Overclock.net banner

[Tom's Hardware] Overclocking Intel’s Core i7-7700K: Kaby Lake Hits The Desktop

24K views 186 replies 91 participants last post by  nycgtr 
#1 ·
Quote:
Intel's new Kaby Lake CPU delivers on the clock speeds the company promised, but the power consumption and thermal characteristics were disappointing based on a leaked sample of the new chip we received and tested weeks ahead of its official launch.

The greatest gift a review editor can receive is unexpected hardware, especially when that hardware is an anticipated CPU. What would you do with an early sample of Intel's next-gen Desktop CPU? Overclock, of course! So would I.




Source

UPDATED: Tom's Hardware did a re-test on a MSI Z170A Pro Gaming board, the power consumption is better, however the results are still about the same.

Second Test Run Source
 
See less See more
2
#7 ·


This is bad.

Both 6700k and 7700k are running at 1.3v and ambient temperature was 15c. So, 7700k was running at 91c at stock and 97c at 4.8GHz. Holy..
Perhaps it is because of testing it on the Z170 motherboard. Other than that, this is essentially 6700k with razor-thin ipc increase.

Hardware VP9 encode/decode is nice to have though.
 
#8 ·
I am fine with this CPU but in reality is a like 2700K or 4790K. It should not have not been called 7th generation. Also new MB and chipsets for what???
 
#10 ·
Quote:
Originally Posted by Artikbot View Post

Come on AMD, bring Zen please. We're gonna wilt here waiting for another revolutionary desktop release from Intel.
More than anything I want Intel to wake up. Since Sandy Bridge they have been in deep sleep. Every generation since we hope for that magical jump in performance to finally have a reason to upgrade but they refuse to do so.
 
#12 ·
So like I keep saying, an overclocked 6700k
the 6700k uses significantly more power the higher the clock 70w increase @ 4.8ghz :x
 
#13 ·
If I can hit 5GHz again I will be happy.
 
#15 ·
Silicon lottery already sells 5 ghz 6600k and occasionally 4.9 ghz 6700k.

I imagine Kaby Lake is going to only increase the average max overclock of skylake by 100mhz.

This chip is pointless and just an excuse for intel to claim they have something better and help sell "new" z270 motherboards.

I'm also willing to bet this "improved" 14nm process is all but a lie as well.
 
#16 ·
Quote:
Originally Posted by Lex Luger View Post

Silicon lottery already sells 5 ghz 6600k and occasionally 4.9 ghz 6700k.

I imagine Kaby Lake is going to only increase the average max overclock of skylake by 100mhz.

This chip is pointless and just an excuse for intel to claim they have something better and help sell "new" z270 motherboards.

I'm also willing to bet this "improved" 14nm process is all but a lie as well.
"new" 14nm will probably benefit ULV CPU and not desktops parts.
 
#17 ·
Skylake-X / Zen can't come soon enough.
 
#18 ·
Quote:
Originally Posted by ZealotKi11er View Post

"new" 14nm will probably benefit ULV CPU and not desktops parts.
Indeed. Heck, when was the last time any new Intel process node actually resulted in desktop clockspeed increases (not the meaningless stock figures)? We never budged forward even a little past 32nm/Sandy Bridge in that department (and it's been 6 years).

At desktop 4+ GHz clockspeeds, I recall that 14nm/Skylake didn't even improve efficiency much (if at all) over 22nm/Haswell.

It's all over ladies and gentlemen, we've apparently hit the brick wall of physics and/or Intel's ability to give a crap. Ya know why the next "architecture" is called Icelake? It's because it's a frozen, stagnant body of water and I'm only half-joking. The "lake" suffix should tell it all; it may very well be only an incredibly minor tweak of the Skylake microarchitecture (with a new chipset and PCI-E 4.0).
 
#19 ·
Quote:
Originally Posted by Serandur View Post

Indeed. Heck, when was the last time any new Intel process node actually resulted in desktop clockspeed increases (not the meaningless stock figures)? We never budged forward even a little past 32nm/Sandy Bridge in that department (and it's been 6 years).

At desktop 4+ GHz clockspeeds, I recall that 14nm/Skylake didn't even improve efficiency much (if at all) over 22nm/Haswell.

It's all over ladies and gentlemen, we've apparently hit the brick wall of physics and/or Intel's ability to give a crap. Ya know why the next "architecture" is called Icelake? It's because it's a frozen, stagnant body of water and I'm only half-joking. The "lake" suffix should tell it all; it may very well be only an incredibly minor tweak of the Skylake microarchitecture (with a new chipset and PCI-E 4.0).
Intel has been focusing their architecture on power consumption, not clockspeed.

And truthfully, hitting 5GHz is not easy. sure, the FX can do it, but that is on a much longer pipeline with a IPC in the core 2 duo's territory, nowhere neat sandy bridge. On a competent core design, you are not going to be able to just push clock rates willy nilly. I'd wager that the upcoming zen wont be able to do 5GHz easily either. Not to mention that power consumption still increases exponentially at those speeds. you need a high leakage chip to push super high speeds, and leakage and smaller transistors dont mix well. And the longer the pipeline is, the higher the clockspeeds you can achieve, but the poorer the IPC. Core 2 vs pentium 4, phenom II vs bulldozer, ece.

Instead of shouting "MOAR HERTZ!" we should be asking for better IPC. I'd rather have a chip that does 3.8 with 10% higher IPC then skylake then a chip that can do 6GHz but at core 2 duo IPC.
 
#20 ·
I personally think intel is just milking their virtual x86 monopoly more than we've hit the limits of of silicon.

In fact I would not be surprised if they are intentionally crippling overclocking as it hurts future sales.

I guess they are right because many people have stuck with sandy bridge which came out 6 years ago because there is little incentive to buy their newest CPU's.
 
#23 ·
Quote:
Originally Posted by GamerusMaximus View Post

Intel has been focusing their architecture on power consumption, not clockspeed.
I would stress that they've been achieving/focusing on power consumption, but only for mobile/low clockspeeds. To illustrate one of the points I made in my post (about that efficiency going right out the window at performance desktop clockspeeds despite Skylake's process node advantage over Haswell):
Power_02.png


Encoding_02.png

Not even efficiency is moving forward much for those of us interested in high-speed parts.

Quote:
And truthfully, hitting 5GHz is not easy. sure, the FX can do it, but that is on a much longer pipeline with a IPC in the core 2 duo's territory, nowhere neat sandy bridge. On a competent core design, you are not going to be able to just push clock rates willy nilly. I'd wager that the upcoming zen wont be able to do 5GHz easily either. Not to mention that power consumption still increases exponentially at those speeds. you need a high leakage chip to push super high speeds, and leakage and smaller transistors dont mix well. And the longer the pipeline is, the higher the clockspeeds you can achieve, but the poorer the IPC. Core 2 vs pentium 4, phenom II vs bulldozer, ece.

Instead of shouting "MOAR HERTZ!" we should be asking for better IPC. I'd rather have a chip that does 3.8 with 10% higher IPC then skylake then a chip that can do 6GHz but at core 2 duo IPC.
Therein lies the problem. IPC increases = increased pipeline complexity = either a loss of efficiency if not done selectively and/or a reduction in max clock speeds with no clear process node improvements to offset the losses. Hence, IPC increases that don't interfere with Intel's efficiency policy and don't reduce potential clockspeeds too much have understandably been hard-won, but the returns have been disappointingly slow and underwhelming nonetheless.

I'm all for asking Intel to increase performance however possible (and only focused on clock speed because that is the only thing Intel told us Kaby Lake would bring; we knew IPC would be identical to Skylake), but it's not like we're even seeing much movement on the IPC front either (if it's even possible to squeeze out more without offsetting the gains via lower clockspeeds). And Skylake actually had a few substantial improvements to its core design to extract greater instruction-level parallelism, but the resulting performance gains have been small anyway.

I wouldn't be surprised if Icelake barely has a 5% improvement in IPC over Skylake, and even then only situationally. Other than that, max clockspeeds are stagnant and more cores/cache seem like the last-ditch effort to try and gain some substantial performance gains. Maybe the adoption of some of the newer instruction sets (like AVX) would help too, but progress has been really slow on that front for consumer software.

Doesn't help that Intel still don't even enable AVX on the whole lineup. :/
 
#24 ·
Quote:
Originally Posted by darealist View Post

Looks like a poorly-mounted cooler. I refuse to believe it runs DAT HOT at stock.
No, cooler has nothing to do with it. Look at power consumption.

This is clearly a leaky cpu (not surprising, it is in fact an engineering sample), not representative of the final product. At least regarding thermals and power.
 
#25 ·
So they used the same board as me and they can only hit 3600mhz with 2 DIMM on while I think tweak town shows 3733mhz is also possible. I have the similar set up and I can only achieve 3600mhz RAM 2x8gb as well while 3733 and 3866 will refuse to boot no matter how much voltage applied to VCCIO and VSSA. I'm still a little bit disappointed with the 7700k at this moment for only being able to hit 4.8ghz @ 1.3v with 0 IPC increase. I was hoping majority of the sample can do 5ghz @ 1.35-1.4v so I can sell my 6700k @ 4.8 1.452v and replace it with a 5ghz 7700k.
mad.gif
 
This is an older thread, you may not receive a response, and could be reviving an old thread. Please consider creating a new thread.
Top