About time I make a blog post about the CPU, the most well-known part of a PC, and perhaps everything that can be classified as a “computer” ever. However, I won’t introduce to you what a CPU is, since you should have already been familiar enough with this thing, even if you are just a complete newbie to this field.
Instead, I’ll provide some commentary and explanation on some more obscure stuff, like Moore’s Law and Intel’s Tick-Tock model, since Intel has just officially released their Skylake (a.k.a. Intel 6th Generation) chips, and announced their Kaby Lake chips a while before that. More on that later.
Introduction to Moore’s Law
Moore’s Law (named after Intel co-founder Gordon Earle Moore) states: “The number of transistors in a dense integrated circuit doubles every two years.”
This, of course, is not a scientific law but more of an observation. What this means to the computing industry is that the number of transistors in a processor (CPU, GPU, etc.) will double roughly every two years, which also means that there will be a component roughly twice as powerful every two years. I mean, just look at graphics cards!
But of course, in order to fit twice as many transistors in the same place, the transistors will need to get smaller. That’s where manufacturing processes come in. A “smaller” manufacturing process will allow the production of a smaller transistor, and that’ll be our discussing topic for today.
The Intel Tick-Tock Model
Since the P6 microarchitecture (the CPU generation with the legendary Pentium 4), Intel has adopted a Tick-Tock Model, which means that for every microarchitecture, there’ll be a “Tock”, which is a chip using that architecture, and later a “Tick”, which is a die shrink of said chip (a die shrink is basically the same chip architecture made with a smaller manufacturing process). Why the “Tock” comes first is because it’s the stronger sound of the two. It goes like this:
P6 (Tick, 65nm manufacturing process) –> Merom (Tock, “Core” architecture, 65nm) –> Penryn (Tick, “Core” architecture, 45nm) –> Nehalem (a.k.a. “Intel 1st generation“, signaling the birth of the Intel Core ix series. Tock, “Nehalem” architecture, 45nm) –> Westmere (was considered a modification of Nehalem rather than a new generation. Tick, “Nehalem” architecture, 32nm) –> Sandy Bridge (Intel 2nd Generation. Tock, “Sandy Bridge” architecture, 32nm) –> Ivy Bridge (3rd Generation. Tick, “Sandy Bridge” architecture, 22nm) –> Haswell (4th Generation. Tock, “Haswell” architecture, 22nm) –> Broadwell (5th Generation. Tick, “Haswell” architecture, 14nm) –> Skylake (6th Generation. Tock, “Skylake” architecture, 14nm).
So how does the Tick-Tock model benefit everyone? Well…
- First of all, it encourages people to upgrade every other generation, depending on whether the user likes Tick or Tock. Tick offers a smaller chip with lower heat generation and power consumption, which also allows it to be more mobile, and a small performance boost. Tock on the other hand offers a completely new chip with a big performance boost. I personally prefer Tock since heat, power consumption and mobility are not really my problems.
- Second, Intel’s CPUs use a new socket and chipset for every new Tock. This also means that you’ll need a new motherboard, and occasionally even a new generation of RAM (for example, Skylake will use DDR4 and DDR3 RAM where as most CPUs these days can use DDR3 or DDR2 RAM). What this means is that the end user will most likely buy/build a new computer completely for every other Intel CPU generation, which generates a lot of profit for Intel, since they’re partnered with many pre-built PC companies.
By the way, a new Tick or Tock is expected to come out every year.
However, even though technology’s speed of progress is said to be ever accelerating, Moore’s law is not, and over time the processors industry is having a harder and harder time catching up.
For example, the i5-2550K, the best i5 Sandy Brdige chip, contains 1.16 billion transistors, whereas the i5-4690K, the best i5 Haswell chip, made roughly 2 years later, contains 1.4 billion transistors, which is… far from double that amount.
And not just that, but Userbenchmark also showed that the i5-4690K is only about 19% faster than the i5-2550K on average. Granted, both of these are overclocking CPUs and depend a lot on what the end user decides to do with them, but still…
But what about the graphics cards? Well, although graphics cards are not in our topic here, I’ll still give an example. Here’s the R9 290X compared to the HD 7970, both are the best single-GPU cards from AMD of their generation, and the R9 290X was released roughly 2 years after the HD 7970, despite only being one generation ahead:
As you can see, graphics cards can catch up with Moore’s Law much better than CPUs, but still not quite there yet. Also, while I favor AMD’s benchmarks in posts like this (since nVidia tends to ditch their old cards and leave them to rot, while making everything for their new cards, and AMD tends to keep their old cards running), here’s a comparison between the GTX 980 and the GTX 680 if you like:
Breaking the Law
Ok, so Moore’s Law is slowing down and is not quite correct anymore, but the most prominent, and also most recent, example of this deceleration is Intel announcing that they’ll break the Tick-Tock model entirely and instead release the Kaby Lake, a “2nd Tock”, about a month ago. The Kaby Lake CPUs will be the successor to the Skylake CPUs, but instead of being a die shrink, it’ll instead still use the 14nm manufacturing process, with some modifications to the architecture. Although it won’t be an entirely new architecture, it seems to still be considered a new generation. Meanwhile, Intel’s next Tick, codenamed Cannonlake, has been delayed to late 2017. Why? Because they just can’t start a 10nm manufacturing process.
This clearly shows that Moore’s Law is reaching its limit, and that processing power in a CPU won’t be increasing much for a while, not until quantum computing finally comes to the consumer market.
However, this can also be seen as good news for those of you who are planning to upgrade to a Skylake PC, because the Skylake architecture (along with its H170/Z170 chipset and DDR4 RAM) will persist for quite a while, so if you build your PC now you’ll be able to use it for much longer before you have to upgrade.
Still, this makes me think… are we really accelerating as fast as we are always said to be? What are your take on this matter? Are you excited or puzzled with Kaby Lake? Again, thank you very much for reading and supporting!