Von Neumann Architecture


Video Script

Up to this point in this class, we’ve talked about a certain type of computer called a fixed program computer. However, a fixed program computer has some limitations. A fixed program computer can only perform one task without being completely rebuilt and redesigned for another task. While this may seem very powerful, it is actually very limiting. So examples of fixed program computers would be Babbage’s Difference Engine. It is designed and built for one particular purpose. Of course, Babbage did design another computer, the analytical engine that would have been different it would have been programmable.

And this lies in the modern work of John von Neumann. John von Neumann was a researcher, a mathematician and engineer he was involved in a lot of fields, and his research was directly involved in the Manhattan Project, among other things. His work was also inspired by the work of Alan Turing, and he ended up working on the Edvac which was a successor To the computer. And as he was working on these systems, he started to see a way that he could design a computer that would not only be able to perform tasks based on this wiring, but also it could store program instructions in memory, just like it stores data and use those instructions to change what the computer is doing. This is the idea behind what we call a stored program computer.

In a stored program computer, the computer program itself can be stored in memory, just like the programs data. So no longer do we need to have separate bits of memory for storing the code that we’re running on our computer and the data we’re operating on, we can treat them as one in the same and this is a really revolutionary idea because it vastly simplifies the architecture of our computer down to just a few simple parts. We call this type of architecture von Neumann architecture. And in von Neumann architecture, a very simplistic view of a computer looks like this diagram. We need to have Some sort of an input device where we can collect data and input from the user. We have a central processing unit that contains a control unit that keeps track of the instructions we’re executing, and an arithmetic and logic unit that actually performs the calculation. That CPU is connected to a memory unit that stores not only the data that needs to be operated on, but the program instructions that make up the program that it is running. And then finally, the device needs some sort of output so that it can render its output out to the user either through a printout or a monitor or a sound. These parts make up modern von Neumann architecture. And if you think about your modern computers today, they are all built using this same idea. We have input devices, such as mice and keyboards, and speakers and microphones. We have a CPU, we have RAM and hard drives for our memory units. And then we have our output devices, our monitors, our speakers, all the different ways that we get data out of our computer is von Neumann architecture. In fact, there’s a really Bad joke in computer science that asks, Is there anything new in computer science? Yeah, not much since von Neumann. And it actually is kind of true.

Of course, over time, computer architecture has changed a little bit in some of the details. For example, a lot of older computer systems use what’s called a system bus to connect the CPU, the memory and the input and output devices. So in this case, when the CPU wants to get some sort of data from memory, it sends a command to the control bus that both the memory and input and output are watching. And the CPU can send a control that says I would like data, then in the address bus, it can place the data address that it would like to receive. And then the input or output device or the memory can react to that control and place the data desired into the data bus so that the CPU can receive it. And a lot of different computer systems use this particular setup and it’s really powerful if you want to add more memory are more input and output devices. As long as you can connect them to the system bus. They can all communicate.

Of course, this particular system might have a flaw, can you see what it is? One major flaw with this system is if the system bus becomes overloaded, or if you have too many devices connected to it, or if the system bus is the slowest part of the computer, it very quickly can become a bottleneck. And so more modern computer designs have changed things a bit, so the CPU and memory are more directly connected, so that we don’t have this system bus that becomes the limiting factor in our computer speed. This also leads to the concept of what we call the computer memory hierarchy. One of the things we have to remember with computer memory is the faster and more powerful the memory is, the more expensive it becomes. And therefore as much as we’d like to fill our computers up with the fastest, most powerful memory available, it would very quickly become too expensive. So instead, we try and we try and create a hierarchy of our memory where we have a little bit of the very, very fast memory Such as your processors registers, and the cash on your processors, usually in the size of a few kilobytes to a few megabytes. And then as we go down, we have some fast memory. But it’s a lot slower than that, such as our Ram or random access memory, where we might have a few gigabytes of that. And then we go further down the capacity where we have larger capacity, but slower. And so your hard drives or SSDs, are usually an order of 40 times as slow as the random access memory, which is again, an order of magnitude slower than the cache and the registers built into the CPU. And so one of the things we deal with a lot in computer science is building programs that can take advantage of this computer memory hierarchy. Can we design a program that acts upon data that’s stored in the processor cache, instead of constantly loading and unloading data from that cache? If you study things such as high performance computing, you’ll deal with this issue very, very often.