Computer Performance

We all seem to want our machines to be fast, and we measure performance by how fast the clock is for a start. But what really makes a machine perform effectively, and how do we improve things to get better performance out of one?

We will be examining much of the internal structure of a modern computer, but speed will not be a big part of what we look at in this course. We will examine a few tricks designers use to make things go faster, but our look at performance will be somewhat limited.

Let’s start off by looking at performance in a simple way.

Clock Speed

The rate at which the processor clock “ticks” is measured in terms of the number of times the clock switches from a “0” to a “1” and back again. One full cycle of the clock is one “tick. If you like, think of the “tick” as the moment the clock circuit moves from a zero to a one. We will call the “tock” the moment where the clock circuit moves from a one to a zero. A “cycle in this setup is a “tick-tock”. (Too long for us lazy humans, we will stick with just “tick”).

The modern Pentium processor “ticks” at a rate of around 3-4 billion ticks per second. The term we use for one cycle is a Hertz, named after a famous physicist, Heinrich Rudolf Hertz, who was the first to prove the existence of electromagnetic waves. His work directly led to the development of radio. (see [Dictionarycom17])

Other processors cannot handle a clock going that fast, so they must go slower. The Arm processor we will work with runs at about 1 GigaHertz (or GHz) - one billion ticks per second, and the AVR chip we will also examine runs at 16 MegaHertz (MHz) or 16 million ticks per second.

To get a faster computer, why cannot we just run the clock faster?

The answer has to do with how fast signals move through component in the machine. You got a taste of that when you did the “dance” earlier. It takes time to get all those atoms in the “conductor” happy!)

We will be looking at these components in some detail soon enough. For not, just accept the fact that it takes time for a single switch (transistor) to change from on to off, or from off to on. If we do not let that time pass, the machine starts to fail, and gives flaky results.

Tuning the design of a particular processor is something of an art form. Engineers work very hard to create the fastest machine they can, but they run into other limiting factors that may contribute to slowing the machine down.

Heat

As electrons move around, they generate heat. If we are not careful, that heat can build up in an area of the chip, and cause things to fail. The Cray-1 supercomputer I worked on in the 1970s had a huge refrigeration unit that pulled the heat out of the boards in the machine (hundreds of them). That worked well enough, but Seymour Cray (see :cite::dictRef7), who designed the Cray-1, wanted more speed. So the Cray 2, which powered the supercomputer center I got to run in the early 1990s, was cooled by putting the entire machine in a canister full of an electrically inert coolant! The whole machine ran while bathed in that coolant!

../_images/cray2.jpg

Those towers in the background were used in the cooling system, the coolant circulated between the machine and those towers, just like the cooling system in your car!

Over-Clocking

Some serious computer geeks like to push their machines harder than the manufacturer planned. It turns out that most machines will run faster than the speed printed on the boxes. These folks push the clock up to higher and higher speeds to see if the machine still works properly. Some even install fancy liquid cooling systems, to let them push the clock even faster. Chips can fail under this kind of stress. But sometimes they can get much faster than expected. I am not a real fan of doing this, since you never know if the machine is still working at 100%, or about to go “poof” and not run again.

Other Performance Measures

Speed is not always the most important factor in measuring a machines performance. Other factors are just as important, sometimes more so.

Power Consumption

The fastest machines around usually are desktop systems (or “big iron” like mainframes and supercomputers). The reason is simple. They need a lot of power to go fast, and being powered from the wall is important there.

But machines that must run off of a battery cannot get speed and long running times at the same time. Usually, they compromise speed to run longer. Modern laptops do a pretty good job these days. My top MacBook Pro can run over eight hours on a charge, but it is not that fast. It only runs at around 3.1 GHz to get that kind of battery life. I can push it up to over 3.4 GHz, but the battery life goes way down if I do that!

Heat

As mentioned before, making machines fast makes them hot as well. I have had laptop computers I did not ever want in my lap, they just got too hot! Again, slowing them down makes them run cooler. A big part of the design of a chip involves making sure the heat generating parts are spread over enough area that cooling technologies can keep them running cool enough.

That Cray-2 I ran back in 1992 would literally melt if the coolant stopped circulating. Since that machine cost over $8,000,000, we sure did not want that to happen - ever!

Keeping Parts Busy

An interesting part of performance involves keeping all the parts busy. We have dozens (hundreds) of components in our systems, and we really want all of them to be working as much as possible, to maximize the processing being done.

The speed we are interested in here is not just a clock tick, it is effective processing. Terms like “floating point operations per second (FLOPS) are used here. My Cray-2 was the first machine to top one billion FLOPS, but todays supercomputing machines can do a million times faster than that!

Parallel Processing

You were trained to think sequentially. Do this, then do that! The modern processor can do this and that (and several other things) at the exact same time. However, letting it do that leads to all kinds of problems if not done right. We will look at this topic in our studies.

We are not quite ready to get into this topic in detail, but this overview will get us started. We need to explore the inner workings of the machine before we can examine ways designers work to speed them up,. Aspects of performance like power consumption and heat control will be topics for an advanced course.