4 ms·

Ask HN: Why a Computer Need a “Memory”?

4 points by terrycody 8y ago

sry guys I am a totally newbie on computer science, lately I am learning the basic knowledge of computer science from a Youtube channel Crash Course, in their #6 episode, they talked about the hardware level of how a RAM built, I have a very hard time to get the idea why a computer needs a memory/RAM For example, due to billions of transistors inside a CPU, say if we get a number 1110001010101 in somewhere of those transistors connected wires, why not just leave those wires as is (don't touch these wires), is this the same way as "store a value" as RAM does? I am trying hard to visualize this hardware in my brain and why it is necessary to build so called "latch" or whatever, compare with just leave those wires (because we got enough wires inside) Could anyone help on this topic? Any life example like play games, type words, watch videos or whatever, to prove a computer has to have a RAM to work?!

PaulHoule 8y ago
If the memory was coded in the connections between the transistors, it would be a read-only memory, or a ROM.
- terrycody 8y ago
  thanks for your reply, but this is not the answer I am looking, I want to know the reason why RAM is necessary for a computer, why not just leave those circuits untouched, because if a number is 10001011, those wires already got current, I don't need to store this number elsewhere, right?
  - pwg 8y ago
    You can't leave wires 'untouched' in an electrical system. Unless something continuously supplies energy to maintain the charge, the charges on those 'untouched' wires will simply fade away, and with it, your data. [1] The data is stored in the electrical charges on the wires, not in the actual wires, which appears to be one source of your confusion. Further, those wires will get reused on the next cycle for the next piece of data moving through, which means if you don't save the values represented by the current charge state of the wires, the data will be gone when the next cycle happens and new charges are forced onto the wires for the next piece of data. RAM (memory) is the place for holding onto those particular charges (which is what is storing the data) that were on the wires, until a later time when you want to go pick those pieces of data up again. [1] ignoring flash memory, which is a very specially designed "charge holder" that does not need the continuous energy supply to be able to maintain the charge states.
    - AnimalMuppet 8y ago
      Amusing and somewhat related story: I was trying to (programmatically) tell the difference between two versions of a circuit board. At a certain address, version 1 had an 8-bit read-write register, and version 2 had a 16-bit read-write register. So I wrote a 16 bit value to that address, read it back... and read the correct value, even from the board that only had an 8-bit register there! What happened was, when I wrote the value, the bus driver chips drove all 16 bits of the bus to the levels specified by the data. Even if the data wasn't latched by the register, that bit of the data bus still had the voltage level corresponding to the data bit. When I read the value, the bus driver chips reversed direction, and wrote out what the voltage level on the data lines said. For the 8 bits that were not driven by the register, the voltage level they had was the voltage level that I just wrote to them (because every chip on those bits of the data bus was in a high impedance state, both for the read and for the write). So I was literally using the wires of the data bus as memory! I found that it could remember data for roughly two seconds this way. So, why do we need memory? Because we want the computer to remember for more than two seconds. And because we want it to remember more bits of data than we have transistors in the CPU chip. And because we want to use the transistors in the CPU chip for something else - for computing, not for memory.
      - terrycody 8y ago
        thank you for your story, I am too early to understand such a case lol, but I get the idea, which means, a wire can hold current a while, but not forever, yes this is clear, but why not just keep give the wires current flow? Then a "1" state stored! If we want to store another number, just use other wires! We have billions of transistors and bus wires right? Why a RAM is a must when we have near unlimited wires already ?
        AnimalMuppet 8y ago
        The wires are a "data bus", which connects multiple things. You could think of it as a street, which has multiple houses on it. Sure, I could use the street to store my stuff in, but then other people can't use the street to get to their houses. It's better that I store my stuff in my house. This leaves the street free so that multiple people can use it to get to and from multiple houses.
salawat 8y ago
So, you seem to be laboring under some drastic misunderstandings. Let me try to lay out an example that may make things clearer. I have a CPU. A CPU consists of a collection of general registers (data is generally read from and into here), a special set of registers containing metainformation about the state of the CPU, an arithmetic processing unit, and a clock circuit. This is a dirt simple CPU. Now. If you're using a Von Neumann based architecture, your CPU doesn't distinguish data from instructions. (Modern ones do, but ignore that for now.) An "instruction" consists of an op-code (basically a number), plus addresses indicating where input data is being read from, and where it's being written to. So a set of commands for this CPU might look like load $1 1 load $2 1 Add $1 $2 $3 Assuming we're "wired" into some input pin the CPU knows to listen to, we're running a rudimentary set of computations. We're sending in some number that the CPU recognizes as the beginning of a "load" command, it looks at the 8 bits or so following, and loads 00000001 into register $1. Done for that cycle. Next cycle, same deal, but in register $2. Next cycle, we send the contents of $1 and $2 to the adder, which outputs the result into $3. Easy right? Now. Let's do some REAL computation. Let's say I have a ridiculous amount of data to crunch. I have 30 8 bit registers for general use, and my 5 or 6 special registers, and my input pin. But I have 300000000000000 8 bit numbers that I want to add! I can make this work with just the CPU logic. I can do a bunch of serial computations, then record the results by hand, and re-input that result as an input to the next computation over and over again, But dear God, that's boring! Wouldn't it be great if somehow, I could stage all of my information ahead of time, and somehow have the processor look up that information to do it's calculating, with out me having to remember everything for it? The answer to that conundrum is Random Access Memory. It's a separation of concerns if you will. CPU's do computation. That's what they are for. Doing the stuff that might cause a state change. We don't want values sticking around in the processor that aren't changing if we can help it. Once something is done we want it out of the processor so we can accommodate the next state changes. In CS, we tend to have two main axes on which we measure the footprint of computation: CPU time, and space (memory footprint). If I have high space constraints, I can set up the CPU to handle all the state, but for every new computation, I have to reset everything back to basics, and still keep track of all my information. This quickly turns a simple one cycle operation into many cycles worth of operation and bookkeeping about where everything is. Or, If I have memory to write to, I can do my computation, write it out to somewhere that's memory managed, and let the magic Memory Management Unit feed me new data and instructions while also cleaning up results, and holding onto them elsewhere on the motherboard so I can use my full CPU die to do computations more quickly. Look at it like having a helper to get books for you at a library while you're sitting at home reading books and doing math problems. The library is RAM, you are the CPU, and your home is the CPU cache. Without your helper fetching books for you, you have to stop all the time, and take the books you're done with back, check out new ones, go home, and only then start doing math again. (You're not independently wealthy, or psychic, so absurdly big house, or magical access to all of the books by telekinesis is not allowed, sorry!) With your helper, you're doing math (yay!), and he/she's swapping out books to/from the library to keep you running smoothly. Try not to get caught up thinking about wires. Circuits have about as much to do with Computer Science as a scalpel does to Surgery. They're a tool. Nothing more. The important thing is to understand how all the pieces work together to create a functional whole. If you're still confused about why RAM is a thing, let me introduce you to https://en.m.wikipedia.org/wiki/ENIAC https://en.m.wikipedia.org/wiki/ENIAC A computer that, quite literally, was absolutely dependent on the people around it to play the role of "memory" since each computation required rewiring the damn thing. Ain't nobody got the patience for that anymore!

terrycody 8y ago

thanks so much sir! Really appreciated your help! You must be at least a very high-level expert in big company like Google or whatever, or you won't know these things so clearly. To be honest, I have a feeling your answer is 100% what I need to find, but my head is still spinning like hell, I am very old guy now and I have no background of STEM, want to be a self-taught CS person, I just started recently :( I already watched some basic videos like transistors, logic gates, and can fully understand them (what a luck!), and yesterday when I watch that video series:https://www.youtube.com/watch?v=fpnE6UAfbtU&list=PL8dPuuaLjXtNlUrzyH5r6jN9ulIgZBpdo&index=8&t=0s&ab_channel=CrashCourse https://www.youtube.com/watch?v=fpnE6UAfbtU&list=PL8dPuuaLjX... I really have a hard time to understand it clearly in my head, the video is going to build a RAM from scratch, but didn't explain why computer MUST need RAM from a hardwire level, thats why I ask the question. ------------------------------------------------- I will listed the parts I still don't understand in your answer, no idea if u will back at this or not :( -------------------------------------------------- "The data is stored in the electrical charges on the wires, not in the actual wires, which appears to be one source of your confusion." From what I learned the data indicates 1 when that wire got current flow, but since the voltage is still there I think this wire will always has current flow to keep show "1" to our human, until the next circle e.g. a 0 state input come, then no flow anymore because no voltage to send current, am I right? You said its the charges which I try to accept but can't. Because later we will talk about RAM, when data is stored in RAM, a 1 bit number like "1" means that wire will keep getting current in order to keep the state. Sry my English is not good enough hope u can understand, what I mean is, I can see you describe 2 different logics in your answer, input wires only flow a sudden flow of current then fade away, whereas RAM wires will keep let charges flow, but in my eye, all these things build with transistors, so why design like this? (lets go to 2nd question) "those wires will get reused on the next cycle" If the final purpose of design like above is re-use the "input wires" which I can understand! but is this necessary? I mean, in today's CPU, we has billions of transistors inside, for example, if we want to store the number, 1101, just keep current flow in 1, 2, 4 wires, and done! If we need store another number, 11000101, just choose another 8 wires and keep the states, why not just this? We have billions of such wires! ---------------------- Sry for these question, I know myself is a stubborn guy and love to dig things from the very ground up, I think my problem is how to visualize these hardware levels, why they design like this or that, why all these things necessary, I know today's programmer most of them don't even know these things and won't harm to be a good developer, but I don't know why, I want to learn them! Thank you very much sir, at least I will read your answer again and again and again. PS: I already opened a ton of taps to read similar answer online hope I find sth :)

pwg 8y ago
> You must be at least a very high-level expert in big company like Google or whatever, or you won't know these things so clearly. No, not at all. This is all basic electronic circuits and basic electrical engineering material. But there are a LOT of basic foundational items required to understand it, and without those foundational items, it is very confusing. > but didn't explain why computer MUST need RAM from a hardwire level, It probably did not explain 'why' because this is one of those building block fundamentals the video assumed a watcher possessed. A computer needs /somewhere/ to store data for later access and use. The conventional name for that is RAM. > From what I learned the data indicates 1 when that wire got current flow, but since the voltage is still there I think this wire will always has current flow to keep show "1" to our human, until the next circle e.g. a 0 state input come, then no flow anymore because no voltage to send current, am I right? Your sentence above shows one item of confusion. Voltage and current are not the same (your statement above almost uses them interchangeably). Voltage and current are related (by Ohm's law and resistance for current flows) but they are not the same. In an ideal system, a voltage can exist without a current flowing (this is one of the reasons that the word "potential" is often used to refer to voltage, a voltage provides the 'potential' for a current to flow, but no current need to flow to have a voltage. (note, I said "Ideal" system at the start, an "ideal system" is one that is perfect, but does not exist in the real world). So keep in mind that a voltage can exist, with a minuscule current flow (small enough to be considered zero, but not really actually being zero current). In old 5v TTL systems, a "one" is represented by a 5 volt signal on a wire (with very minimal current flow) and a "zero" is represented by a ground level (0 volt) signal on the wire. So you don't, per. se., have ones or zeros because of current flow, but because of voltage levels (with, in an ideal system, zero current flow). > You said its the charges which I try to accept but can't. Because later we will talk about RAM, when data is stored in RAM, a 1 bit number like "1" means that wire will keep getting current in order to keep the state. The fact that you are having a hard time "accepting" that it is charges that are responsible implies that you are missing another fundamental bit of electrical/electronics knowledge. A voltage level is created by having some conductive element contain an excess of electrons (and electrons supply "charge"). An excess number of electrons (relative to some other point in the circuit) creates a "potential" between those two points, and if there is a conductor between the two points, then a current will flow (and the current magnitude is determined by Ohm's law). > a 1 bit number like "1" means that wire will keep getting current in order to keep the state. Well, a transistor network will be supplied with a voltage, and the transistor network will be setup such that a "1" being stored holds the transistors in the correct switching arrangement to continue to hold the voltage at the "1" level. But in an idea RAM, there would be zero current in this state, and in a real world RAM, there will be a minuscule current (the leakage current of the transistors). > input wires only flow a sudden flow of current then fade away, whereas RAM wires will keep let charges flow, but in my eye, all these things build with transistors, Ah, here might be another source of your confusion. Wires are not transistors, transistors are not wires. Transistors exist, wires exist, and wires interconnect transistors. But they are not the same thing, and both are needed for a RAM. The transistors are connected together using the separate wires to create a particular circuit element (a latch) that can hold itself in two different states. > I mean, in today's CPU, we has billions of transistors inside, for example, if we want to store the number, 1101, just keep current flow in 1, 2, 4 wires, and done! If we need store another number, 11000101, just choose another 8 wires and keep the states, why not just this? Because those wires you want to 'use' there are already serving a different purpose. They are the "transportation" network for the data, not the storage. A useful (albeit stretched) analogy might be to think of a road network. Cars are data bits (and to make it easy, there are only white cars (one bits) and black cars (zero bits). The roads are the wires (transportation to move the cars (bits) from place A to place B), and the parking lots connected to the road network are the RAM (places to store the cars (data bits) until they are needed later). And assume no designated parking on the roads (wires). So if you attempted to "store" data on the wires, you'd park the cars on the roads. But if you park the cars on the roads, then they block other cars from using those same roads to go from point C to point D. > why they design like this or that, In the case of your starting question of "why not just store on the wires" the very simple reason is that the wires are for transportation, not for storage, and using them for storage would create a transportation log-jam (nothing else could use them for transport until their storage function was over).

terrycody 8y ago
hey sir, first thank you so much for your typing, this must cost you some times, really appreciated actually trying to help a stupid guy! I googled another several hours till now, yes still confusing as hell, but I indeed learned a lot new things regard "Memory", I think my problem lies in PC architecture, why engineers design a PC like this bla bla, and I indeed know its because practical usage to do this lol, just can't get the real CORE of WHY this. I understand most part of what u said, yes its all about efficiency and best practices. What I still confused about is am I right about the wires things? For example, I don't care about the efficiency or engineering reasons to have "individual Memory" for computers, now I want to build my own PC, I just want to use the wires inside a CPU to do everything! Say, we have a ton of wires now in my CPU, fist input 2 8 bits number, 00000011 and 00100101, because the wires got voltage when it got "1", this wire will always keep at "1" if I don't rewrite any of these wires to another value. So now let's send this 2 number to an adder, so maybe adder need locate the wire numbers (in this case, #1-16), or adder need physically wired onto those 16 lines or whatever,(I think this way is impossible, because if connected, the charges will flow through already, make it harder to add them together?! ) Anyway, if first method is doable, then in my CPU, I done the calculation without any Memory system lol. If I need more numbers, just use more wires! (sry I don't know how many such circuits inside a CPU, I think its limited though) ---------------------------------------------------------------- I think there must be errors of this such logic, just I don't where :( I can't find the exact thing online, I think I am just a fool, because maybe nobody think like this, I fall into some wrong circle. Maybe at the very end of day, just try to convince myself like this: why PC need memory system? because we need store data. Why not just use wires inside a CPU? Because its not efficient lol. Thank you for your help again, I think I will face more problems along the way, what a though journey to self taught CS :(
- pwg 8y ago
  > What I still confused about is am I right about the wires things? No, you are not right about the wires thing. In my other answer, one of your statements implied you seem to think of wires and transistors as the same, but they are not. Wires transport voltages around between the transistors, the transistors switch on/off depending upon what voltages the wires transport to them. The combination of the wires transporting different voltages at different points in time, and the transistors switching on/off dependent upon the transported voltages, is what creates the basic building block of a CPU. > Anyway, if first method is doable, then in my CPU, I done the calculation without any Memory system lol. Is your real question here this one: "Why is the memory located in a separate box apart from the CPU"? Because if that is your real question, then there is a good explanation of why this is so. It is so because the best design of the circuits for the CPU portion is different from an engineering perspective than the best design for the circuits for the RAM portion. And due to restrictions on integrated circuit manufacturing, it is easier and cheaper to build the two components separate, using different integrated circuit techniques, and then connect them together via bus lines, than to build them all together in one single component. The circuit designs for the CPU center around fast operation often at the expense of additional energy consumption. The circuit designs for bulk memory (RAM) center around much lower energy consumption (often at the expense of some speed of operation) and very high density packing of storage cells. Attempting to mix both of these techniques on a single chip is possible, but much more expensive and difficult than building them independently on separate chips. But, also, there have been experimental CPU designs that attempt to do some of what you suggest, the (generally) external RAM cells are actually built on the CPU itself, right near where the data stored therein will be used by the CPU. So what you suggest has been done, it is just not the design that gained traction for our current IBM-PC architecure PC's. Also, one other item to consider. If you build a CPU and RAM on the same chip, then to offer different amounts of RAM in the market, you have to design and build different chips. But if you build the CPU separate from the RAM, you can change the amount of RAM in a system by adding/subtracting RAM modules (presuming the system was designed with plugs for adding/subtracting RAM modules) while only having to build one kind of CPU chip, and one kind of RAM module. Lower cost of production, much more market flexibility.
- salawat 8y ago
  Hmmm. Well. You're getting closer at least, and you're at least poking the right places of the field for answers, as computer architecture should get you a great deal more insight into what constitutes a CPU. One thing that you might help is to look into Integrated Circuits as well. One thing many don't actively realize is that what we call a CPU in a computer in not the only "processor" in a computer. There are other, more specialized processors which are responsible for providing interfaces between the CPU and hardware peripherals, but which are far more tightly constrained in what they can do. Best I can tell you for now though is that for the most part, you'll rapidly run into non-computing related problems the more you Try to cram into a processor die. Physics still apply, and computation creates heat. Heat degrades the performance of the system, and the more transistors you pack into the same area, the harder it is to keep cool, fabricate accurately, and guarantee the life time of. Also, on die memory opens you up to problems in term of actual circuit layouts. There have been designs where certain alignments of values in the physical chunk of Silicon can have undesired side effects on values stored near them. Rowhammer I believe the phenomena is called. So, specialization ANF separation of concerns won the day. Hope you find what you're looking for!