Why the Intel Core i5 is actually a Core i9, or what is processor binning
We’ve gotten used to not having one or two desktop processors in every generation, but sometimes a dozen. They differ in number of cores, frequencies, cache sizes, integrated graphics and many other things, which makes us think that manufacturers have to create a different production line for each CPU, which is expensive and quite difficult.
In fact, as surprising as it may seem, all modern Intel or AMD CPUs are made with the same chips. Yes, physically an 8-core Core i9 and a 6-core Core i5 are as close as possible. Even apple does the same. But how and why do manufacturers do it? Let’s get to the bottom of the matter.
It all starts with silicon wafers
The path of most processors we are used to starts with a standard 300mm silicon wafer called a wafer. The wafer is made of high-purity silicon, so in addition to wearing masks, all employees wear special suits to keep contamination levels as low as possible.
Later, lasers and masks are used to etch layers of metallization, insulation, and the semiconductors themselves onto these wafers, creating the beautiful silicon wafers we sometimes see in presentations:
The cost of one of these wafers is simply fabulous, often running into several hundred thousand dollars, and the production process can take months. Hundreds of processors can be engraved on such a wafer, which are then cut up and sold. However, in reality, things are more complicated.
At least part of the wafer (5-15%) goes directly to scrap: as we remember, the CPU crystals are rectangular or square, but the wafer is round. You cannot make it square: the manufacturing process consists of creating cylinders with pyramids at the ends:
So it turns out that a part of the wafer is guaranteed not to contain the CPU and goes directly to recycling.
Okay, but surely the rest of the wafer can be cut into individual CPU crystals, placed on substrates, and sold. In practice, again, no.
Our world is not perfect.
No matter how hard engineers try, it is impossible to make a perfect wafer and “burn” processors on it perfectly. There will always be defects: somewhere from silicon inhomogeneities, somewhere from dust on the wafer surface during laser operation.
So the wafers are tested for defects after production using various methods, and finally some part of the wafer is marked for further analysis: binning. Yes, it can literally be translated as “binning”, and in fact it is not far from the truth.
Let’s go back to Rocket Lake itself, for example. The highest representative of this line, the Core i9-11900K, has 8 cores, three cache levels, various controllers and integrated graphics:
The 8 rectangles in the center are the cores, the cache is between them, the dark area on the left is the integrated graphics.
These are exactly the type of crystals that should be obtained ideally. But, what to do if a failure detects that a core of some processor crystal is not working? The easiest option is to discard that defect after cutting the wafer into separate crystals with a laser cutter.
But, as we remember, such wafers are very expensive and take a long time to produce, so engineers had a simple idea a few decades ago: let’s disable bad blocks and create cheaper processors from such crystals.
Bad integrated graphics? Let’s disable it and mark said processor with index F – it will still work, but you will need a discrete graphics card for image output. One or two cores? Let’s disable them along with the cache and call this Frankenstein a Core i5 instead of a Core i7 or a Core i9.
Thus, it turns out that “garbage digging” gives results: only crystals that are unlucky are discarded: for example, they have a bad L3 cache common to all cores or a ring bus. The rest is trimmed and sold on a “waste for profit” basis.
Even apple is not deprived of this: for example, the company sells its M1 SoC in two versions, with 7 and 8 GPU cores. In fact, from a physical point of view, the chips of both versions are, of course, the same, the company simply does not want to throw away the almost working SoC. Considering that not everyone needs high graphics performance, this decision makes sense:
You can also sell half the beer
It seems that everything is in place now: the simpler CPU lines are the rejections of the older ones. But what about the Core i7 and Core i9 of the latest generation of Intel CPUs, for example? The Core i7-11700K only differs from the Core i9-11900K in clock rates, exactly the same as the Core i5-11500 and Core i5-11600K:
How does this work? It’s simple: there is such a thing as purity of silicon: the closer a wafer-cut crystal is to its center, the purer the silicon it contains. This affects the so-called leakage currents: the lower they are, the cooler the processor will be, but also the worse it will be overclocked.
This is why companies select crystals with high leakage currents to build high-frequency CPUs: yes, they will get hotter, but they will also support higher frequencies. And for simple solutions, especially without overclocking, crystals with low leakage currents can be taken, which at the same time will reduce the power consumption of such CPUs.
How big is the actual difference in achievable frequencies between these CPUs? With well-established technological processes such as 14 or 7 nanometers, no: thus, while the Core i9-11900K reaches an average of 5.1GHz, the Core i7-11700K stops at around 4.8-4.9GHz. A 200MHz difference at these frequencies gives less than 5% real performance.
A marriage of will.
No, it’s not what you’re thinking. It’s no secret that low-end processors like the Core i3 or Core i5 outsell the older Core i7, let alone the Core i9. But the first ones are defective, you will say. Is it possible that more than half of them are defective when the tokens are made?
Of course not. Producers don’t usually disclose exact figures, but in fact the proportion of defective wafers barely exceeds 10-15%. So where is the bоMost of the basic lines of the CPU?
Of the complete Core i7 or Core i9. Yes, manufacturers’ software or hardware disable cores or integrated graphics that are running at full capacity, labeling those processors as Core i3 and Core i5. But what is the benefit then? After all, the representatives of higher ranges cost more than the basic ones, that is, the companies deprive themselves of benefits?
Due to downsizing bugs this happens: The 4 core Ryzen 3 1200 has all 8 cores.
Of course not. That’s where the economy comes into play. Let’s say a wafer chip costs Intel $50. You can sell it as a Core i9 for $400, bringing it down to a Core i5 for just $200.
But at the same time the Core i5 sells, say, 5 times better than the Core i9. That means the company will only earn $350 after selling a top line rep. Therefore, it is more profitable to sell 5 representatives of the middle line and earn only 150 dollars for each of them, but your total profit is already 750 dollars.
unlock magic
And that may have already led some users to the thought: since AMD and Intel often lock down absolutely working kernels, why not try unlocking them?
And sometimes this is successful. For example, more than ten years ago AMD produced the Athlon with two cores and the Phenom with three and four cores. You can often try to unlock Athlon to Phenom directly in the BIOS, ie get a processor with twice as many cores for the same price. Of course, this was not always possible, but there are also many lucky people on the Internet.
The Athlon is becoming… Athlon becomes… a full-fledged Phenomenon!
Or, for example, not so long ago, about 5 years ago, due to a bug from Intel on the Z170 chipset boards, it was possible to overclock 6th generation CPUs with a locked multiplier. As a result, the basic Core i5-6400 at only around 3GHz could be overclocked to 4GHz and sometimes even higher. Of course, the initially overclockable Core i5-6600K carried higher frequencies – the thing, as we’ve already found out, is leakage currents – but still a free performance bonus of up to 30% never hurts.
In anticipation of questions: no, you cannot “unlock” AMD Ryzen and the latest generations of Intel Core. Companies have gotten smarter, and now crystals initially create special bridges, which are cut off during “forced” rejection. So there is no way to enable disabled kernels programmatically.
In summary.
What is the end result? Companies will never stop making a profit, and they will literally squeeze the juice out of silicon, sometimes selling chips with almost half of their processing units turned off. On the other hand, we are better: the more active the binning is, the less money companies will lose due to rejections and the lower the cost of the tokens will be for us. So if you buy a Core i5 in the future, just know that it was a Core i9 at heart, it’s just bad luck.
