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.
Yes. That's a good example of an architectural improvement from one generation to the next. Kaby Lake, Coffee Lake, Coffee Lake Refresh, and Comet Lake are not. They're just Skylake with a better iGPU and some minor node improvements that amount to nearly nothing compared to a true node shrink.
Only if the initial release contained an issue and you experience that improvement only when the software suffers from that flaw, similar to how the p67 and h67 chipset significantly impacted SATA performance. This led to degraded bandwidth-sensitive operations or loading tasks that faced hard page faults, causing their speed to drop on Sandybridge. As a result, all affected components needed replacement. It wasn't a CPU concern but an ASIC matter; CPUs are essentially newer versions of ASICs running on refreshed or upgraded nodes with PDK updates that don’t necessarily boost IPC. They can achieve faster clocks using a redesigned VLSI core, allowing more clocks at lower voltages. This approach was implemented during the 14nm Skylake era, involving more than just VLSI redesigns.
The i7-4090k performed roughly 10% better than the 4770k. It’s not too surprising since they used the same chip. I believe the switch to solder for the thermal paste contributed to a noticeable increase in speed.
AMD suggested an average 18% boost in IPC, which aligns with what comes from architectural changes. Comet Lake to Rocket Lake showed a comparable figure in my memory, and similar numbers likely appeared in other instances. Other factors might also play a role. Data from that time isn’t extensive, but Sandy Bridge brought AVX, offering roughly double performance for tasks it supported. Broadwell saw minimal desktop adoption—didn’t you mean Haswell? There’s always a delicate balance; how much gain comes from improving the same thing versus alternative methods. This explains why more specialized functions have been added over time. Zen 5 seems to have the strongest AVX-512 support to date, finally surpassing Intel’s decent implementation since 2017 before it was removed with Alder Lake. Recently, AI acceleration features appear to be the next focus area. Switching from 4-core CCX to 8-core CCX equals 8-core CCD, which significantly aids workloads needing tighter core collaboration. Yet we’re stuck at an 8-core ceiling today if you want to sidestep complications. Early on, many weren’t fully aware; everyone was preoccupied with benchmarks like Cinebench R15 that ignored these details. I haven’t directly tested a Zen 3 desktop, but indirect data hinted at about a 10% IPC increase for Prime95-like tasks when bandwidth wasn’t an issue. I do own a Zen 3 laptop in use, though it’s challenging to compare directly with Zen 2. Edit: I haven’t examined it closely, but AMD’s hybrid APUs seem to separate the big and small cores into CCX, which isn’t ideal. CCX would be better if the processors had solid internal bandwidth.