Intel Kaby Lake Core i7-7700K IPC Review @ [H]

IPC have been known to be the same for like a year. So if you expect 2 equally clocked CPUs to perform different you set yourself up for a disappointment long ago.

The 14nm+ process allows for higher clocks and lower power consumption. Out of the box a 7700K is 5% faster in 2-4 threads and 7.5% faster in 1 thread over 6700K. It also seems to overclock better in average with those 200-300Mhz.

the biggest change is in mobile, with higher clocks and higher sustained boost. So you end up with 20-25% performance increases. For server the same pretty much applies. First server part is Skylake-EP tho and Skylake-X for HEDT. 32 cores with a 2.1Ghz base vs Broadwell with 24 cores and 2.2Ghz base.

With Coffee Lake in a year you will see the same IPC again. However you will get 6 cores instead of 4 on LGA1151. If you are looking for IPC increases outside caches, you have to wait till Icelake.
Seems to overclock better?

Please show me a large sample of retail chips that prove this. Oh wait, its not released yet. Intel would never send out cherry chips though would they.............

That said, the higher clocks are not proven stable. They are quick test runs. Anyone can get a higher clock running a short 3d app or cpu test. Show me it can maintain constant gaming (or whatever it is you need it to do) for a period of time.
 
Seems to overclock better?

Please show me a large sample of retail chips that prove this. Oh wait, its not released yet. Intel would never send out cherry chips though would they.............

That said, the higher clocks are not proven stable. They are quick test runs. Anyone can get a higher clock running a short 3d app or cpu test. Show me it can maintain constant gaming (or whatever it is you need it to do) for a period of time.

Private people already got retail 7700K chips here and there and they clock well.

And sending out cherry samples only works the first time. Next time its cherry picked vs cherry picked. If we presume it was the case to begin with.
 
Private people already got retail 7700K chips here and there and they clock well.

And sending out cherry samples only works the first time. Next time its cherry picked vs cherry picked. If we presume it was the case to begin with.
Time will tell but dont assume its some magical clocking chip. History shows this has happened in the past with only mediocre clocks being the norm.
 
Wish I had grabbed the 6700K when it was on sale for $260 after Thanksgiving. :/
If Kaby can reliably hit 5+ GHz that will be some consolation.

Bring on Zen, I guess.
 
Are there any new hex/octo cores on the way? Kaby Lake-E or something. Maybe those will be good.
 
Thanks for the review. Looks like Intel just released their Bulldozer.
 
I'm going to be on this 4770k for a while... If it doesn't stop trying to kill itself. Lost a motherboard a few months ago when my MSI 1080 blew up and on Friday my H110i GT pump shat the bed and now I'm back on a stock cooler.
 
What Kyle does to less than "golden" Kaby Lake 7700Ks..

H_Kyle_smashing_intel_kaby_lake.jpg
 
Hi Kyle, I'm writing an app that will read the IPC directly from the processor and display it in real time. I'm trying to get it ready in time for Zen's release. Is this something you would be interested in for your Zen review? I don't think anyone else is able to do this.

I've attached a screen shot from a few different benchmarks to give you an idea of what it will do. There is all this talk about IPC (especially with Kaby Lake and Zen) but so far nobody that I know of has actually measured it. So I think I'm the first. Anyway, I think it would be cool to see this in your Zen review. Let me know.
 

Attachments

  • CoreDamageX-blender.jpg
    CoreDamageX-blender.jpg
    171.3 KB · Views: 103
  • CoreDamageX-cpuz-178.jpg
    CoreDamageX-cpuz-178.jpg
    170.2 KB · Views: 75
  • CoreDamageX-aots.jpg
    CoreDamageX-aots.jpg
    165.9 KB · Views: 74
I think that Intel actually wants people move to Zen, they might have something lot faster in stores and after people have moved to Zen, they release their Zen killer, which will again make people to pay a lot to get fastest and newest.

I regret of getting Skylake to replace Core2Duo, I should of moved to X99, but as Skylake was the newest I though it will get faster CPU's (especially for single core performance), but looks like Kaby Lake is not going to give a lot. Maybe Kaby Lake i3 K model would be more interesting for single core performance needs as there is not much point paying from i7 when only situations where I find CPU needing more power are those where software uses only 1 thread.

For majority of enthusiast certainly Kaby Lake i3 would be too weak even when clocked, some like from soundless pc builds and for those it might be interesting option as is current Skylake i3, I'm running semi passive cooling on mine and only noise is buzz from Seasonic fanless PSU (don't buy), my room is enough quiet to hear power adapters making noise at wall socket or cat breathing, so even slight noise from pc is heard quite well, maybe that is why Seasonic was a disappointment.

So in contrast to this background Kaby Lake does give some improvement over Skylake as it might be possible to get bit more computing power for close to 0db, but there are not many (if at all?) reviews that try to find that, maybe because there are not so many with these needs. However Kaby Lake does not deliver much improvement even on that, so don't see point from moving to it from Skylake. So maybe just wait for socket 2066 then, maybe that is where Intel wants people move from Zen, after all more money to market would help revive pc.
 
I suppose perspective is subjective. Care to elaborate why you feel it's a success, despite it being a carbon-copy of Skylake?

Take a look on notebookreview with 20-25% higher speed.

And else look at the desktop lineup. The speedbump isn't bad at all +200-300Mhz here and there.

To call it Bulldozer is simply ignorance.
 
I think that Intel actually wants people move to Zen, they might have something lot faster in stores and after people have moved to Zen, they release their Zen killer, which will again make people to pay a lot to get fastest and newest.

I regret of getting Skylake to replace Core2Duo, I should of moved to X99, but as Skylake was the newest I though it will get faster CPU's (especially for single core performance), but looks like Kaby Lake is not going to give a lot. Maybe Kaby Lake i3 K model would be more interesting for single core performance needs as there is not much point paying from i7 when only situations where I find CPU needing more power are those where software uses only 1 thread.

For majority of enthusiast certainly Kaby Lake i3 would be too weak even when clocked, some like from soundless pc builds and for those it might be interesting option as is current Skylake i3, I'm running semi passive cooling on mine and only noise is buzz from Seasonic fanless PSU (don't buy), my room is enough quiet to hear power adapters making noise at wall socket or cat breathing, so even slight noise from pc is heard quite well, maybe that is why Seasonic was a disappointment.

So in contrast to this background Kaby Lake does give some improvement over Skylake as it might be possible to get bit more computing power for close to 0db, but there are not many (if at all?) reviews that try to find that, maybe because there are not so many with these needs. However Kaby Lake does not deliver much improvement even on that, so don't see point from moving to it from Skylake. So maybe just wait for socket 2066 then, maybe that is where Intel wants people move from Zen, after all more money to market would help revive pc.

If you want something faster than Kaby Lake its called Skylake-X.
 
Intel actually wants people move to Zen

Zen with its lower IPC (remember this is supposed to be between Ivy Bridge and Haswell if AMD was telling the truth about the 40%) and lower clocks for the 8C / 16T versus Intel's mainstream will not make it the CPU that many people are dreaming about.
 
Last edited:
Take a look on notebookreview with 20-25% higher speed.

And else look at the desktop lineup. The speedbump isn't bad at all +200-300Mhz here and there.

To call it Bulldozer is simply ignorance.

You're entitled to your opinion.

IMO; Those miniscule speed increases, while offering no IPC increases, makes this a pointless release.

Hopefully the first 10nm parts will provide at least a 10% IPC gain while significantly lower power consumption...then I'll be impressed.
 
Hopefully the first 10nm parts will provide at least a 10% IPC gain while significantly lower power consumption...then I'll be impressed.

If that happens I expect to see a clock speed reduction.
 
If you want something faster than Kaby Lake its called Skylake-X.

Yeah, there was also Kaby Lake-X on some news, both for socket 2066, I believe, but I guess it is still quite bit of rumors, we know such is coming, but what those will deliver remains to be seen.


Also kind of agree that Zen probably is not going to be up to hype, but few fanboy reviews and it will sell and after that Intel will sell more again as those who made the move will move back.

Back in time I made one such upgrade, believing what magazine told and actually lost performance on that upgrade, well at least I helped to keep development of CPU's alive with my money.
 
Yeah, there was also Kaby Lake-X on some news, both for socket 2066, I believe, but I guess it is still quite bit of rumors, we know such is coming, but what those will deliver remains to be seen.

Kaby Lake -X is a 112W quad, comes from the Kaby Lake-S dies. So dual channel and 16 lanes.

Skylake-X is a bit changed compared to Skylake-S, 1MB L2 cache, 1.375MB L3 cache per core. AVX512 perhaps enabled on Skylake-X as well. 6, 8 and 10 cores.

You can find it all in Intel information. In terms of release its confirmed for Q3 in the last financial conference call.
basin-falls.jpg

kaby-lake-x.jpg
 
You're entitled to your opinion.

IMO; Those miniscule speed increases, while offering no IPC increases, makes this a pointless release.

Hopefully the first 10nm parts will provide at least a 10% IPC gain while significantly lower power consumption...then I'll be impressed.

For you, yes. But so have it been forever haven't it?

First(Cannon Lake) 10nm is unlikely to give 10%. Second generation perhaps yes (Icelake).

IPC history:
Farber123.jpg
 
For you, yes. But so have it been forever haven't it?

First(Cannon Lake) 10nm is unlikely to give 10%. Second generation perhaps yes (Icelake).

IPC history:
Farber123.jpg

What are you talking about? It hasn't been that way forever...there are IPC gains between most generations plainly shown in that graph you posted.

Kaby Lake = Westmere. Didn't give shit in terms of IPC gains over its predecessor.

Westemere was simply a die-shrink of Nehalem.

Kaby Lake isn't even a die-shrink.
It isn't even a socket change.
And it doesn't provide any IPC improvement over its predecessor.
To me, it's a completely pointless release because of those major factors.

If you feel otherwise, then more power to you and let us all know how your Kaby Lake performs.
 
How is this anything other than maybe a dot release/minor revision of Skylake? Wow.
 
Kyle is there going to be anything in this new chip that will boost the 270 boards when they are released at the same time, maybe more lanes or anything we just can't use in 170?
Quad channel memory on the 270?
Heck if the only thing they have is new to be new then this is bad, very bad.

I wish AMD's new processor sets them on their ass so we can get some competition back in the game and maybe get new because its a lot better.
 
This i had my 3930k rig for 4 or 5 years and only upgraded when the mb started to die. Its sad how litle cpus have come in the last few years higher ipc but lower clook room and less power thats about it.
At least gpus still need upgrading

Yep. I just had to leave my 3770k behind when my motherboard died. I was perfectly happy with the processor. The new 6700k really isn't better enough that it makes much of a difference. The m.2 SSD though... That's a difference maker.
 
Zen with its lower IPC (remember this is supposed to be between Ivy Bridge and Haswell if AMD was telling the truth about the 40%) and lower clocks for the 8C / 16T versus Intel's mainstream will not make it the CPU that many people are dreaming about.

I think while Zen may not have a faster IPC I do believe AMD will offer more cores and threads than Intel at a cheaper price and multi threaded applications may thrive. Zen may be a great choice for video editing and photochop at hundreds less than Intel. Even if Zen comes close to Intel for gaming they will win back all of the AMD fans. I know I have held out for AMD for many years.

We'll find out soon.
 
I think Kaby Lake is a win, not the desktop but just the chip. The combination of Kaby lakes great clock speed to power usage for mobile, and nvidia's current power usage for pascal, I think those are laptops and devices that the Expendable income that Kyle mentioned in his review will be spent. I'm happy with my 5960x and 5930k on the desktop, it's just to damn annoying upgrading chipsets. Unlike some people though I dont look at Intel fucking me over every new gen because they released a product I wasn't gonna buy anyways. I see them as holding immense value over time that has paid over and over and over in cost. I've given my family 2500ks, 3570ks, and all that and those chips are STILL amazing. Literally intel has created generations of very reliable and nicely performing parts that go over a half decade back and are still relevant today. As an asian guy who's owned camries passed down from my parents, and would love to give a camry to my daughter when she turns 16...this is the f'ing dream right here boys :).

Concerning this chip. I'm actually interested in a 7700k because I found personally more cores affects what I do pretty much negligently. I have an 8 core system and I do basically shit with it. Yeah I have photo editing and stuff like that...but no one programs the stuff I do to use a system I have well. I think it's cool though I could grab a 7700k and mess with the idea of 5ghz...that's nice. So selling the 5960x and system, upping the 5930k back to the photo editing rig would make it nice to get a 7700k to concentrate on as a gaming monster. I remember getting 5ghz clocks on the 4790k and loving the experience running 5ghz with a itx build and getting killer framerates.
 
Last edited:
Really disappointing. I need to build another system for upstairs but at this point, it's a dead end build with either Skylake or maybe even Ivy Bridge. I hate it when the build cycle is crap like this. There is nothing worth upgrading until Q4 17/ Q1 2018 so it's a one build system.
 
I think while Zen may not have a faster IPC I do believe AMD will offer more cores and threads than Intel at a cheaper price and multi threaded applications may thrive.

I have no doubt the 8C / 16T will have an advantage in 8+ (maybe even 6+) threaded applications versus Intel's 4C / 8T.

Even if Zen comes close to Intel for gaming they will win back all of the AMD fans.

There are a lot of these that want $100 CPUs. The 8C / 16T was designed to be a $300+ CPU. Although there will be lower core count Zen processors in that price range.
 
Last edited:
Thanks for the review. Looks like Intel just released their Bulldozer.

This is hardly the case. Intel's "Bulldozer" was the first "Netburtst Microarchitecture" based processors which would be the basis for the entire Pentium IV family. It started with the Pentium IV Willamette processors back in the and ended with the Cedar Mill based processors.

I suppose perspective is subjective. Care to elaborate why you feel it's a success, despite it being a carbon-copy of Skylake?

First and foremost, Kaby Lake isn't a carbon copy of Skylake. The iGPU is different and it can maintain higher frequencies with less power. I'm not sure what he exact optimizations are but there are obviously some changes to the design which account for this. It's showing better than expected performance gains in the mobile applications. I think Kaby Lake is going to be a success in that market.

Zen with its lower IPC (remember this is supposed to be between Ivy Bridge and Haswell if AMD was telling the truth about the 40%) and lower clocks for the 8C / 16T versus Intel's mainstream will not make it the CPU that many people are dreaming about.

Exactly.

Loved the article! Intel really are doing everything they can to allow AMD to catch up so it's not a monopoly. What i think would make an amazing article would be to do a comprehensive round up of IPC 'gains' of all Intel quad cores dating back to Core 2 Extreme QX6700. Simply choose a clock all CPUs can reach and see how the benchmarks fall. There will be a little help from RAM development but it would make for such an interesting read don't you think?

I'm fairly certain that Intel isn't remotely concerned about what AMD does or does not do. It isn't letting AMD catch up. People don't seem to understand that developing for "performance per watt" is much more difficult than building CPUs for raw performance. In the two markets that demand this above raw performance, these products have been and will continue to be successful. If Intel wanted to release a pure performance CPU for the desktop, I'm sure they would be able to do so. The market isn't there so it isn't going to bother.
 
I'd honestly like to see a completely NEW CPU, utilizing a better "infrastucture" than the x86. There are so many transistors "wasted" to get around the limitations of the x86 architecture it isn't even funny.

If a company sat down and actually devised a new instruction set, with the idea of muli-core/multi-thread execution in mind, especially with an eye towards massive multitasking. None of these things were on the minds of CPU developers when the x86 architecture was designed. And companies have done amazing work with adding instructions, register renaming, task switching, but it's patch upon patch upon patch. Which in no way can be as efficient as designing something from the ground up to have these abilities.

Honestly, i am thinking with a really new design, the machine would be fast enough to "decode on the fly" old x86 code, or even just run a VM with an x86 emulator setup.

I mean we are still stuck with clocks, when the articles i read are all about clockless designs being the future.

I know what i am talking about has been attempted, but it wasn't by a company with the brain power and deep pockets of intel.
 
I'd honestly like to see a completely NEW CPU, utilizing a better "infrastucture" than the x86. There are so many transistors "wasted" to get around the limitations of the x86 architecture it isn't even funny.

If a company sat down and actually devised a new instruction set, with the idea of muli-core/multi-thread execution in mind, especially with an eye towards massive multitasking. None of these things were on the minds of CPU developers when the x86 architecture was designed. And companies have done amazing work with adding instructions, register renaming, task switching, but it's patch upon patch upon patch. Which in no way can be as efficient as designing something from the ground up to have these abilities.

Honestly, i am thinking with a really new design, the machine would be fast enough to "decode on the fly" old x86 code, or even just run a VM with an x86 emulator setup.

I mean we are still stuck with clocks, when the articles i read are all about clockless designs being the future.

I know what i am talking about has been attempted, but it wasn't by a company with the brain power and deep pockets of intel.

It was called IA64 and died with x64 unfortunately.

I dont think we are going to see anything new as long as ARM and x86 exist and when there is more than one company making them. The risk vs reward from a business perspective is simply not there and learned the hard way with IA64. And its a shame when looking far ahead.

Competition rarely means quality.

And these issues just gets bigger and bigger as we see more cores with very limited scaling and less benefit from wider cores.
 
Last edited:
they did make it 24 pci lanes up from 20. I can't find anything they did on the cpu making a hill of beans.

We can only hope that amd has something and makes intel get off their ass and give us something for our money
 
I'd honestly like to see a completely NEW CPU, utilizing a better "infrastucture" than the x86. There are so many transistors "wasted" to get around the limitations of the x86 architecture it isn't even funny.

If a company sat down and actually devised a new instruction set, with the idea of muli-core/multi-thread execution in mind, especially with an eye towards massive multitasking. None of these things were on the minds of CPU developers when the x86 architecture was designed. And companies have done amazing work with adding instructions, register renaming, task switching, but it's patch upon patch upon patch. Which in no way can be as efficient as designing something from the ground up to have these abilities.

Honestly, i am thinking with a really new design, the machine would be fast enough to "decode on the fly" old x86 code, or even just run a VM with an x86 emulator setup.

I mean we are still stuck with clocks, when the articles i read are all about clockless designs being the future.

I know what i am talking about has been attempted, but it wasn't by a company with the brain power and deep pockets of intel.

You mean like ARM?
 
I'd honestly like to see a completely NEW CPU, utilizing a better "infrastucture" than the x86. There are so many transistors "wasted" to get around the limitations of the x86 architecture it isn't even funny.

If a company sat down and actually devised a new instruction set, with the idea of muli-core/multi-thread execution in mind, especially with an eye towards massive multitasking. None of these things were on the minds of CPU developers when the x86 architecture was designed. And companies have done amazing work with adding instructions, register renaming, task switching, but it's patch upon patch upon patch. Which in no way can be as efficient as designing something from the ground up to have these abilities.

Honestly, i am thinking with a really new design, the machine would be fast enough to "decode on the fly" old x86 code, or even just run a VM with an x86 emulator setup.

I mean we are still stuck with clocks, when the articles i read are all about clockless designs being the future.

I know what i am talking about has been attempted, but it wasn't by a company with the brain power and deep pockets of intel.
You might want to look into POWER by IBM. It's pretty epic and designed completely around parallelization. It supports up to 8 way SMT and is user configurable for different thread counts based on workload.
 
i just re read some of the review but is it me or this review is missing voltage used on either chip? sure comparing 4.5ghz on both seems fair but power consumption is more or so based on voltage used yet it mention nothing about voltage. it seems kabylake use less power at first glance but since no chips are the same, what if kabylake is at 1.2v and skylake is at 1.35v?
 
Back
Top