You may have heard about these words or seen them in some articles before, but asked yourself: What do they even mean? In this article we’re going to talk about some CPU Terminologies and what they mean
1. Clock (“Frequency”)
CPUs don’t randomly process instructions every time they have a new instruction set, they work in clocks. Think of it as the following:
.The CPU does a “refresh” or clock, it fetches for instructions that need to be worked on.
. Next, the CPU does another clock where it gets the information and puts in the the Cache (We’re going to explain it later)
. Then it does another clock, it fetches information AND moves the information from the Cache to be processed
. In the next clock, the CPU does the same things to keep the pipeline full do it doesn’t waste power and outputs data
That’s how CPUs work, they work in steps, and the number of those steps are measured by how many times they reoccur in a second and that’s why we Hertz (Hz) for that. Hertz describes how many times something happens again and again a second. So instead of saying “Oh this CPU refreshes 4 Billion times a second!” you can just say: “This CPU is clocked at 4GHz”
2. IPC (Instructions Per Clock)
Contrary to what most people think, higher clocked CPUs even of similar core count ARE NOT always better. Because of this thing: IPC or Instructions Per Clock, which describes how many instructions or “information” a CPU can actually get in one clock or “refresh”. For example: Let’s say that the CPU named “X” can get 10 IPC and is clocked at 4.5GHz and CPU “Z” can get 20 IPC but it’s clocked lower at 3.5GHz. If you do the math, the CPU “Z” gets more instructions even at a substantially lower clock! Let’s do it on our own and let’s use the clocks in GHz to simplify thing:
4.5GHz * 10 = 45 Billion Instructions
3.5GHz * 20 = 70 Billion instructions
So as you can see, lower clocked CPU can still outperform higher clocked ones. So don’t use the CPU clock as a measure and look for benchmarks yourself
Traditionally, back in the early 2000s, CPUs only had one core, which is self explanatory: You have one single ASIC doing everything. Then Intel pushed the industry further by introducing the “Core 2 Duo” series, which technically meant that it had 2 processors inside one single “CPU” so it can have a much higher computing horse power. And the core count has been increasing since. But like clocks, more cores DO NOT ALWAYS MEAN more performance as most applications, like games, don’t use more than 4 cores most of the time and even if they got optimized for a higher core count, they will still be limited by the APIs until we all shift towards DX12 or Vulkan
A lot of people mistake the thread count on their CPUs with core count. Cores are physical processors present on the CPU die, and threads are like the “logical” processors, what software actually “sees” when you run it. Although it’s true that a lot of CPUs have the same number of cores as threads, one with Intel’s “HyperThreading” technology have double. It doesn’t mean that Intel did some kind of magic to make their CPUs have more numbers when they don’t or something, but rather “tricking” the software with a piece of technology that makes one CPU appear as two. This is was used for one sole reason: More efficiency. You can think of it as giving the CPU two hands: Make it grab as much instructions as it can so it works as much as it can and you don’t “waste Silicon” (More on that later)
5.TDP (Thermal Design Power)
A lot of people confuse TDP for power consumption (Mainly because they’re both measured in Watts and the power requirements also look like the same) but TDP number means the required thermal solution has to at least be able to dissipate as much heat, and by the laws of thermodynamics everything that uses power loses it in some from or another, in this case when a CPU uses electricity it loses it under the form of heat, also measured it Watts.
Cache is a super-fast type of memory only found in CPUs, it’s even faster than RAM but it’s also really expensive to manufacture and that’s why it only appears in CPUs. It’s basically a little pool of storage (Always in MBs because it’s super expensive) to store important instruction awaiting processing or needed in order to complete an existing process
Think of transistors as wires, wires that are linked to each other in a certain way to form what called a “logic gate” which are used to actually do things with the CPU. Transistors are made out of semi-conductors, most famous and popular is Silicon. Manufacturers simply take a quantity of sand (Pure sand to get the best possible result) heat it up to insane temperatures so it melts into Silicon which later gets gathered into one big cylinder of Silicon, then thing layers of it are cut to make the transistors. The process is a bit complicated and may require a lot of explanation so I’ll do a separate article on it.
8.14nm (14 NanoMeter)
When looking at the spec sheet for a CPU you may have stumbled upon something called “lithography” followed by a number like: 14nm, 28nm etc… Typically, that number was used to describe the size of a transistor. But after the fast and huge advancement in technology and shrinking processes it no longer meant that, since transistors now in something like a 6700K are something like 8nm wide, but rather the size of the tools used in the cutting process and thus the gaps between the transistors.
9.ASIC Quality (Or the “Silicon Lottery”)
Nothing we make is perfect, and that implies to Integrated Circuits as well. Some cuts into the transistors may sometimes be heavily flawed and result in “voltage leaking” meaning that it doesn’t deliver or transfer the current efficiently, and it’s called ASIC quality which can vary from a CPU to another and it’s measured in %. While it usually doesn’t matter in locked CPUs, overclockers look for the best possible ASIC quality, because the higher it is the more effecient the CPU will be, meaning that some CPUs may require MORE Voltage than another CPU with theg same model number (Same SKU) to deliver the same clocks. Think of it like this: We have two 6700K, one has a 75% ASIC quality and the second has a 90% ASIC quality. If you try to overclock the first, it will require about 1.35V to reach 4.6GHz but teh second one will reach it at 1.25V. That is a SUBSTANTIAL difference in power consumption and also heat because as you increase voltage you’re also increasing your temperature. And that’s why it’s called the Silicon lottery: Because you’re taking a gamble whether or not you’ll get a a really good CPU, a typical one or a very bad one for overclocking
ASIC quality also has something to do with binning. In fact, it’s the sole reason why this term exists. When you look at the high-end server grade parts from Intel for example you’ll see that they have to pass more “certification” or “validation” tests. And because this type of hardware needs to be as efficient as possible, there are higher standards when it comes to ASIC quality. So when a Xeon chip fails to exceed the minimum ASIC quality or has a flout out faulty core, it gets “binned” or downgraded to a consumer product. For example: a Xeon has a core that has a very low ASIC quality. Intel disables the additional cores and makes it a 4-core 6700K. And because Xeons are of much higher quality, you will have a VERY high ASIC quality percentage, thus making your CPU very good for overclocking. And people like AMD have been sorting out highly binned CPUs for decades. For example: The 8320, 8350 and 8370 are basically the exact same chip, but with different ASIC quality percentages.
The topic is still deep and we’re not even scratching its surface in terms of the techniques and physics behind it. Leave a comment if you want us to cover something in particular
Here are some websites that could help you gain more information about CPUs: