU didn't necessarily require any heatsinks! I guess the Pentium generation CPUs were the first ones to really require it.
Because the most important thing from their point of view is how many bits of information a chip operates on at the same time (old CPUs "IPC" can be directly link to it's width, or bits size it can load/store per cycle). Today CPUs use VERY many tricks under the hood to make process more efficient, since making it wider (than 64-bit), is unnecessary (and can make old software run too slow on such CPUs). We also get wider execution of code via newer and newer instruction sets (SSE, AVX). In case of 4Mbit specifically, I think they asked about ammount of memory (cache, not RAM) a CPU has at it's disposal. Sidenote : A pretty well known chip with 4Mbit memory is Pentium Pro (with 512kB od L2), and a 4MB memory chip could be Core 2 Duo from 2006 (two cores with shared memory).
I think one day we will have hard drives with no moving parts that's the size of a dime that can hold hundreds of gigabytes. Let's see where the future takes us
I wonder if chips made on larger feature size (or older chips) is reliable/lasts longer than today's extremely small and dense chips. A machine made in 1994 that I worked on has a computer board (?) that has no heatsink and covered in dust, but it has never been replaced, it still works. Meanwhile the CPU in my PC just died with no apparent problem, the PC is well-maintained. 🤔
I think it depends on the fab. MOS technology (Commodore) posthumously known to be awfully poor production chips that fail a lot. I believe the Japanese made chips were super high quality though.
"We're able to put thousands of transistors on a single chip"....in 2021 we have nearly 12 billion transistors on the CPU in an iPhone 12. Take that, 1983!
@ 4:43 LOL! Have we exhausted the technology? We've been asking this question since forever it seems. Believe it or not even if we still had the same technology as back then the evolution of technology would still occur in fact probably more so since we would actually have to think when designing a product (more of a challege vs today's throw it together). I respect people's work but just as an example (maxwell and kepler) there are so many optimizations that we could do with current tech, instead we hope to solve the problem with "new" tech instead of optimizing hardware and software. SOC's would be alot more prevalent early on for computers in general as well as HBM. Just now getting this technology because the thought is on profit and not innovation. Good job AMD for bringing the tech to the table (bad for making a "decent" gpu).
+decimat777 I'm guessing you're not an engineer. To get to the level of tech we have today has required ungodly innovation, science, and cognition.... Not just "hey, let's just do it a different way. That'll be so much easier." Innovation is a form of optimization - especially after the Nth level; Merely semantics
+Gatsby Light Upcoming engineer not a pro i will give you that, I've designed two ASICs with SMT by myself but that's all I've done so far . I think you missed the point of my comment. I was referring to the optimizations that could be made with a current set of tech before discarding it to move to another gen hoping that speed and the natural efficiency effects of shrinking a chip will solve the problem. There are many things that could be done to optimize hardware before it could be considered outdated. Intel's tick tock strategy is a small example as I think they could do more. Especially so with software. I'm working on my compsci degree atm and seeing other people's software twice the size as mine and buggy as hell is disheartening (yes i know we're learning but I've seen examples of code and heard stories that its similar in the workforce). So all i was saying was that it was funny that they said this back then and also that we could do so much more with a generation of tech than we care to try.
+decimat777 Actually companies and science departments of many institutes are constantly researching fot new tech. New tech is the only way to achieve new techniques and improvements in resourses capabilities and efficiency, also research is the only way to discover new things or more virtudes and better use of existing things. Research is the only way to obtain more knowlege. Also I disagree about Intel, I think they doing a great job creating processor that use a lot less electricity, and more reliable than years before, they get less hot, and honestly are very impressive compared to older pcs I had and used (Since 1989). I think AMD makes excelent products too, but Intel is ahead of them all the time.
+xplus93 a perfect example of what I was referring to this whole time has recently occurred. AMD has been on a roll releasing drivers for their somewhat dated technology and it has proved to be quite successful. Take the r9 290 as more direct example, at first it was sort of the mid-high range competitor and now it can even compete with a gtx980 in some cases, this card was released in 2013 *take a moment to process for the sake of logic. on the flip side another example is nvidias Maxwell architecture vs Kepler same tech just redesigned and reorganized to be more efficient. AMD had software issues but a great architecture, nvidia had the software down but only decent efficiency, whereas now it's stellar after the redesign. Jumping to the next tech before solving the current issues doesn't solve the problem guys, it creates more. it's ok though, technology is reaching (or currently experiencing, which would explain these two instances above) a plateau, which will be a good humbling experience for us in general, before we can exponentially rise again.
Just watching this reminds me just how amazing the fundamentals were for instance that a stock 1982 C64 can do this ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-jGAsiHY1VCs.html That says a lot about computers of today
