Discussion Half of the Core "gens" have all been Skylake

[NTMBK]

In the first five generations we went through Nehalem, Sandy Bridge, Ivy Bridge, Haswell and Broadwell. The following five generations have been... Skylake. Crazy.

 

[SPBHM]

That's right, Skylake was roughly 5% faster than Haswell. It was not even a good generational leap.

the original Skylake reviews at launch pretty much all used DDR4 2133, when you compare a Skylake with high memory clocks and Haswell with maxed memory speed I think the difference is much larger

 

[IntelUser2000]

the original Skylake reviews at launch pretty much all used DDR4 2133, when you compare a Skylake with high memory clocks and Haswell with maxed memory speed I think the difference is much larger

I've seen many reviews on it. Some reviews tested using higher DDR speeds without much benefit. We talked about this in the Cometlake HOT thread as well. Some of us concluded its likely due to latency settings.

And if its true Skylake scales too well with memory then it supports the theory that its a mediocre core. We all know CPU is way ahead of memory since the 90s - most of all the uarch advancement since then has been focused on reducing on the reliance of memory.

And again if that's true for Skylake that won't necessarily be true for future generations, because CPU architects address weaknesses. Typically this is among the reason why new uarchs are especially a boon to value CPUs since you don't need to spec your system in a fancy way to get the benefits as it did the previous generation.

 

[NTMBK]

I've seen many reviews on it. Some reviews tested using higher DDR speeds without much benefit. We talked about this in the Cometlake HOT thread as well. Some of us concluded its likely due to latency settings.

And if its true Skylake scales too well with memory then it supports the theory that its a mediocre core. We all know CPU is way ahead of memory since the 90s - most of all the uarch advancement since then has been focused on reducing on the reliance of memory.

And again if that's true for Skylake that won't necessarily be true for future generations, because CPU architects address weaknesses. Typically this is among the reason why new uarchs are especially a boon to value CPUs since you don't need to spec your system in a fancy way to get the benefits as it did the previous generation.

Or, we are putting 10 cores into a dual channel platform that was designed for a quad core, so we need more bandwidth to keep them all fed.

 

[liad]

It's kinda silly to say that Skylake was a bad core and we didn't noticed because AMD was worse.. because both are literally 99.99% of the x86 market and all of the servers/home computing market.

 

[DrMrLordX]

It's kinda silly to say that Skylake was a bad core and we didn't noticed because AMD was worse.. because both are literally 99.99% of the x86 market and all of the servers/home computing market.

In all fairness, what else were people going to buy?

 

[Atari2600]

4 years later, they might have to backport Rocket Lake to 14nm.

Backports don't... indeed cannot work, or if they can, they are likely only fractionally better than the existing architecture.

The transistor, power and thermal budgets just throw all the top level assumptions that went into the design straight out the window.

 

[IntelUser2000]

Or, we are putting 10 cores into a dual channel platform that was designed for a quad core, so we need more bandwidth to keep them all fed.

The same was actually claimed during the OG Skylake days. I didn't do enough comparisons to see if its due to latency or not.

In all fairness, what else were people going to buy?

Yea really. Small advancement over the current best is still an advancement. 10% in desktops, and in laptops we sometimes got a regression on the battery life department.