Kaby Lake Overclocking Revealed: Simultaneously impressive AND disappointing

[MageTank]

So we have a few leaks from WCCF (Take this with a grain of salt if you must) showcasing a 7700K on LN2, and a 7600K on air. One of these results is pretty impressive. The other, not so much. 

 

Quote

The Intel Core i7-7700K was overclocked by HKEPC on LN2 cooling. The chip used might be an engineering sample but we have already seen retail chips ready. The chip was coupled on an ASRock Z170M OC Formula motherboard along with GALAX HOF DDR4 memory. The multiplier on the chip was adjusted to x67 and the bus speed remained untouched. This resulted in a clock speed jump of 6.7 GHz. There’s no reporting on the voltage used but it should be high considering the use of LN2.

6.7ghz on LN2. Keep in mind, we don't know the voltage required to achieve this, but the current WR on HWbot for the 6700k is 7ghz on LN2. This is slightly disappointing in comparison, but we know too little to be absolutely certain. If we knew its voltages, and had other bins to compare against, it would clarify whether or not this is a good or bad result, either way, I am currently not impressed with this LN2 result.

 

As for the 7600K, it's results on air, are pretty impressive. 

 

Quote

The Intel Core i5-7600K on the other hand was overclocked to 5.1 GHz on air. This test was performed on an ASUS Maximus VIII Gene motherboard by facebookHK. The chip had multiplier adjusted to x40 and bus speed adjusted to 127.51 MHz. This pushed the clock speed to 5.1 GHz. The chip voltage was adjusted to 1.504 Volt which seems rather high. The tester says that the chip was very stable on 5.1 GHz and the temps were pretty good considering the voltage used.

Now, we know the voltage required to achieve this overclock. It's very high. HOWEVER, keep in mind this was done on air. Thermal runaway dictates that the hotter something gets, the more power it consumes, which again leads to more heat and more power consumed. With a better cooling solution, this voltage requirement might be reduced. The fact that it was "very stable" at this clock speed on air, might lend some credit to the new fin design. To put this into perspective; one of the more prolific overclockers on this forum, @Lays has a 5ghz 6700K that runs at 1.43v on a custom loop. 

 

While the overall clock speeds have not changed from what we can achieve on Skylake (as predicted by most people), the impact of the new fin design on overclocking is still not known. Perhaps the more mature 14nm process in Kaby Lake provides a higher voltage tolerance over Skylake, or at least offers more stability of lower clock speeds at lower voltages. I am sure @patrickjp93 will know more about this, so I might as well tag him while I am at it. I still stand by my previous statements, in that Kaby Lake will have zero IPC improvements over Skylake. As we get more details about Kaby's release, I'll be prepared to eat my words if I am wrong. 

 

7700K clock speed image:

 

 

 

7600K clock speed image: 

 

 

 

 

Source: http://wccftech.com/intel-core-i7-7700k-core-i5-7600k-overclock/

 

UPDATE:

 

5 minutes ago, NumLock21 said:

More pics!

 

http://itbbs.pconline.com.cn/diy/53223899.html

 

 

I can't tell if the memory kit they are using is trash, or if the IMC on this specific CPU is bad, but this is a very lackluster memory result. DDR4 4000 has a peak theoretical memory bandwidth of 64GB/s in dual channel. This result is only showing 83% write efficiency, and only 64% read/copy efficiency. Hopefully it's the kit itself. I would hate to see a step backwards in IMC quality. 

[Praesi]

I wanna see every day OCs. Dont care about LN2 results.

[vanished]

So for haswell the commonly accepted safe voltage was 1.35, and then for skylake it strangely went up... had that happened again here?  My gut would tell me that it should be the same as skylake but I haven't heard anything yet.

[Dabombinable]

1 minute ago, Ryan_Vickers said:

So for haswell the commonly accepted safe voltage was 1.35, and then for skylake it strangely went up... had that happened again here?  My gut would tell me that it should be the same as skylake but I haven't heard anything yet.

Its could be that the FIVR in Haswell degraded above 1.35V. Though its possibly just Intel's 22nm process, since Ivybridge had a safe maximum of 1.2V

[MageTank]

1 minute ago, Ryan_Vickers said:

So for haswell the commonly accepted safe voltage was 1.35, and then for skylake it strangely went up... had that happened again here?  My gut would tell me that it should be the same as skylake but I haven't heard anything yet.

I feel it shouldn't have changed, but with that new fin design, it may have. The biggest piece of the puzzle to me, is exactly which air cooler they used on that 5.1ghz i5 result, and what exactly "very stable" means? In my eyes, "very stable" means days of heavy AVX2 load, something you will NEVER see on a 5ghz chip. As for the cooler, I am expecting something giant, like a Phanteks TC14PE or Noctua D15. 

 

If this chip is hitting 5.1ghz without being delidded with CLU, on air, then Kaby Lake just might be impressive after all. 

[vanished]

1 minute ago, MageTank said:

I feel it shouldn't have changed, but with that new fin design, it may have. The biggest piece of the puzzle to me, is exactly which air cooler they used on that 5.1ghz i5 result, and what exactly "very stable" means? In my eyes, "very stable" means days of heavy AVX2 load, something you will NEVER see on a 5ghz chip. As for the cooler, I am expecting something giant, like a Phanteks TC14PE or Noctua D15. 

Yeah I'd like to know the cooler as well.

 

And to me, I'm not sure why people feel the need to put qualifiers on an absolute.  It's like saying "very unique", or "very round", or "very on".  It either is, or it is not.  If it's stable, that implies it ran for 24 hours plus doing an intense stress test.  If it's not stable, it's unstable, and to what degree doesn't necessarily matter (to me at least).

 

1 minute ago, MageTank said:

If this chip is hitting 5.1ghz without being delidded with CLU, on air, then Kaby Lake just might be impressive after all. 

Agreed

[Dabombinable]

2 minutes ago, MageTank said:

I feel it shouldn't have changed, but with that new fin design, it may have. The biggest piece of the puzzle to me, is exactly which air cooler they used on that 5.1ghz i5 result, and what exactly "very stable" means? In my eyes, "very stable" means days of heavy AVX2 load, something you will NEVER see on a 5ghz chip. As for the cooler, I am expecting something giant, like a Phanteks TC14PE or Noctua D15. 

 

If this chip is hitting 5.1ghz without being delidded with CLU, on air, then Kaby Lake just might be impressive after all. 

Days of AVX2 is far beyond the coolers most people have for their CPU. I've already had my Siedon 120V almost die from trying AVX2 loads with my 4790K at 4.8GHz

[MageTank]

1 minute ago, Ryan_Vickers said:

Yeah I'd like to know the cooler as well.

 

And to me, I'm not sure why people feel the need to put qualifiers on an absolute.  It's like saying "very unique", or "very round", or "very on".  It either is, or it is not.  If it's stable, that implies it ran for 24 hours plus doing an intense stress test.  If it's not stable, it's unstable, and to what degree doesn't necessarily matter (to me at least).

 

Well, My logic has always been "stable until unstable". What was stable before, can become unstable over time. I also demand my overclocks be stable under the most stressful situations. If I have to underclock to be stable when folding GROMACS, then my overclock isn't stable. I simply do not believe in "gaming stable". 

[MageTank]

2 minutes ago, Dabombinable said:

Days of AVX2 is far beyond the coolers most people have for their CPU. I've already had my Siedon 120V almost die from trying AVX2 loads with my 4790K at 4.8GHz

Extremely efficient math is hot, lol. It gets even hotter when you overclock your memory too. Going from 2133 to 3600 was an 8C jump in CPU package for me. I literally had to delid in order for my 6700k to run Prime95 on my 45nm ITX cooler. 76C under load at stock clocks AND undervolted to 1.1v. 

[Dabombinable]

1 minute ago, MageTank said:

Extremely efficient math is hot, lol. It gets even hotter when you overclock your memory too. Going from 2133 to 3600 was an 8C jump in CPU package for me. I literally had to delid in order for my 6700k to run Prime95 on my 45nm ITX cooler. 76C under load at stock clocks AND undervolted to 1.1v. 

I'm stuck with lowly DDR3 1333 as I bought it for my i5 4440, and high speed kits (even DDR3 1600) costs far too much in comparison. Although I do run the cache at 4.6GHz.

[Lays]

16 minutes ago, Ryan_Vickers said:

So for haswell the commonly accepted safe voltage was 1.35, and then for skylake it strangely went up... had that happened again here?  My gut would tell me that it should be the same as skylake but I haven't heard anything yet.

Fivr went away and vcore became cache voltage and core voltage, that's why voltage went up and heat went down. 

[BuckGup]

If AMD has taught us anything it's clock speeds don't matter as much as you think

[vanished]

9 minutes ago, SamStrecker said:

If AMD has taught us anything it's clock speeds don't matter as much as you think

Well, yes, to a point, but if you take any given chip and speed it up, it's gonna help

[patrickjp93]

2 hours ago, MageTank said:

I feel it shouldn't have changed, but with that new fin design, it may have. The biggest piece of the puzzle to me, is exactly which air cooler they used on that 5.1ghz i5 result, and what exactly "very stable" means? In my eyes, "very stable" means days of heavy AVX2 load, something you will NEVER see on a 5ghz chip. As for the cooler, I am expecting something giant, like a Phanteks TC14PE or Noctua D15. 

 

If this chip is hitting 5.1ghz without being delidded with CLU, on air, then Kaby Lake just might be impressive after all. 

6 hours under IBT and it'll be rock solid for anything Joe Consumer is doing.

[BuckGup]

16 minutes ago, Ryan_Vickers said:

Well, yes, to a point, but if you take any given chip and speed it up, it's gonna help

But then there are other issues with heat, power usage, and so on. 

[patrickjp93]

25 minutes ago, SamStrecker said:

If AMD has taught us anything it's clock speeds don't matter as much as you think

And if Intel has taught us anything, it's that Intel doesn't sacrifice IPC to get high clocks, at least not since Netburst.

[BuckGup]

1 minute ago, patrickjp93 said:

And if Intel has taught us anything, it's that Intel doesn't sacrifice IPC to get high clocks, at least not since Netburst.

So would higher clocks with lower IPC yield better performance in only certain areas or the opposite of what we have now?

[vanished]

8 minutes ago, SamStrecker said:

So would higher clocks with lower IPC yield better performance in only certain areas or the opposite of what we have now?

afaik lower IPC is just worse no matter what.  And if you have ramp up clocks to compensate, this becomes very tangible as the higher clocks will almost certainly require more voltage, which means more heat and more power consumption.

[Energycore]

Same IPC, higher clocks and lower TDP, this really looks like what can be expected of an Optimization step.

 

Of course, getting to brag about 5GHz on air is nice

[Energycore]

11 minutes ago, SamStrecker said:

So would higher clocks with lower IPC yield better performance in only certain areas or the opposite of what we have now?

In very simple terms, the increase in clocks has to be more than the decrease in IPC to yield better performance.

 

Veeeeery simple terms.

[patrickjp93]

17 minutes ago, SamStrecker said:

So would higher clocks with lower IPC yield better performance in only certain areas or the opposite of what we have now?

It would provide higher performance for scalar code where instructions already have a 1-cycle latency and would be very difficult to drop, but just about everything else would drop, especially if cache clocks can't keep up.

[MageTank]

22 minutes ago, patrickjp93 said:

6 hours under IBT and it'll be rock solid for anything Joe Consumer is doing.

That is very true, for now. However, with what you discussed in your blog, if AVX gets more use, that might change. Now, I don't question the validity of peoples claims when they say their system is stable, because as I said earlier in this thread: It's stable until it's unstable. This is just a simple fact. What I do tend to question, is peoples claims of temperatures under specific loads. "My 5ghz 1.5v CPU never goes above 70C". If a claim like this is made, then they've clearly not exhausted all options for different load types. Find me a system at that clock speed and voltage that can survive large memory Linpack and not reach thermal junction (or let out the blue smoke). 

 

I used to criticize XTU for not being stressful enough, but you pointed out that they switched to a more efficient Linpack build, and it is quite stressful for the average consumer. I would consider that a great middle-ground stress test for people to use to stress their systems. I just don't expect people to get away with running their insanely high overclocks under more unrealistic stress levels (48k FFT P95, Large Linpack, GROMACS, etc). 

 

On a side note: Do you think it's possible that Kaby Lake will have higher voltage tolerances than Skylake? 

[patrickjp93]

1 minute ago, MageTank said:

That is very true, for now. However, with what you discussed in your blog, if AVX gets more use, that might change. Now, I don't question the validity of peoples claims when they say their system is stable, because as I said earlier in this thread: It's stable until it's unstable. This is just a simple fact. What I do tend to question, is peoples claims of temperatures under specific loads. "My 5ghz 1.5v CPU never goes above 70C". If a claim like this is made, then they've clearly not exhausted all options for different load types. Find me a system at that clock speed and voltage that can survive large memory Linpack and not reach thermal junction (or let out the blue smoke). 

 

I used to criticize XTU for not being stressful enough, but you pointed out that they switched to a more efficient Linpack build, and it is quite stressful for the average consumer. I would consider that a great middle-ground stress test for people to use to stress their systems. I just don't expect people to get away with running their insanely high overclocks under more unrealistic stress levels (48k FFT P95, Large Linpack, GROMACS, etc). 

 

On a side note: Do you think it's possible that Kaby Lake will have higher voltage tolerances than Skylake? 

Actually, based on the taller fins and refined process, the variant tolerances are probably higher.

[Dabombinable]

2 hours ago, SamStrecker said:

If AMD has taught us anything it's clock speeds don't matter as much as you think

That pretty much only applies for their post 2011 CPU and Netburst. CMT scales like shit at higher clock speeds, and actually loses IPC.

[Prysin]

4 hours ago, MageTank said:

Well, My logic has always been "stable until unstable". What was stable before, can become unstable over time. I also demand my overclocks be stable under the most stressful situations. If I have to underclock to be stable when folding GROMACS, then my overclock isn't stable. I simply do not believe in "gaming stable". 

play the right game and you learn that "Gaming stable" is a bullshit concept

 

Tomb Raider games, Guildwars 2, Far Cry 4 will crash much faster then almost any other game ive ever played if your OC isnt stable. These games/game engines has almost zero tolerance for errors caused by instabilities.