Given same Clock/Cores: difference between Intel CPU Generations?

rasmasyean

Distinguished
Mar 15, 2008
264
1
18,795
What's the different in performance given the same clock and cores of an Intel iX CPU?
Like if you compare say: Core i7 Quad Core 3.4 GHz (Sandybridge to SkyLake).

Is the performance difference inherent in the CPU itself? Or is it that it can support other peripheral components (e.g. faster RAM) that make the difference?

Thanks for any! :)
 
Solution
More transistors does help but that's not really an architecture improvement. You can see amd cpus with more transistors per core but less ipc. Smaller process node helps to fit more transistors and keep power usage down so tdp isn't an issue. An example:
Sandy Bridge 95W 32nm 995M transistors 216mm² die size
Ivy Bridge 77W 22nm 1.2B transistors 160mm²
Haswell 84W 22nm 1.4B transistors 177mm²
You can see with the smaller process is better tdp. More transistors on 22nm was a bigger die size and tdp. If you look back, sandy to ivy was 6% ipc increase but still roughly the same 200M increase in transistors like ivy to haswell which was an 11% ipc increase.

The improvements are much more complicated but here is an example...
There used to be only a small performance increase from one gen. to the next gen. Around 5-10%, where most was because of the higher clock speed of the new gen. CPU. However with the new Coffee Lake CPUs, that have changed a bit. The performance increase is higher than what we are used to see between the different generations.
 

rasmasyean

Distinguished
Mar 15, 2008
264
1
18,795


Is that due to the Coffee Lake's Turbo Boost going up to the high-4 GHz's?
On a side note, would you need a good cooling system to take advantage of this?
 


That and it has 6 cores (I5 and I7) and also better IPC.
A decent cooler will do unless you plan to overclock (K models only)
 
Improvements in architecture. I thought we were talking no difference in clock and cores. Coffee has 0 ipc increase.

5-10% was the ipc increase and not from higher clocks. I haven't seen a collective review like this for coffee yet but it'll be the same as sky/kaby. Sky > kaby > coffee is no performance difference per ghz because they are all the same architecture. https://www.anandtech.com/show/10959/intel-launches-7th-generation-kaby-lake-i7-7700k-i5-7600k-i3-7350k/8 All cpus get clocked to 3ghz to test ipc. There's also the difference with ram but you can see from 2133 to 2400 sky to kaby it has little effect.

Ipc increases:
Sandy Bridge to Ivy Bridge: Average ~5.8% Up
Ivy Bridge to Haswell: Average ~11.2% Up
Haswell to Broadwell: Average ~3.3% Up
Broadwell to Skylake : Average ~2.7% Up

http://www.tomshardware.com/reviews/intel-coffee-lake-i7-8700k-cpu,5252.html Kaby and Coffee were clocked the same in this review and if you check the single threaded benches, you see no difference because of the ipc.
 
Well, that is yet to be seen. Going from 2133MHz to 2400MHz RAM does affect some games quite a bit.
But anyway, you still get 2 more cores, and to be honest, who cares about clock to clock performance, no one is going to underclock their new CPU.
Coffee Lake CPUs are quite a bit faster in games than Kaby Lake, even the I5-8600K beats the old I7-7700K and that's not because of clock speed because the 8600K runs slower, it's also not just because of 2 more cores as most of todays games wont benefit from that and the 7700K has HT.
 
Yes of course some games are affected more by ram speed but generally it doesn't. We can't go by a few cases when most of them are not. Most of what the ipc bench covers wasn't gaming anyways. Cpu performance isn't limited to just gaming.

Ipc (what you now mention as clock for clock) is completely important and relevant and you even mention ipc previously yourself because of the importance. Many articles from many different sources go over ipc for a reason. When there was no clock increase like from 3770k to 4770k, the difference in performance was ipc. There was no ram difference there either. Downclocking was only relevant to test ipc and no one was suggesting someone do it for normal usage.

The question wasn't what cpu is better, it's what causes the performance increase over different gens. But we can compare them anyways. http://www.tomshardware.com/news/intel-coffee-lake-gaming-i5-8600k,35722.html Overall it's not quite a bit faster. Many of those games are taking advantage of more than 4 threads. Ht doesn't stand up to real cores so even if just 5 threads are used, the 8600k wins over a 7700k.
 

rasmasyean

Distinguished
Mar 15, 2008
264
1
18,795
How exactly do you get more IPC in the newer generations? Are some instructions are changed in such a way that they take say: 2 cycles vs. 3 cycles to execute? Is it because there are more transistors in there to make like a "division" command operate faster, whereas in a previous generation, you wouldn't be able to fit all that stuff into the CPU without making it much larger?
 


That was the article I was looking for, I knew it had been done.
 
More transistors does help but that's not really an architecture improvement. You can see amd cpus with more transistors per core but less ipc. Smaller process node helps to fit more transistors and keep power usage down so tdp isn't an issue. An example:
Sandy Bridge 95W 32nm 995M transistors 216mm² die size
Ivy Bridge 77W 22nm 1.2B transistors 160mm²
Haswell 84W 22nm 1.4B transistors 177mm²
You can see with the smaller process is better tdp. More transistors on 22nm was a bigger die size and tdp. If you look back, sandy to ivy was 6% ipc increase but still roughly the same 200M increase in transistors like ivy to haswell which was an 11% ipc increase.

The improvements are much more complicated but here is an example. https://www.anandtech.com/show/3922/intels-sandy-bridge-architecture-exposed/2
 
Solution