Intel Core i5 10600K Comet Lake vs. Core i5 Skylake / Haswell / Sandy Bridge

[Setif]

Its going to get worse before it gets better. Intel's 10nm is a mess, and they just today announced that their 7nm has been pushed to end of 2023 at the soonest. That means intel is *five years* behind the competition in lithography process.

Intel 14 nm = 10 nm of rivals
Intel 10 nm = 7 nm of rivals
Intel 7 nm = 5 nm of rivals
Intel 5 nm = 3 nm of rivals

[CochainComplex]

Intel 14 nm = 10 nm of rivals
Intel 10 nm = 7 nm of rivals
Intel 7 nm = 5 nm of rivals
Intel 5 nm = 3 nm of rivals

I agree on the fact that both processes are not comparable just by raw nm numbers. But it is obvious that intel has issues to keep up with the performance. Intel needs around 45W+ to achieve what AMD does with 15W+. This is not reflected by your lithographical numbers even if we take into consideration that powerdraw or current functions often have some square law behaviour. I would assume that 10+++nm = 7nm.

[PerformanceExpert]

Intel 14 nm = 10 nm of rivals
Intel 10 nm = 7 nm of rivals
Intel 7 nm = 5 nm of rivals
Intel 5 nm = 3 nm of rivals

This is correct when comparing theoretical density, however it ignores all other factors like power and achieved density in real chips (which is generally higher for TSMC - eg. Renoir is 30% denser than Lakefield). Also important is timeline, it's correct to say Intel is at least 3 years behind with 10nm and 7nm, and this may get worse still.

[pinguinpc]

Good test Michael but same test in gaming will be usefull and dont forget 1080p (cpu performance)

[arQon]

you can clearly see that from 2500k to 7600k there is some kind of saturation function, then Ryzen impacts and suddenly performance is overproportional developing.

Yeah, no kidding. For me, the most remarkable (and important) takeaway from the article is that if you exclude the AVX/etc crap that has almost no impact on any real-world applications (just fabricated benchmarks and a tiny handful of scientific/crypto programs), the gen-over-gen improvements since *Sandy* are basically zero other than the node shrink that came in with Ivy; and even WITH the AVX/etc aspects the generational improvement from *Haswell* (which feels like a decade ago already) to whatever Lake 7xxx is are somehow even smaller than that!

It's not like we didn't already know Intel spent years just jerking off at the top of their mountain of money and market share, but I actually thought they'd at least made SOME progress. I guess that mistaken impression just came from the clockspeed bumps on the highest-binned high-end parts. This makes me even more glad that AMD eventually got their act together again.