29
$\begingroup$

I rewatched terminator 2 and got intrigued by a particular quote:

[microchip]

scary stuff, radically advanced. I mean it was smashed, it didn't work, but it gave us ideas, took us in new directions. I mean things that we would've never thought of

So, suppose in some time travel shenanigans modern processor was brought back into 80s. How much progress would be possible to get from that?

$\endgroup$
  • 6
    $\begingroup$ Why would anybody pursue technology that they know ends in an apocalypse within their lifetime? "Hey, my games will run a bit faster for a few years before the whole family and I get incinerated by kill-bots." $\endgroup$ – user535733 Dec 5 '19 at 22:47
  • 3
    $\begingroup$ @user535733 In the movies the engineers didn't know it would end in an apocalypse, they just knew that a highly advanced robot came from the future. Actually I'm not entire sure they new that much either. $\endgroup$ – TitaniumTurtle Dec 5 '19 at 22:54
  • 12
    $\begingroup$ "Hey look at this pretty paper weight." - dude finding a processor and having no idea what the hell it is. $\endgroup$ – Michael Mano Dec 5 '19 at 23:08
  • 3
    $\begingroup$ Find technology in a silicon micrograph taken by a multi-billion dollar 1 megawatt camera? 1980's technologies would jump ahead by 1-2 years from a 2019 chip. If they had a picture of the EUV laser technology, that's the other 39 years of RnD. $\endgroup$ – com.prehensible Dec 6 '19 at 12:06
  • 5
    $\begingroup$ I'm surprised so few people have pointed out that (most of) the photolithography process happens outside the actual chip, and I'm skeptical that the end product can tell you much about the equipment and methods used to produce it. How much could someone from the bronze age learn from a piece of modern steel? How much can steel tell about the furnace(s) used to make it? These questions are rhetorical, but they are also sincere because I don't know the answers. $\endgroup$ – adfaklsdjf Dec 8 '19 at 11:04

17 Answers 17

46
$\begingroup$

A single, broken IC of any sort is unlikely to create any change in direction of research, or spur any new breakthroughs. Almost all of the advancement since the 1980s has been in miniaturization, which has allowed more power for less.... power (I'm talking computing power per amp, etc). This has enabled new things like WiFi and video streaming and the like.

But they wouldn't know what those were if it's broken. They might be able to identify it as a radio transmitter, but not know what it is or what it's for. It'd be smaller, sure, and work on frequencies they probably aren't using, but the concept of a built-in radio transmitter isn't really that new.

Additionally, with a single device, there's only so much you can get out of things without destroying it.

So, it would reinforce the paths that were being investigated - Perhaps speed them up a little, but it wouldn't be anything revolutionary.

Now, for some speculation.

If you sent back a functioning cell phone and charger (Which would function just fine in the 80s), THAT could change things. Not the technology, but how we use it. There were plenty of projects in the 1980s that were outright FLOPS, because people thought they'd be popular, and they weren't. Similarly, there were things that were neglected that have been dug up and people go "That would have been AWESOME! and would have changed a lot of things!"

I'm specifically excluding any historic information - since that's a whole different ballgame in itself - but knowing that, in the year 2020 we aren't always video calling even though the technology is there and clearly capable of doing it would change some people's outlooks. Early mobile phones were not super popular - for numerous reasons, but knowing that they would be everywhere would likely change a lot of priorities. Similarly, the Lithium Polymer batteries found in every modern phone are a little more advanced than the rechargeable batteries of the 1980s - but knowing that they became the battery of choice could acceleate that research, and possibly edge out NiCad and other rechargeable battery types that have their own issues.

The processes for producing chips is iterative, small advancements that really can't be accelerated with just an example of a chip. How they're used can enlighten people into many, many new ideas.

$\endgroup$
  • 2
    $\begingroup$ While I was thinking about the idea of sending back a fully functioning computer, I got to thinking. If I look at the number of operations that computer can do before end of life, and compare it to the total number of operations that can be done by all computers worldwide combined, how many years worth of global computer operations can be crammed into that one computer? $\endgroup$ – Cort Ammon - Reinstate Monica Dec 6 '19 at 4:07
  • 4
    $\begingroup$ A single broken IC would still show you how to achieve that kind of miniaturisation. Things like CMOS, high-k dielectrics (e.g. hafnium), BGA packages etc. It would show you where to focus your research. $\endgroup$ – Michael Dec 6 '19 at 8:31
  • 21
    $\begingroup$ @Michael The broken chip isn't going to show them how it is done. There's no way they would know the how from looking at the end product even if it is perfectly intact. They would already know smaller is better, but how did they make it so small? I guess they can use this broken chip to get more funding, but that's about it IMO. $\endgroup$ – Nelson Dec 6 '19 at 9:16
  • 7
    $\begingroup$ @Nelson: I partially agree. I’m no expert but some things like the materials used or transistor layout (fin FETs etc.) will show them how they got the IC that small. Things like which element to use for p doping to create a CMOS; which element to use for insulation to make gates smaller (or with less capacitance) etc. Of course you still have to be able to achieve the required photolithography precision in the factory, but until 7nm with EUV that seemed like less of a problem. $\endgroup$ – Michael Dec 6 '19 at 9:51
  • 2
    $\begingroup$ But what kind of IC? The OP just says "modern processor", they don't specify a CPU. What about the massive parallelism and on-die memory of a GPU? Customizability of an FPGA? What about memristors and other data-close-to-processing tech and exotic (non silicon) semi-conductors and deep learning / neural net hardware for tensor processing like TPU? What about hyperthreading and multicore processing? I agree with others most of the challenge is fabrication, however the advances have certainly not all been about miniaturization. $\endgroup$ – Davos Dec 11 '19 at 11:46
19
$\begingroup$

Not as much good as you think, most of the important computer hardware stuff that happened between 1980 and now already happened by ~2000.

In many respects, software and society has spent decades trying to catch up to the potential of the explosion of good hardware as it is. Even if you gave Bill Gates access to i7 processors with solid state drives, Windows 95 would still be a buggy, insecure, platform, that crashes at the drop of a hat because isolating memory properly is a process that is STILL being refined to this day [See Meltdown Vulnerability]. It would likely even be more unstable since the protocols by which software and hardware communicate at the lower levels are more complex now than they were back then.

Modern programming is largely about isolating the programmer from the guts of the computer and building on yesterday's successes to avoid yesterday's failures. When a programmer calls on any native function, that function comes from libraries that are build on top of lower level libraries that have been rewritten, reworked, and refactored a dozen times over the course of many years. For example: this text you are reading right now uses a font that is rendered by an extension that is updated nearly every year to a library that was actively development during the 80s and 90s to solve problems created by older font rendering methods created in the 60s. Modern fonts could have easily been rendered on 90s machines, but innovation and market demands simply did not make them progress to their current levels until now.

There is also the issue of educating programmers about best practices on a large scale to make sure they don't try to work around what is already successful due to their own ignorance. The number 1 problem I see in programs written before 2010 is that only a small number of programmers from that time period actually seemed to know how to program. https://xkcd.com/2030/ The theory of how to do things right all existed somewhere in the world, but most programmers before that were actually plumbers, electricians, graphic artists, engineers, annylits, etc. who just so happened to wondering into the realm of programing and knew nothing about these theories. It's taken many years for programming to really solidify as a profession, for standards to actually start to become universal, and for colleges to actually have large enough of a pool of good programmers to pick from to properly teach the next generation.

There is also the early adoption hurdle. An iPhone released in 1995 would have flopped as bad as Google Glass. Most people struggled with the concept of "Why do I need to spend 100 dollars on a pocket phone when I have one at home", expecting them to spend 700 dollars on a pocket computer that can spy on your every action sounds like stupidity in a time where people barely trusted PCs. When transition happens too fast for people to get used to it, they reject it out right.

All that said, yes, there would still be niche improvements. Certain technologies like learning AI and big-data analytics would have arrived much sooner with better chip technology, but software stability and proliferation of technology as a cultural element would have taken just as long. Without that proliferation, there would not be enough people to develop the technology beyond the realm of government and big business use.

$\endgroup$
  • 2
    $\begingroup$ Some of the reasons why moderns OS crash less is because of features of the modern CPUs $\endgroup$ – Antzi Dec 6 '19 at 7:59
  • 4
    $\begingroup$ Windows 95 was buggy, insecure, and crashy basically for two reasons: it had to retain compatibility with what came before it (DOS software that expected access to the full system), and it had to do this within the limitations (mostly memory rather than CPU) of computers of the time. $\endgroup$ – Muzer Dec 6 '19 at 10:04
  • 2
    $\begingroup$ @Antzi: You are correct only if by "modern CPU" you mean those designed in the 1960s. The misdesign of the 8086 was an aberration, not the rule. $\endgroup$ – AlexP Dec 6 '19 at 10:07
  • 5
    $\begingroup$ The main reason Win95 crashed so much was that its OS had no memory protection. Every program on the system had full access to any memory location (including hardware memory mapped locations used by device drivers to control addon hardware like video, audio, or disks) Any little bug in any program running on the machine was likely to crash the entire machine. $\endgroup$ – T.E.D. Dec 6 '19 at 14:08
  • 3
    $\begingroup$ @Nosajimiki speaking as a programmer, I can confirm you are wrong, we had algoritms even before computers, IA wasn't able to be developed many years ago because a lack of hardware, frameworks are just a very high level of abstraction and we develope it because we have enogh hardware to use them. Less hardward less software layers we are able to use. More hardware means more freedom for software development $\endgroup$ – Erick Silva Dec 6 '19 at 17:58
13
$\begingroup$

No doubt it would be informative and encouraging. They should be able to extract the actual chip and examine it under and electron microscope. It would then indicate that fantastic amounts of miniaturisation were possible and that was the way to go. By 1980 they were already on that track but it would have been a great encouragement.

Unfortunately it would not tell them how to make the chips via advanced photolithographic means.

$\endgroup$
  • 5
    $\begingroup$ I suspect that the geometry of modern FETs (significantly different from early 1980s electronics IIRC) would be pretty illuminating. $\endgroup$ – ikrase Dec 6 '19 at 9:11
  • $\begingroup$ @ikrase I don't think mere geometry of the resulting structure would tell anything about how to achieve it. E.g. how would you guess to use multiple patterning if you saw a modern hi-speed transistor in a microscope in 1980? $\endgroup$ – Ruslan Dec 9 '19 at 5:35
  • $\begingroup$ I was thinking more along the lines of "finFets" and "CMOS is the future, they're still using lithography". $\endgroup$ – ikrase Dec 9 '19 at 11:37
13
$\begingroup$

The only thing that will change would probably be the approach to Moore's law.

Since I have started chewing on computers, I have recurrently heard that "we have reached the limit of what is physically possible with miniaturization".

First it was contact exposure, "we can't go smaller!"

Then it came projection, "we can't go smaller!"

Then it came immersion lithography, "we can't go smaller!"

Then it came double exposure, "we can't go smaller!"

Then it came extreme ultra violet, "we can't go smaller!"

Then it came something which is probably still heavily guarded in some safe, "we can't go smaller!"

In all this chain of "we can't go smaller!" we have gone from 1 micrometer (1000 nanometer) of critical dimension to the current 3 nanometer.

Seeing a current state of the art microchip won't tell you how it has been made, but it will tell you that it can be made. Therefore you will be able to devote more energy into answering the "how can I make it?" instead of answering the "can I make it?".

Then of course looking at the details of the structure itself can give you valuable hints. Modern build up of a transistor is far from the one it is found on most didactic texts.

$\endgroup$
  • 5
    $\begingroup$ Was there really much doubt at every step of the way that we could go smaller? After all, it has always been an engineering problem to make a smaller transistor. We are only now looking at actual physics getting in the way, since transistors are so small they are effectively only a few atoms across. The problem until now has been how to make something so small, but now it's whether something so small can be made to work at all. $\endgroup$ – Turksarama Dec 6 '19 at 12:03
  • 3
    $\begingroup$ @nelson I recall a whole lot of angst at various points trying to figure out the next steps. I have a friend who did chip design at AMD. There was a lot of dead end projects before they found workable solutions. $\endgroup$ – SRM Dec 6 '19 at 18:09
  • 1
    $\begingroup$ @Slarty, what most people miss is that Moore's law is not a law of nature like Newton's law. It is just a convenient empirical extrapolation which has proven useful in dictating the roadmap for electronic, and which has been revised already a number of times $\endgroup$ – L.Dutch - Reinstate Monica Dec 6 '19 at 21:04
  • 1
    $\begingroup$ I don't think I've ever heard someone claim we were at the endpoint of miniaturization (or maybe I did and dismissed it so fast that I forgot it even happened). Moore's law no longer holds when the rate of miniaturization can no longer be sustained--which I think happened at least 5 years ago--but miniaturization will surely continue in the post-Moores-law era, albeit at a slower pace. I think we can expect the rate of change to follow an S-curve, not exponential growth right up to a hard stop.. $\endgroup$ – adfaklsdjf Dec 8 '19 at 10:55
  • 1
    $\begingroup$ I think most of that "we can't go smaller" is marketing. Just imagine how you look when you can announce "They said they can't go smaller, but we can!" $\endgroup$ – Ruslan Dec 9 '19 at 5:39
6
$\begingroup$

I don't think that quote is that meaningful.

A microchip today is an evolution of chips of the 80s. But fundamentally they aren't different; it's based on the same principals, we made semiconductors on silicon wafers in the 80s, we still make them on silicon wafers today. With progress made on a myriad of disciplines coming together we've learned to make them smaller and faster, and that's the difference.

We haven't made a fundamentally different technological advancement in this field. So, seeing a smaller and faster chip can't really give you ideas, so to speak.

I know there's a lot of talk about quantum computing, and chips that use quantum states to store not just two types of bits, but many. However as far as I am aware, they're not yet in any way a reality. But if they (or even us) to actually get our hands on something like that, now that could open up paths to many things.

$\endgroup$
  • 2
    $\begingroup$ there've been a lot of patents about logic, "guts" of CPU, and there's been at least great progress in hardware support for multi-threading. There's got to be more than just "fundamentally not different" $\endgroup$ – Noone AtAll Dec 5 '19 at 22:55
  • 2
    $\begingroup$ The multi-threading thing is really just an obvious solution to running into light speed limits w.r.t. clock speed: if you can't make it twice as fast, make it small enough to fit two. $\endgroup$ – manveti Dec 5 '19 at 22:57
  • 1
    $\begingroup$ @manveti there's a lot more to multithreading than "make communication fast". At the very least it wasn't until ~2014 that everyone agreed how much freedom is acceptable in concurrency and hardware was specialized for that $\endgroup$ – Noone AtAll Dec 5 '19 at 23:07
  • $\begingroup$ Sure, but systems with multiple processors aren't anything new. The only difference is that now we have a trend to put multiple cores on a single die, which is driven largely by the fact that we hit a point where the speed of light said we couldn't make cores faster without making them smaller. It's the same reason I wouldn't think they'd be floored by modern processors' pipelining: our pipelines are much longer now than they were then, but the concept was already well-established. $\endgroup$ – manveti Dec 5 '19 at 23:12
  • 2
    $\begingroup$ @NooneAtAll Not so much. There was a lot of work done in the 1980s on multi-core processors (most notably with transputers) and they were all stomped by single-core clock rate increases. Turns out that it's easier to push clock rates than develop new infrastructure. Until the wheels fell off the clock-rate-increase bus about 10 years back, of course, and then they had to relearn all the 1980s lessons again. $\endgroup$ – Graham Dec 6 '19 at 12:42
5
$\begingroup$

The most immediately actionable information would be the manufacturer's name. In the 1980's, especially the early 1980's, it was not obvious which chip manufacturers would survive. Knowing in the 1980's e.g. that Intel would be a survivor would be worth millions.

$\endgroup$
  • $\begingroup$ And if it were a TSMC 7nm chip and was sent to 1982? $\endgroup$ – CGCampbell Dec 6 '19 at 14:15
4
$\begingroup$

Modern processors probably would not give all that much of an advancement. Much of what we have learned requires high transistor counts to make them worth their while. Take FPGAs as an example. The idea is great, but when the number of transistors is low, brethren like CPLD are more efficient. Or look at FLASH memory, which is currently beating Moore's law. It's beating it because we made it too small, and then we use fuse bits to select flash cells that aren't broken. That only works when the flash memory blocks are big enough that the block selection logic doesn't consume too large of a percentage of the chip real estate.

However, if you sent back one of the state of the art neural network chips, you might find something. There are some chips on the bleeding edge which are running faster and smaller than "possible" because they are willing to entertain some mistakes -- error due to voltage fluctuations dopant diffusion rates, parasitic capacitance, etc. Speed (and low power consumption) is more important for some problems. They can always gang together multiple copies of the operation and use voting if reliability is important. Indeed, even if we look at more mainstream chips, like GPUs, we see FLOPS/Watt is a driving metric for performance.

Which means one of these neural network chips could be interesting in the way the chip was in Terminator 2. This idea of "it's okay to get the wrong answer" is a radical departure from the thinking of the 1980s. It would have a chance at actually changing the way software developer think about programming and what hardware they want to see built to support those programs.

$\endgroup$
3
$\begingroup$

Good sides and the big issue that a chip doesn't tell you how to make it have been pointed out already. Now for some words of caution - it might be even a trap. The big problem is that sending just a chip wouldn't offer any reason WHY changes were made - which is a crucial bit to improve stuff.

For example, back then CPUs were planar and used SiO2 and copper, now chips have fins, high-k dielectrics are used and Co-Cu mix is used for the smallest wires. When your engineers see all these differences, they have no idea when and why were they introduced. Trying to implement cobalt to a 1 um chip is stupid as it would perform worse than just copper - but we know that now due to a lot of material research. They can just copy us. Finfets might be considered mandatory at huge sizes to improve contact area or whatever... but they would work like crap due to higher variability in the process. Etc etc.

From the chip architectural point of view, I can't see any immediately obvious traps that could make troubles if you take a typical desktop chip. Weird ISA that has to be decoded would be puzzling, but likely ignored as some weird compatibility layer they don't need. They would see many cores of different types in the same chip, out of order execution with greatly improved branch predictors and whatever else, various trust levels, ability to power off parts of the chip and so on. Some improvements here would be obvious and help with chip development, but I believe they wouldn't try to implement many of these ideas because they take more space than they have available with their manufacturing process. The biggest potential trap is if these people still try to implement too many of these features, leading to huge chips which would be nearly impossible to manufacture - dealing another blow to already troubled fabs.

However, if these people receive some of our more experimental chips like nervana, this could make all sorts of issues as they could abandon traditional chips and believe stuff that mimics our brains is required to make a good robot. These chips would be a far greater learning experience in various fields but they would also have much higher chance to take wrong ideas from them.

$\endgroup$
3
$\begingroup$

It is very hard to guess, almost impossible.

They could likely figure out it was some sort of computer chip. I assume it would get into the hands of somebody smart & generous enough to take it apart and put it through a sufficiently powerful microscope and distribute their findings.

Assuming they were really devoted to studying the thing, they could probably figure out some of the really high level stuff. They had to go to multiple cores at some time. Cache seems to used extensively. Parallelism generally seems to be a huge deal generally. They could probably identify some data processing elements -- we have better adders and multipliers now, but I bet they could figure them out.

They'd know where things are going generally. But, the tradeoffs we make nowadays are very specific to our current state, where memory is big and slow, transistors are tiny and fast. They'd have lots of work to do to fill in the gap.

The business/social aspect might be most interesting. Computers were kind of a big deal by the 80's, right? Like, they weren't everywhere, but there was a nascent industry. The money shoveling process had begun. Knowing that the end state is really good will result in other people becoming interested. More money might speed up the R&D process. But, in our universe the people who worked in and ran the computer industries were interested, curious, ok with the uncertainty. They were to some extent rebels. Telling the business community where it is going to end might result in an injection of folks who are less passionate, more driven by a guaranteed ROI. It isn't clear to me that this will have a positive effect on the industry, at all.

$\endgroup$
  • 1
    $\begingroup$ In the same note, imagine those assembly programmers put in front of overly potent hardware. There would be no more incentive to optimize code. My guess would be that the software industry would evolve more slowly because sloppy coders would have a chance to make their crappy code into mainstream products, thus generally decreasing the overall software quality. It's what we experience now (resource hogs like electron-based apps), except that we have a solid foundation built by the early hardware restricted programmers who had to get inventive. $\endgroup$ – brett Dec 7 '19 at 12:21
2
$\begingroup$

I disagree with the answers given so far.

In hardware-design (and especially processors) there are some very hard mathematical problems. The problem is that if you change the specs of a single element this will also change the specs of everything that is connected. If you want to calculate the "sweet-spot" for all attributes of all transistors there are innumerable possibilities and you have to try a big chunk of them to find a good one. So to build a better processor, you already need one that is almost as fast.

Getting your hands on a processor that is much faster than everything you have might enable you to make calculations that you would otherwise not be able to and thereby allowing you to jump ahead.

One restricting factor would be that you only have one and you might need a lot of them to enable a lot of people working simultaneously.

But assuming that someone chooses to use the processor for hardware-development it might enable that person to solve some problems that were unsolvable before (or would take so long that you could not realisticly wait for it).

$\endgroup$
  • 1
    $\begingroup$ how can you use a processor if you don't have nor know the related power supply specs? $\endgroup$ – L.Dutch - Reinstate Monica Dec 6 '19 at 9:01
1
$\begingroup$

Quite a lot actually.

There is a lot of information to be gained, for which it is not necessary to have a complete, working understanding of the circuit. These include the IC packaging, it's chemical composition, the size of it's process technology. How the die or dies connect to each other, how the component is clocked, and how the clock in propagated around the chip. How heat and power are managed. How the product is grounded.

Even high level visual observations of the die would give good ideas of how the chip is organized. For example in multi core chips you can visually see the repeating patterns of the cores. And the cache memory banks are regular, repeating groups

They were capable of this type of analysis in that time period. Here is a report about the "product evaluation" of a Zilog Z80 in 1979

IC Reverse Engineering, 1970's style, Hackaday

And the actual report from the agency that performed the evaluation.

$\endgroup$
1
$\begingroup$

The first divergence wouldn't be on the hardware level, but in the way software is engineered.

The hardware they get shows a strong focus on parallelism and a heterogenous architecture. CPU, GPGPU, the IO/memory controllers are seperate, smart components. A lot of current days struggles goes back to the then popular idea of having one powerful CPU, offload only stuff you'll never need again.

They would adapt a development approach that is more geared towards parallelism. Think functional programming. The object orientation hype would be significantly reduced. Instead of the GHz race and the efforts to deal with its problems they would gravitate towards growing in breadth. Also we would likely see a wider range of more specialized yet programmable computation units early in the game. Perhaps skip the whole fixed pipeline 3D accelerator phase?

At the same time the programming languages and predominant algorithms would coevolve to make developing on such a platform easier. We spend a lot of effort to keep parallel execution pipelines filled by reordering serial code. The same amount of work put in keeping memories only in sync when needed would have a better scalable benefit.

On a side note: Those neural networks are an excellent candidate for parallel processing.

$\endgroup$
  • 1
    $\begingroup$ A broad fabric of parallel computational units, you say? Like the Transputer, or perhaps even a Connection Machine? Parallelism isn't a new concept, and wasn't a new concept then. It just wasn't necessary until we started to run out of clock-speed headroom -- until then, it was always more productive (of market success) to speed things up instead of broadening them out. $\endgroup$ – jeffB Dec 6 '19 at 14:31
  • $\begingroup$ How is OOP different from functional programming CPU-wise? After all under the hood it is generally just one extra argument to otherwise identical function. $\endgroup$ – Oleg V. Volkov Dec 6 '19 at 23:24
  • $\begingroup$ It is not about the machine code per se. OOP allows for side effects. A pure function doesn't. That has far reaching consequences for anything cache or shared memory related. It makes it easy for a compiler to use parallel execution transparently. Our timeline pushed the concept into the "there be dragons"-territory, We didn't know which fork would be the right one in the long run. They would. $\endgroup$ – Robert Martinu Dec 7 '19 at 2:10
  • $\begingroup$ @OlegV.Volkov Despite the name, "functional programming" is not simply programming with functions. It's programming without mutable variables (as in, all variables are either function arguments, or constants). Removing mutability is useful for parallel programming, because you no longer need to coordinate usage. There's no risk of two threads overwriting the same variable, if overwriting variables is impossible. $\endgroup$ – SlugFiller Dec 15 '19 at 19:10
  • $\begingroup$ @SlugFiller This doesn't change anything at assembler and lower levels. $\endgroup$ – Oleg V. Volkov Jan 6 at 21:48
1
$\begingroup$

Most of the answers seem to incorrectly focus on the idea of sending back a CPU. I can agree that the advances from that would be limited.

However, the Terminator's processor is not simply a CPU. It's a custom-built chip, designed to run the Terminator's neural network. A current-day equivalent, would either be a GPU, or perhaps one of those ASIC chips used in crypto mining machines.

Seeing a GPU would certainly give scientists in the 80s something to think about. While CPUs aim to maximize clock speed per chip, GPUs actually use lower speed chips, but at a much greater number. Also, the chip focuses more on optimized floating point operations, and less on memory access, branching or integer operations.

Seeing this would certainly give scientists of the past new ideas and directions. Like "What if instead of trying to make our CPU faster, what if we cram a whole lot more CPUs into one computer?" This could also lead to quicker advancement in software designed for parallel processing, since our tooling for that is STILL lacking today.

As an added bonus, even if the die is smashed, some of the cores may still be intact.

ASIC would actually be an even better example, however. These are designed to only run one specific algorithm. Thing is, unlike a general purpose CPU or GPU, they would actually contain the algorithm. A bitcoin miner from today ending up in the past, would give the people of the past the SHA256 algorithm, which would not be developed until 2001. At least for cryptography, this would be a huge leap, as the likes of MD2 and its successors (and all of the issues inherent in them) could be completely skipped.

Now, admittedly, an ASIC for SHA256 isn't quite as exciting and revolutionary as an ASIC for a deep learning algorithm. Except for those specifically working in the field, that is.

$\endgroup$
  • $\begingroup$ Back in the 80's they had ROM chips bolted next to poor CPUs. When you see ASIC with that mindset, you think that next you need to go manufacturing custom CPUs instead of using ROM chips with general purpose CPU. In reality, modern computing mostly uses generic processors with fully programmable flash memory (that is, the firmware is on the flash, too). $\endgroup$ – Mikko Rantalainen Dec 16 '19 at 9:29
0
$\begingroup$

I think it's important to realize that they didn't recover a chip from the technology era used to create SkyNet. They recovered a chip developed by SkyNet to run Terminators.

There's no way of knowing how advanced the chip was, relative to technology from our own time. A machine intelligence with the ability to enhance itself can geometrically advance in a very small number of improvement iterations. For all we know, it had achieved the equivalent of 1000 years of human scale advancement by the time it manufactured that Terminator model.

$\endgroup$
0
$\begingroup$

It might actually slow technological progress

I recently started working in this area (ASIC research, not time-travelling death robots*), so I'm by no means an expert. But many ideas get patented and go nowhere, for many reasons. If you showed someone from 20 years ago a finfet or gate-all-around transistor, or a phase-change memory cell, they'd be impressed, but have no idea how to build it.

Technology doesn't exist in a vacuum, and a radically different type of chip wouldn't run any existing programs. Operating systems, compilers, interpreters, etc. have to be written to fit the new chip. At best, someone will see the value and start building the infrastructure which will one day hopefully work with the future chip.

But there is a danger to seeing how one particular chip is implemented, because it gives you tunnel-vision and you can't see the other ways that might be better. Future chip probably came into existence through an iterative, competitive evolutionary process, and probably has quirks that only exist because of the path it took to get there. Some of those quirks are mistakes to be fixed in the next generation of chips.

Finally, if you're sending something back in time to do battle, why send all your good secrets for the enemy to find? Instead send back a decoy chip that will keep the humans busy trying to replicate a failed line of research.

Imagine if someone from 30 years ago found the Intel chip with the floating-point bug!

*to the best of my knowledge

$\endgroup$
  • $\begingroup$ Imagine getting a working Itanium processor without any software. They might figure out that VLIW is the correct way to proceed. $\endgroup$ – Mikko Rantalainen Dec 16 '19 at 9:30
0
$\begingroup$

It might speed things up a great deal - not due to being able to find out the hows and whats of the technology, but because they'd be able to get a BUNCH of investment dollars.

One of the hardest things to do when developing new tech is getting investors to put money in when theres no guarantee that it's even possible. And when they do put in, they want a high level of return/ownership depending on the risk level.

If you can show that 'This is DEFINITELY possible, because as you can see, it's already been done - we just need to work out how.', then the risk to the investor is much lower.

$\endgroup$
-1
$\begingroup$

They wouldn't have the software to make good use of it.

$\endgroup$
  • $\begingroup$ On the contrary, many of the modern advances in sortware are not new, but were invented some time ago (many back in the 80s) and simply remained impractical until hardware performance caught up with the software's needs. $\endgroup$ – Starfish Prime Dec 12 '19 at 10:38
  • $\begingroup$ In theory many, but not all. And there is a difference between theory and practice. $\endgroup$ – Jonathan Rosenne Dec 12 '19 at 10:44

Not the answer you're looking for? Browse other questions tagged or ask your own question.