They haven't improved much because of limited resources and technical constraints.
They haven't improved much because of limited resources and technical constraints.
I notice most new processors recently show similar single-core scores in Geekbench. My 8-year-old i5 matches the Threadripper 1950X in that metric. Core counts are rising, yet single-core performance seems flat. It’s puzzling why GPUs keep improving while CPUs lag behind.
GPUs rely heavily on parallel processing. Their speed improves with more cores added each generation, not just single-core performance. When examining how computers handle assembly and binary code, you'll notice a natural limit to calculation speed. Optimization techniques like path prediction help, but they can't overcome this boundary. The only way to boost core performance directly is by raising the clock speed.
Performance is more than just clock speed. IPC determines how much work each clock cycle can handle. Currently, AMD edges out Intel in IPC, meaning their chips deliver better results even when running at the same frequency. A small difference, but a clear advantage.
IPC and clock speed are the main constraints, with IPC struggling more as transistor sizes shrink (we're likely approaching 5nm soon), while clock speed is influenced by silicon maturity. Intel is boosting clock speeds on Comet Lake chips thanks to advanced manufacturing. When Moore's law was still relevant, significant gains in per-core performance were observed. The 9590 could run at 5GHz, but its single-core and multi-core performance lagged behind Intel's offerings, especially in multithreaded workloads.
I5 chips using 32nm technology typically have a die size about five times larger than a TR3’s 7nm core. There’s no limit to increasing it further, but scaling width instead of height is common. GPUs follow the same idea—smaller dimensions let you pack more processing units onto one chip. A smaller footprint also reduces heat output, allowing a slight increase in clock speed.
intel must surprise everyone soon, I’ll know when i’m convinced. i’m cautiously optimistic that zen 3 will finally let amd keep up. based on the 3600xt’s 4.6-4.7ghz, a 16-core chip running at 4.5 would be impressive thanks to better IPC. i still feel a bit behind with zen 3’s single-core performance, but i’m hoping for something unexpected. the real draw of zen is its power consumption, which stays just under max ocs—even at single core it’ll be quite impressive.