Intel's 8th Generation Core Family - Coffee Lake (LGA 1151, 6C/12T)

Where do you expect Core i7-8700K's Turbo to land?

  • 3.8/3.9 GHz

    Votes: 0 0.0%

  • 4.0/4.1 GHz

    Votes: 3 23.1%

  • 4.2/4.3 GHz

    Votes: 6 46.2%

  • 4.4/4.5 GHz

    Votes: 3 23.1%

  • 4.6/4.7 GHz

    Votes: 1 7.7%
  • Total voters
    13
  • Poll closed .
Nightfire said:
Lol he is so mad!!! Ahha nice rebutle with the FM series.

He was very successful in convincing people that CFL will work with z270.
Just go back to the beginning of this thread. Then when announced it wouldn't be, he clammers on saying nobody replaces their cpu only any how.
Click to expand...

This isn't the first go around with the topic. Every single time you mention backward compatibility and AMD sockets he goes, "BUT THE FM SERIES!!!! OMG AMD SUX!!!!!" I already got a week ban for calling him out on the CFL/z270 support earlier in this thread when he refused to answer the question about why there was a change in backward compatibility.

Now our two friendly Intel Zealots are preaching about power delivery, faster RAM, and better overclocking as if a high end Z270 board wouldn't run circles around the low end Z370 with all these new "features." I'd have more respect for them if they just came out and said it was a money grab on Intel's part.
 
juanrga said:
Some few people, including myself, claimed that CoffeeLake would have slightly higher IPC in memory-bound workloads, thanks to larger L3 cache and higher memory clocks. There was even an attempt to estimate the gain: it would be in ~4% range or so.

Everyone, including myself, said that the main gains in performance would come from moar cores and higher clocks.
Click to expand...

tbh thats the wrong way of thinking. increasing cache as well as faster IMC would not mean IPC is increase. though imo its a gray area because increasing cache is still part of the CPU so could still be called ipc increase depending on the workload like you said.

thinking of TLC SSD with SLC flash caching. increasing cache do make it feel like it is faster, up until the cache is filled up while the CPU cores itself isnt really any faster so ultimately once cache saturated it'll still be at the same performance, if a workload last long enough etc.

honestly i dont think it works the same way as SSD but it should be similar in some.
 
oleNBR said:
tbh thats the wrong way of thinking. increasing cache as well as faster IMC would not mean IPC is increase. though imo its a gray area because increasing cache is still part of the CPU so could still be called ipc increase depending on the workload like you said.
Click to expand...

It does as such. Same reason people see benefits with faster memory etc.

It doesn't change the actual core IPC capabilities, but you can end up with some odd results. A 5775C with L4 cache matching a 6700K for example.
 
oleNBR said:
tbh thats the wrong way of thinking. increasing cache as well as faster IMC would not mean IPC is increase. though imo its a gray area because increasing cache is still part of the CPU so could still be called ipc increase depending on the workload like you said.
Click to expand...

This is why when we're talking about IPC these days its actually "Effective IPC", and is what people want to know 99% of the time. Things such as memory speeds, cache strategies and sizes, and even scary things like register shadows and branch prediction updates will change your effective IPC. The core itself may be processing the exact same things just as quickly (same "true" IPC), but more is getting done.

All I'm saying is don't get too hung up on the idea of "true IPC", as you don't have visibility of it anyway. All that matters is instructions retired in a time interval, and everyone calls that IPC now.

Edit: yeah, what Shintai said while I was typing.
 
When the term IPC is used (or misused?), is it a measure of single precision arithmetic or dual precision? Is it correct that all microprocessor operations are in integer or FXP (fixed point) format?
 
FPGA said:
When the term IPC is used (or misused?), is it a measure of single precision arithmetic or dual precision? Is it correct that all microprocessor operations are in integer or FXP (fixed point) format?
Click to expand...

The term is not widely used except as a handwaving "this CPU gets more done in a time interval than that one". There is no precise definition, which is why it is never listed or measured in a review.

Internally, we don't talk in terms of IPC, we describe times to completion in terms of clocks for a particular operation, such as an integer add for example.

Edit, clarifying types:
x86/x64 CPUs operate on multiple sizes of integers or floating point values. With SSE and now AVX instructions they can perform the same operation on multiple data values, making them very fast for cases which can support them. Without that, you fallback to the old-style floating point ops, but the CPU still does them natively (not emulated).
 
Last edited:
juanrga said:
Some few people, including myself, claimed that CoffeeLake would have slightly higher IPC in memory-bound workloads, thanks to larger L3 cache and higher memory clocks. There was even an attempt to estimate the gain: it would be in ~4% range or so.

Everyone, including myself, said that the main gains in performance would come from moar cores and higher clocks.
Click to expand...

I went to quote the post you said that the 8700k would have faster IPC. What a shock when I learned you edited the post today!

Either way your Intel PR Marketing is failing today. Better power delivery then the Z270? Come on man now even you are grasping at straws LOL.
 
oleNBR said:
tbh thats the wrong way of thinking. increasing cache as well as faster IMC would not mean IPC is increase. though imo its a gray area because increasing cache is still part of the CPU so could still be called ipc increase depending on the workload like you said.
Click to expand...

IPC means Instructions Per Cycle. If you alter the cache or the main memory, the IPC is altered. For instance if you disable the cache on a modern CPU its performance will be greatly reduced even when it is clocked at the same frequency, because the IPC has reduced.

If you check the details of a modern model of simulation of a CPU, you can see that engineers split the IPC into components: internal (core) vs external, where external includes the effect of external caches and main memory, and the internal component of the IPC is split into core subcomponents again: often branch prediction, caches, and rest of core pipeline. Again with each component being split into subcomponents. For instance the cache IPC term splits into L1d, L1i, and L2 components. And so on. The more detailed the model the more components are simulated.

IPC --> IPC_core ; IPC_ext

IPC_ext --> IPC_L3 ; IPC_mem

IPC_core --> IPC_ss ; IPC_bp ; IPC_intc

IPC_intc --> IPC_datac ; IPC_insc

IPC_datac --> IPC_L1d ; IPC_L2

IPC_intc --> IPC_L1i; IPC_L2

...
 
Last edited:
Brackle said:
I went to quote the post you said that the 8700k would have faster IPC. What a shock when I learned you edited the post today!
Click to expand...

What? Check the thread before posting BS. I have been saying the same thing since August...

juanrga said:
Larger cache and faster RAM increase the IPC.

i7-8700k: 218cb / 4.7GHz = 46.38

i7-7700k: 200cb / 4.5GHz = 44.44
Click to expand...

Next you have an ancient attempt to explain the numbers:

juanrga said:
So far as I know CoffeLake is the same microarchitecture than Kabylake; therefore the only possibility to explain the higher IPC has to come from the larger L3 cache and the faster RAM. The faster RAM reduces access latency, which increases IPC because reduces the cycles that the cores aren't doing useful work, and a similar explanation comes from the larger L3. How much the IPC is increased depends on each workload: the IPC can vary from zero for compute-bound workloads to probably a nice quantity for memory-bound workloads.

The 7700k has 33% more L3 than the 7600k. The 8700k has 50% more than the 7700k.

IPC 7700k: 43.78
IPC 7600k: 43.33

Assuming linear relationship between L3 and CB scores

43.78 + ( 43.78 - 43.33 ) * [ 12 - 8 ] / [ 8 - 6 ] = 44.68

which implies a 2% higher IPC. The data leaked shows a 4.4% higher IPC. I don't know if the remaining 2% can be explained by the fast memory/controller. I don't have any data.
Click to expand...

I continue saying that the 8700k would have slightly higher IPC thanks to larger L3 and faster memory. I continue saying the gains would be smaller and in the ~4% range.
 
juanrga said:
What? Check the thread before posting BS.
Click to expand...

If you would only listen to your own advice....Think about how many more people might actually read your posts instead of thinking your spewing BS.

Either way 8700k Is not faster then the 7700k. The only thing it has going for it is higher clocks and more cores. IPC is still the same. So Intel has not made any progress in CPU's since the 6700k. Hell you can even say they glued on 2 extra cores on the 7700k, and called it a 8700k.....Sound familier?
 
Brackle said:
If you would only listen to your own advice....Think about how many more people might actually read your posts instead of thinking your spewing BS.

Either way 8700k Is not faster then the 7700k. The only thing it has going for it is higher clocks and more cores. IPC is still the same. So Intel has not made any progress in CPU's since the 6700k. Hell you can even say they glued on 2 extra cores on the 7700k, and called it a 8700k.....Sound familier?
Click to expand...
And it's basically the same price as the 7700k so what's the problem?
 
I mean the 8700k is a 7700k with two more cores made on 14++, the performance gains largely come from 2 more cores and higher clocks. default DDR4 speed gains is moot on a K cpu sense even skylake can do DDR4-3600 and higher no problem so long as you got ram and a motherboard good enough to do it.

I'm only excited to see if delidded we can all core 5 ghz pretty reliably, if that is the case its a good process improvement and the extra two cores is an awesome bump in all around performance easily closing the gap between consumer intel and ryzen on heavily multithreaded tasks which back when ryzen was released i predicted a 6 core 12 thread sky/kaby lake arch cpu should match or beat ryzen and so far early review leaks out of china are saying that is the case the vast majority of the time as i expected.
 
juanrga said:
What? Check the thread before posting BS. I have been saying the same thing since August...



Next you have an ancient attempt to explain the numbers:



I continue saying that the 8700k would have slightly higher IPC thanks to larger L3 and faster memory. I continue saying the gains would be smaller and in the ~4% range.
Click to expand...
FYI larger cache does not mean more IPC. L2 was increased in new HEDT but IPC took a hit because the cache is slower*.

As always it depends the task and what architecture changes happen and how they affect each task. Some tasks will increase while others will decrease with every change. Thats why we need +/- between each new generation and each CPU handles different programs differently. Thats larger why ASICs for a single use case are so damn effective.
 
Last edited:
Brackle said:
If you would only listen to your own advice....Think about how many more people might actually read your posts instead of thinking your spewing BS.

Either way 8700k Is not faster then the 7700k. The only thing it has going for it is higher clocks and more cores. IPC is still the same. So Intel has not made any progress in CPU's since the 6700k. Hell you can even say they glued on 2 extra cores on the 7700k, and called it a 8700k.....Sound familier?
Click to expand...

Wait, an 8700k is demonstrably faster than the 7700k - higher clocks and more cores like you say. I get you want a major step forward and new architecture, but you'll just have to settle for a whole lot better this iteration.
It is not like gluing things on, and takes a shit ton of work. I assure you, AMD does not think Intel did nothing this round.

This is what drives EEs to drink, you know. :(
 
kirbyrj said:
Don't even start with the Intel marketing bullshit. Better power delivery? Better OCing? Faster RAM? So do you really believe a $100 ECS Z370 board has better power delivery, better OCing, and faster RAM than a Z270 Asus Maximus IX Apex? I didn't think so.
Click to expand...

Considering how long AsRock, Asus, MSI and Giga have been making boards and specialising in overclocking, I find it EXTREMELY hard to believe these boards aren't actually amazingly well designed for power delivery, consistency and ability to tweak the voltage to very precise amounts.

It does indeed feel like total bullshit to me.

I did see the bottom of the 8700k vs 7700k (I think?) and it does have a few more resistors or capacitors or whatever the things on the bottom are.
 
AbRASiON said:
Considering how long AsRock, Asus, MSI and Giga have been making boards and specialising in overclocking, I find it EXTREMELY hard to believe these boards aren't actually amazingly well designed for power delivery, consistency and ability to tweak the voltage to very precise amounts.

It does indeed feel like total bullshit to me.

I did see the bottom of the 8700k vs 7700k (I think?) and it does have a few more resistors or capacitors or whatever the things on the bottom are.
Click to expand...

yep no doubt power delivery is perfectly good enough on good z270 boards. to this day the only possibly legitimate reason i have heard and i can't for the life of me remember where i read it now is that intel changed the management engine on coffee lake, probably due to the NSA leak detailing how to disable it and that is the sole reason for needing new motherboards but only insiders that work for mobo makers can say/tell for sure why exactly there is a change and so far only leak on that front is unnamed mobo reps stating that intel made them do it is all and theres no technical reason they can't really which i most definitely agree is a very lame move but certainly kind of normal for intel.
 
Frito11 said:
yep no doubt power delivery is perfectly good enough on good z270 boards. to this day the only possibly legitimate reason i have heard and i can't for the life of me remember where i read it now is that intel changed the management engine on coffee lake, probably due to the NSA leak detailing how to disable it and that is the sole reason for needing new motherboards but only insiders that work for mobo makers can say/tell for sure why exactly there is a change and so far only leak on that front is unnamed mobo reps stating that intel made them do it is all and theres no technical reason they can't really which i most definitely agree is a very lame move but certainly kind of normal for intel.
Click to expand...
interesting if you ever come across a source on this please pm me it.
 
SomeGuy133 said:
interesting if you ever come across a source on this please pm me it.
Click to expand...

found where i read it at.

https://www.pcgamesn.com/intel/intel-14nm-coffee-lake-release-date

their source they reference is the dutch site that attempted to run cpus cross platforms and I don't read dutch but google translate doesn't result in them making any sort of mention of it though so not sure where they came up with it to be honest. I will tell you though in my recent google searches looking for that info i see as of the past day multiple downloads are coming up for an updated version of management engine drivers for z370 so there very well could be something to that but given its backdoor/nsa leak info i doubt we'll see much if any coverage on this for a long time if its indeed true.
 
Shintai said:
It does as such. Same reason people see benefits with faster memory etc.

It doesn't change the actual core IPC capabilities, but you can end up with some odd results. A 5775C with L4 cache matching a 6700K for example.
Click to expand...

why did intel cheap out on the cpus after 5775c again with no L4cache. they need to give that to 8700k


juanrga said:
IPC means Instructions Per Cycle. If you alter the cache or the main memory, the IPC is altered. For instance if you disable the cache on a modern CPU its performance will be greatly reduced even when it is clocked at the same frequency, because the IPC has reduced.

If you check the details of a modern model of simulation of a CPU, you can see that engineers split the IPC into components: internal (core) vs external, where external includes the effect of external caches and main memory, and the internal component of the IPC is split into core subcomponents again: often branch prediction, caches, and rest of core pipeline. Again with each component being split into subcomponents. For instance the cache IPC term splits into L1d, L1i, and L2 components. And so on. The more detailed the model the more components are simulated.

IPC --> IPC_core ; IPC_ext

IPC_ext --> IPC_L3 ; IPC_mem

IPC_core --> IPC_ss ; IPC_bp ; IPC_intc

IPC_intc --> IPC_datac ; IPC_insc

IPC_datac --> IPC_L1d ; IPC_L2

IPC_intc --> IPC_L1c

...
Click to expand...

well yes, majority of software do infact store stuff in ram and then flow into cache etc so i guess ultimately the "true" core ipc will not increase, overall performance is still a jump. however what i was hoping for was some real optimization to reduce loop latency and such i guess intel already had their core architecture so optimized they simply cant do it anymore, or even if they did has very little improvement. for example broadwell to skylake, barely anything. (donno if its called loop latency not sure what the term is, basically removing unncessary things to reduce latency.)
 
im getting answers everywhere. first 10nm delayed then here says 10nm delay is now debunked but then people started saying its delayed again? so we gonna have 8core mainstream on 14nm++ or 10nm+? cause 8 cores on the current process is gonna be power hungry af
 
oleNBR said:
im getting answers everywhere. first 10nm delayed then here says 10nm delay is now debunked but then people started saying its delayed again? so we gonna have 8core mainstream on 14nm++ or 10nm+? cause 8 cores on the current process is gonna be power hungry af
Click to expand...

It sounds like 10nm is verging on impossible to me.

If they ever do pull it off and have 8 cores, it will be a big leap over coffee lake (if they can cool the thing) but I suspect it'll be the last big tech leap in many, many,many years. It sounds like 14nm to 10nm has been a real nightmare.

I reckon Intel will be on 10nm for at least 5+ years.
(I mean REAL 10nm, not AMD "7nm" stuff)
 
AbRASiON said:
It sounds like 10nm is verging on impossible to me.

If they ever do pull it off and have 8 cores, it will be a big leap over coffee lake (if they can cool the thing) but I suspect it'll be the last big tech leap in many, many,many years. It sounds like 14nm to 10nm has been a real nightmare.

I reckon Intel will be on 10nm for at least 5+ years.
(I mean REAL 10nm, not AMD "7nm" stuff)
Click to expand...

10nm has been a nightmare for Intel, I would say there is a real chance it got delayed again as they need lots of masks to do it. My guess they jumped ship and went to EUV technology to reduce costs and improve yields.
 
oleNBR said:
im getting answers everywhere. first 10nm delayed then here says 10nm delay is now debunked but then people started saying its delayed again? so we gonna have 8core mainstream on 14nm++ or 10nm+? cause 8 cores on the current process is gonna be power hungry af
Click to expand...

they is no new delay...it was click bait sensalization. 10nm ULV laptop parts are on track as they "always" have been. 10nm desktops have been canned and replaced with coffee lake and possibly a future 8 core coffe lake instead of cannon lake with icelake to follow. on a process and arch step. Assuming intels charts are to be taken literal 10nm+ will be slightly weaker than 14nm ++ due to it not being as mature.

AbRASiON said:
It sounds like 10nm is verging on impossible to me.

If they ever do pull it off and have 8 cores, it will be a big leap over coffee lake (if they can cool the thing) but I suspect it'll be the last big tech leap in many, many,many years. It sounds like 14nm to 10nm has been a real nightmare.

I reckon Intel will be on 10nm for at least 5+ years.
(I mean REAL 10nm, not AMD "7nm" stuff)
Click to expand...

Gideon said:
10nm has been a nightmare for Intel, I would say there is a real chance it got delayed again as they need lots of masks to do it. My guess they jumped ship and went to EUV technology to reduce costs and improve yields.
Click to expand...
can both of you please stop this ridiculous sensationalization of everything...... it provides nothing useful to this thread and its just childish

also gideon stop making things up with 0 factual basis. it is getting tiresome to read and i rather not want to ignore you like ignoring abrasion

abrasion your making stuff up too and tons of PR and historical facts and scinfic data says what you said was beyond retard.

A lot of the info is released and your just spewing pointless ignorant stuff to new members and spreading patently false lies.
 
Umm, if the chipset is actually called Z390, then whatever CPUs that launch alongside it will be Coffee Lake and 14nm++.

There's nothing any of you can say that will convince me Intel will launch a 10nm+ Icelake 8C on a platform called Z390.
 
oleNBR said:
im getting answers everywhere. first 10nm delayed then here says 10nm delay is now debunked but then people started saying its delayed again? so we gonna have 8core mainstream on 14nm++ or 10nm+? cause 8 cores on the current process is gonna be power hungry af
Click to expand...

10nm is not delayed. That's the official statement, including for financial reporting.

I am sure however you see plenty of clickbait and wishful forum posts with delays.
 
TaintedSquirrel said:
Umm, if the chipset is actually called Z390, then whatever CPUs that launch alongside it will be Coffee Lake and 14nm++.

There's nothing any of you can say that will convince me Intel will launch a 10nm+ Icelake 8C on a platform called Z390.
Click to expand...

That's like saying Z370 have to be Kaby Lake CPUs and cant be Coffee Lake on 14nm++.

Icelake is on track for H2 2018 and taped out long ago.
 
TaintedSquirrel said:
Umm, if the chipset is actually called Z390, then whatever CPUs that launch alongside it will be Coffee Lake and 14nm++.

There's nothing any of you can say that will convince me Intel will launch a 10nm+ Icelake 8C on a platform called Z390.
Click to expand...

Shintai said:
10nm is not delayed. That's the official statement, including for financial reporting.

I am sure however you see plenty of clickbait and wishful forum posts with delays.
Click to expand...

so what are we seeing here? the next 8c mainstream will be on 14nm++ and be compatible with z370?
remember also what eurocom said and did mention 8c 10nm on z390..
 
SomeGuy133 said:
they is no new delay...it was click bait sensalization. 10nm ULV laptop parts are on track as they "always" have been. 10nm desktops have been canned and replaced with coffee lake and possibly a future 8 core coffe lake instead of cannon lake with icelake to follow. on a process and arch step. Assuming intels charts are to be taken literal 10nm+ will be slightly weaker than 14nm ++ due to it not being as mature.




can both of you please stop this ridiculous sensationalization of everything...... it provides nothing useful to this thread and its just childish

also gideon stop making things up with 0 factual basis. it is getting tiresome to read and i rather not want to ignore you like ignoring abrasion

abrasion your making stuff up too and tons of PR and historical facts and scinfic data says what you said was beyond retard.

A lot of the info is released and your just spewing pointless ignorant stuff to new members and spreading patently false lies.
Click to expand...

I don't know who this guy is, or what his problem is but ummm lol ok fella, ignore away?
 
Gideon said:
10nm has been a nightmare for Intel, I would say there is a real chance it got delayed again as they need lots of masks to do it. My guess they jumped ship and went to EUV technology to reduce costs and improve yields.
Click to expand...

it's pretty apparent that it's been extremely difficult. Tick Tock is super dead. 14 has dragged on for ages, 10 will also. Just means they need to work on the design much more to eek out performance.
 
Brackle said:
Either way 8700k Is not faster then the 7700k. The only thing it has going for it is higher clocks and more cores. IPC is still the same. So Intel has not made any progress in CPU's since the 6700k. Hell you can even say they glued on 2 extra cores on the 7700k, and called it a 8700k.....Sound familier?
Click to expand...

Moar cores, higher clocks, and higher IPC. I already explained you twice why I expect the IPC to be slightly higher (about 4%) in memory bound workloads. I expect proper reviews to confirm this, just as they confirm that Broadwell and Skylake have higher IPC when memory is overclocked.

Gideon said:
10nm has been a nightmare for Intel, I would say there is a real chance it got delayed again as they need lots of masks to do it. My guess they jumped ship and went to EUV technology to reduce costs and improve yields.
Click to expand...

What part of 10nm CoffeeLake is in production is not still understood?

What part of Intel has made 10nm ARM chips and they run at 3.2GHz is not still understood?
 
Last edited:
SomeGuy133 said:
FYI larger cache does not mean more IPC. L2 was increased in new HEDT but IPC took a hit because the cache is slower*.

As always it depends the task and what architecture changes happen and how they affect each task. Some tasks will increase while others will decrease with every change. Thats why we need +/- between each new generation and each CPU handles different programs differently. Thats larger why ASICs for a single use case are so damn effective.
Click to expand...

Skylake-X increased the L2 but reduced the L3. There are also fundamental changes in interconnecs and in the cache policies (inclusive vs exclusive). All that affects IPC.

That is not the case with Kabylake --> CoffeeLake.

The muarch is the same. L1 and L2 caches are the same, interconnects and cache policies are the same. And the ammount of L3 cache per core is the same. CoffeeLake get more L3 cache only because has moar cores. This L3 cache is shared. Thus memory-bound workloads that don't use all cores would get a small bump in performance due to having more cache per core, and thus rediucing the number of accesses to main memory.
 
oleNBR said:
im getting answers everywhere. first 10nm delayed then here says 10nm delay is now debunked but then people started saying its delayed again? so we gonna have 8core mainstream on 14nm++ or 10nm+? cause 8 cores on the current process is gonna be power hungry af
Click to expand...

2017
Desktop: CoffeeLake (14nm++)
Mobile: CanonLake (10nm)

2018
Desktop: IceLake (10nm+)
Mobile: IceLake (10nm+)

Shintai said:
10nm is not delayed. That's the official statement, including for financial reporting.

I am sure however you see plenty of clickbait and wishful forum posts with delays.
Click to expand...

The funny part is that this 10nm-is-delayed FUD is coming from the same people that promised us Zen2 on 7nm next year, despite AMD official roadmaps said otherwise.
 
Last edited:
You must log in or register to reply here.
Back
Top