Why 10GHz CPUs Are Impossible Probably

If you grew up during the 90s or early 2000s, you probably remember that a computer's value and performance were defined by one number, its clock frequency. Back then, a 500 MHz computer was considered twice as fast as a 250 MHz one. This wasn't just clever marketing, it was a technical reality. For several decades, computer performance increased at incredibly predictable rates, driven by two fundamental principles of engineering. The first is the Moore's law. In 1965, Gordon Moore observed that our manufacturing capabilities were advancing so fast that we could double the number of transistors on a microchip roughly every two years. This consistent doubling of components allowed the industry to build increasingly complex and powerful processors. However, this rapid increase in transistor count created a problem. If you pack more components into the same small space, you risk generating more heat than a silicon can safely dissipate. This is why the second, less known principle was so important. It's called Denard's scaling. In 1974, Robert Denard and his colleagues at IBM discovered a remarkable physical property. As you shrink a transistor, you can also proportionally reduce its supply voltage while keeping the transistor completely stable. To understand why this changed everything, we have to look at the dynamic power formula for a CMOS transistor. In this equation, the power consumption P is determined by three key variables. The capacitance, which represents the electrical charge a transistor must hold to switch states. The voltage, this is the supply voltage which has a large impact on power consumption because it's squared. And the clock frequency, or how many times per second a transistor can switch. When you reduce the physical area of a transistor, you inherently decrease its capacitance. And since the internal components are smaller, the supply voltage can be lowered. Because voltage is squared in this formula, these two reductions dramatically drop the power consumption of each individual transistor. But instead of using these efficiency gains to create ultra-low power computers, engineers realized it was far more beneficial to reinvest those savings. They chose to increase the clock frequency while reducing the transistor size and voltage. This meant that transistor density could double with each generation, and here's the most remarkable part, the chip's total power density remained constant. It was the ultimate get-out-of-jail-free card for computer architecture. This balance allowed engineers to increase clock speeds by roughly 40% with each new computer generation without exceeding the thermal limits of the CPU. The result was a period of almost linear performance growth. During this time, software developers rarely had to worry about optimizing their code. They knew that the hardware would simply be twice as fast in a few years. However, between 2004 and 2006, this era of free performance hit a physical limit and began to collapse. For a transistor to function as a reliable switch, it requires a minimum voltage to distinguish between an on state and an off state. This is known as the threshold voltage. As transistors continued to shrink, engineers eventually hit a wall. By the mid-2000s, the supply voltage had dropped to around 1 volt, and it simply couldn't go any lower. If they pushed it further, the gap between the supply voltage and the threshold voltage would become so small that transistors became unstable or prone to errors. At the same time, a second problem emerged, leakage current. As the physical layers of the transistor became only a few atoms thick, electrons started to leak through the insulation even when the transistor was supposed to be off. This phenomenon, known as quantum tunneling, meant that electricity was constantly slipping through the gates. The result was a huge spike in static power, or in other words, electricity was wasted as pure heat even when the chip is doing nothing. Between the inability to lower voltage and the rise of leakage current, chips began to generate more heat than cooling systems could safely remove. This was the power wall. We had reached a physical ceiling where simply increasing the clock frequency would generate enough heat that would literally cause the silicon to melt. The power wall is one of the reasons computer architecture shifted its focus from a single high-frequency processor to multiple medium-speed cores. In architectural terms, the focus has shifted from latency, how fast a single task completes, to throughput, how many tasks can we complete simultaneously. From this point on, hardware could no longer provide free performance gains. The responsibility shifted to software developers to write code that would handle multiple threads of execution. Nowadays, multithreading is fully integrated into the hardware, operating systems, and user programs, to the point where we largely take it for granted. Learning computer science and architecture can definitely be confusing sometimes. But don't worry. To help with that, I've been creating a series of short ebooks. Each one focuses on just one topic at a time, to keep things simple. I've filled them with illustrations, examples, and some really interesting facts to help everything click for you. If you're looking for a quick way to understand how computers work, you can find a link to my ebooks in the description below. I hope this video was helpful, thanks so much for the support, and I'll see you soon!