I've been a PC hardware enthusiast most of my life, and part of the fun for me was putting the best CPUs into high-end motherboards and overclocking the snot out of them. Sometimes it was to eke a little more performance out of a laptop CPU when that was all I had to game with, other times it was to compete against my friends circle to see who was the best, and others it was simply for the thrill of seeing numbers go up and the PC boot and pass benchmark tests without issue.

But over the years, the thrill of overclocking has waned for me. It's a combination of things that made me realize that I don't need to push the boundaries as much anymore. I wouldn't exactly say I'm happier because of it, but I'm content with the performance that modern hardware is providing straight from the factory, and in some cases, overclocking was actively hurting that performance baseline.

I don't have anything to prove anymore

Architectural changes have reduced headroom unless you have exotic cooling

When CPUs ran at slower core frequencies, it was fairly common to get double (or more) performance by increasing the core frequency, voltages, and a few of the associated voltages to the chipset and parts of the motherboard. If you wanted to chase numbers, you'd even downclock your RAM and only use one stick, so the memory controller wasn't being strained and you could be sure that the thermals of the CPU were solely from the cores. Those days are almost over, and have been really for some time now, ever since AMD released Ryzen in 2016.

Since then, a few things have happened. Architectural improvements have increased factory boost clocks to the "magic" 5GHz or even close to 6GHz. That's the region where you're hitting fundamental physical limits for the silicon and the transistors that make up the CPU, making it necessary for sub-zero cooling with liquid nitrogen and other exotic means to increase past that barrier.

Then the die shrinks from 14nm to 7nm and to 4nm improved things in the factory, where everything is built to increasingly lower tolerances, leaving very little unused headroom for overclockers to play with. And those smaller, tightly-packed transistors mean increased thermal density, which makes cooling CPUs even harder when pushed to their limits.

Plus, automatic boost clocks have changed the equation

A decade ago, I would have had to work hard to get a flagship CPU to run above 5GHz for daily use. Some CPUs couldn't maintain that at all. Compared to modern Intel CPUs that could boost to 6GHz without changing any settings, I started to wonder why I should bother. The challenge with modern CPUs isn't to get to stratospheric core frequencies; it's to find unused room to break past the factory parameters.

That doesn't mean I run my CPUs stock

There's still some performance gains to be had, just not by manually cranking frequencies and voltage

While I like the relative fire-and-forget nature of modern boost clock algorithms, that doesn't mean I'll leave my CPUs at stock settings. Both Intel and AMD respond well to undervolting the core, reducing the power being fed to improve thermals, so those boost clocks can do their job. For Ryzen builds, I use PBO and ECO to optimize the boost curve and reduce the TDP of the chips, giving cooler running while sustaining boost clocks.

Intel is in a weird place right now, and my Core Ultra 9 285K is rarely used. Successive BIOS updates have fixed the memory issues I saw, without dialing up the DDR5 voltage to get the rated XMP settings. But boost performance is inconsistent no matter what I try, and it's seemingly inconsistent between the efficient and performance cores. I do have another board on the way to check this again, but even then, I don't have high hopes for any meaningful overclocking boost.

The huge gains of earlier generations are gone

I've been overclocking CPUs for a long time, although not so long that I had to change the motherboard's crystal oscillator to increase the clock speed. Early CPUs were interesting, but I cut my chops on the AMD Duron and Athlon CPUs of the early 2000s, back when you could unlock the clock frequency using a thin pencil. You could reattach L1 bridges that were lasered off in the factory with a 0.5mm mechanical pencil, and then you could set any clock frequency you wanted, as long as it would boot reliably afterward. You could get a reliable 25% or more out of the CPU like this, and nobody who hasn't seen it done will ever believe you.

The last CPU I've owned that could overclock reliably was a 4th-gen Intel Core i7-4790K. With a base frequency of 4GHz and a boost clock of 4.4GHz, it was no slouch at factory settings, but cranking the core voltage up to 1.4V managed a stable all-core of 5.2GHz that I ran daily for almost two years. Since then, I've owned a couple of dozen CPUs from both AMD and Intel, and the overhead for huge gains has gone.

AMD routinely cranks Ryzen to as high as the silicon can handle in terms of boost clocks, and increasing the core frequency manually often works against you for performance gains. The boost clocks are better when only one or two cores are needed, and I've found time and time again that an all-core overclock hurts overall performance.

And Intel hasn't had a good overclocking CPU since the company got stuck on the 14nm node for years and had to find performance gains by optimizing the silicon and cranking voltages and clocks, essentially factory overclocking everything. The only CPUs that didn't get this treatment were non-K models, and you can't overclock those anyway.

Modern overclocking of CPUs has hit diminishing returns

Before you get the pitchforks out, I'm not saying don't overclock your CPU. Overclocking is a fantastic hobby, and one that I have enjoyed for many, many years. In some ways, I'm sad that modern hardware is already juiced from the factory because the thrill of seeing measurable gains never gets old.

That's why I'd say if you want to overclock a CPU, find an older platform and use that. AMD's Bulldozer and Vishera CPUs are good candidates; they might not have measured up to performance claims, but the single-core overclocking potential exists. Same with early Intel Core CPUs, going as far back as the Sandy Bridge i7-2600K. Those can be a lot of fun, and don't require exotic sub-zero cooling to get impressive overclocks.