Yes, there have been instances where a direct refresh improved CPU performance noticeably.
Yes, there have been instances where a direct refresh improved CPU performance noticeably.
If you're limited to same architecture, same node, same design, you won't be getting any significant gains. I wouldn't be surprised if there's no example matching these criteria that gets >10% boost in performance. Maybe the closest would be on the GPU side with the AMD 7970 which was relaunched as the 280x and then further relaunched again I think, with performance gains mostly being from driver improvements. Same thing happened to the RX480 / 570, but really this isn't attributable to the silicon and moreso the driver improvements.
In short, only clock changes were considered. If the clock stays steady at 10%, a 10% performance gap isn’t achieved. This could be outside the scope unless referring to node families. We’ve seen at least 14, 14+, and 14++ before they stopped using that term. Each step added roughly 100 MHz when increased. It’s not quite 10%, but it’s still meaningful. Checking data shows 7700k had a base gain of 200MHz and turbo boost of 300MHz at the same TDP compared to 6700k. The non-K models also showed similar improvements. Zen 2 was another example with noticeable upgrades. I got the 3700X when it launched, and I was into competitive overclocking then. It’s clear that my sample outperformed later releases by about 200 MHz. AMD hasn’t publicly discussed process enhancements, but they likely made quiet improvements. This applies to at least three process node generations. If you focused on power limits instead of clock, you’d notice differences too. Fixing it around clock only affects architecture. There might be a rare case where you hit 10%, like the RAM speed jump from 2133 to 2666—about a 25% increase. For workloads limited by bandwidth, that difference would show up in tests like Prime95 on large FFTs. Just ensure you compare similar RAM setups for accurate results.
A cheaper option existed, which it did. My argument centered on the fact that Intel’s architectural tweaks from Skylake to Comet Lake were minimal, so any CPU with identical core and cache setups—like Kaby Lake, Coffee Lake, Coffee Lake Refresh, or Comet Lake—could be considered a refreshed version. To acknowledge Intel, their iGPU improved noticeably. For graphics purposes, the iGPU upgrade was substantial enough to justify choosing Core i3 over Core i7. However, functionally, CPUs are quite similar. The main issue with the i3 10300 after four years is its price drop, which fell more than half from $340 to $160. If it had launched at $340, it would have been a significant oversight.
They often face issues with bad thermal performance, so they might have simply adjusted the temperature range instead of changing the model number. Nowadays, CPUs come pre-built with specific features, which means higher prices for faster models. The i7 920 offered solid performance between 2.66 and 4.0+ cores, especially noticeable when running intensive tasks. This is why, in smaller versions, failure rates tend to rise even though resources aren’t wasted. Some designs included multiple signal cores, but the real strategy is prioritizing fast cores with redundancy—if one fails, others compensate. This approach works well for performance-focused builds. It depends on whether they’re targeting desktops or mobile devices, as the most profitable ones usually secure these features. Energy efficiency in laptops differs from desktop CPUs, and speed gains are typically around 10% at best, rarely reaching 15%. Performance improvements usually stay between 1% and 10%, depending on the game. Updated September 26, 2025 by thrasher_565
A substantial enhancement in architecture would involve removing the 3D cache.
The most significant architectural progress in recent years comes from Zen 2 through Zen 3. Identical architecture, identical core count, and matching cache sizes—even when using the same socket. Achieving 20% or more improvements was frequent, sometimes even at comparable clock speeds. Intel previously delivered similar enhancements; the transition from Westmere to Sandy Bridge yielded comparable results on the same 32nm process. Broadwell to Skylake has demonstrated strong performance gains in recent applications, despite weaker initial results.
It's worth noting that the Zen 2 design had its flaws, and Zen 3 would likely be a stronger version if more consideration had been given. For context, though, it's hard to fault someone for not having a groundbreaking idea in their time. Technology usually evolves through repeated improvements. And by the way, do you want some CPU-related content?