Overclocking Intel’s Core i7-7700K: Kaby Lake Hits The Desktop!

[TJ Hooker]

Doesn't seem right that an overclocked 6700K would be cooler (and henceforth use less power) than the stock 7700K...

Yeah, the huge temperature deltas between the 7700k and 6700k both at stock and overclocked seem pretty wonky, and don't seem to be in line with what one might expect based on the power draw measurements. Did Intel replace the TIM with toothepaste for Kaby Lake or something?

 

[turkey3_scratch]

I'd like it if the reviewer had used an oscilloscope for some power readings. TDP is almost never real world power.

 

[TJ Hooker]

I'd like it if the reviewer had used an oscilloscope for some power readings. TDP is almost never real world power.

He measured watts at the wall, presumably with something like a kill-a-watt meter. Now, I believe that Prime95 small FFTs are a fairly consistent, steady load, meaning that things should be in more or less a steady state, making average power draw a suitable alternative to using an oscilloscope to measure instantaneous power.

 

[nitrium]

I remember a time when Intel released the same CPU with higher clocks without attempting via marketing to make it look like a whole new generation. e.g. the i5 750 and the i5 760. What happened to that Intel? These should have been called the i7 6750K, i5 6650K etc. It's even the same socket ffs.

 

[xapoc]

4790k FTW

 

[RedJaron]

Doesn't seem right that an overclocked 6700K would be cooler (and henceforth use less power) than the stock 7700K...

This is completely unquantified. While both were overclocked 400 MHz, the 7700K was running a higher clock ( 4.8 GHz compared to 4.6 GHz ). It's not at all odd that a CPU at a higher clock will burn more power.

I'd like it if the reviewer had used an oscilloscope for some power readings. TDP is almost never real world power.

He measured watts at the wall, presumably with something like a kill-a-watt meter. Now, I believe that Prime95 small FFTs are a fairly consistent, steady load, meaning that things should be in more or less a steady state, making average power draw a suitable alternative to using an oscilloscope to measure instantaneous power.

It's not a perfect steady state. Total load fluctuates some depending on which FFT iteration you're on ( by default P95 cycles FFT size every 3-5 minutes ). My own tests have shown as much as 15W between one cycle and the next. But as long as you're comparing the same FFT size across the CPUs, you're fine.

4790k FTW

How do you figure that? Older platform, older RAM support, lower IPC. If you were building a new computer today and needed to buy every component, why would you opt for last-gen parts? Unless you're in a market where the price difference is prohibited, there's no reason.

 

[turkey3_scratch]

Doesn't seem right that an overclocked 6700K would be cooler (and henceforth use less power) than the stock 7700K...

This is completely unquantified. While both were overclocked 400 MHz, the 7700K was running a higher clock ( 4.8 GHz compared to 4.6 GHz ). It's not at all odd that a CPU at a higher clock will burn more power.

I'd like it if the reviewer had used an oscilloscope for some power readings. TDP is almost never real world power.

He measured watts at the wall, presumably with something like a kill-a-watt meter. Now, I believe that Prime95 small FFTs are a fairly consistent, steady load, meaning that things should be in more or less a steady state, making average power draw a suitable alternative to using an oscilloscope to measure instantaneous power.

It's not a perfect steady state. Total load fluctuates some depending on which FFT iteration you're on ( by default P95 cycles FFT size every 3-5 minutes ). My own tests have shown as much as 15W between one cycle and the next. But as long as you're comparing the same FFT size across the CPUs, you're fine.

No, in the Heat and Efficiency section, looking at those red bars on the graph, the 4.6Ghz I7-6700K is cooler than the stock I7-7700K.

 

[Ryrynz]

"Adds nothing new" Really? The new GPU? The new hardware decoder which allows for up to 2.6x the battery life? Improved clocks? Bug fixes..

 

[Ryrynz]

Also the improved speed shift will give you faster performance which wouldn't be shown well in these benchmarks.

 

[RedJaron]

No, in the Heat and Efficiency section, looking at those red bars on the graph, the 4.6Ghz I7-6700K is cooler than the stock I7-7700K.

Again, that's nothing surprising. Heat and voltage increase abruptly as you approach a CPU's OC limit due to the physical properties of silicon lithography. You hit the point where you need relatively big jumps in voltage to keep things stable for even small clock increases, which increases heat dramatically. I'd be interested to see this Kaby CPU compared against itself, one run at 4.8 GHz and the other at 4.7 GHz. More importantly, I'd like to see Kaby against Skylake at the same clock and then compare power and heat.

Also, as we don't know whether this chip was an ES or retail sample, and since we don't know if the 270 chipsets will include some additional power management in firmware, it's premature to draw any concrete conclusions. This was a preview. True, one that raises some concerns, but nothing to say the sky is falling.

 

[turkey3_scratch]

It's pretty surprising to me... Just to make sure, you are seeing it properly that the overclocked 4.6Ghz I7-6700K uses less power and is much cooler than the stock, non-overclocked I7-7700K, right?

 

[TJ Hooker]

No, in the Heat and Efficiency section, looking at those red bars on the graph, the 4.6Ghz I7-6700K is cooler than the stock I7-7700K.

Again, that's nothing surprising. Heat and voltage increase abruptly as you approach a CPU's OC limit due to the physical properties of silicon lithography. You hit the point where you need relatively big jumps in voltage to keep things stable for even small clock increases, which increases heat dramatically.

... So by that logic the Kaby Lake CPU is hitting its OC limits at a lower frequency (stock clocks no less), compared to the Skylake CPU. Yes, I would say that's surprising, given that the whole point of Kaby Lake is that it's supposed to be a more refined Skylake that can run better at higher clocks (ignoring iGPU stuff for the moment).

Yes, as you said, there are a variety of other factors at play here, not the least of which is simply the silicon lottery. No one is saying "well there you have it, Kaby Lake runs hotter and more power hungry than Skylake" (my toothepaste remark was obviously tongue-in-cheek). But the sample shown in this preview is still an odd and somewhat disconcerting result.

 

[alextheblue]

It's pretty surprising to me... Just to make sure, you are seeing it properly that the overclocked 4.6Ghz I7-6700K uses less power and is much cooler than the stock, non-overclocked I7-7700K, right?

Yeah, the 4.6 OC'd i5 is burning less juice than the stock 4.5 i7. Cost of hyperthreading?

 

[turkey3_scratch]

It's pretty surprising to me... Just to make sure, you are seeing it properly that the overclocked 4.6Ghz I7-6700K uses less power and is much cooler than the stock, non-overclocked I7-7700K, right?

Yeah, the 4.6 OC'd i5 is burning less juice than the stock 4.5 i7. Cost of hyperthreading?

This is an I7 I7 comparison not an i5 I7 comparison.

 

[Pere__]

Conclusion, wait another year, if you want more performance on your 6700k buy a good CPU cooler and overclockit

 

[Guest]

There is nothing interesting here.

 

[RedJaron]

Just to make sure, you are seeing it properly that the overclocked 4.6Ghz I7-6700K uses less power and is much cooler than the stock, non-overclocked I7-7700K, right?

Um, no, the OC'd Skylake is at 189W, the stock Kaby is 183W. After some manual intervention, the Kaby drop dramatically to 141W. This suggests an over-aggressive auto voltage in the firmware, which would be indicative of an unfinalized firmware, not a hardware problem.

Though yes, the stock Kaby does run warmer under load.

... So by that logic the Kaby Lake CPU is hitting its OC limits at a lower frequency (stock clocks no less), compared to the Skylake CPU. Yes, I would say that's surprising, given that the whole point of Kaby Lake is that it's supposed to be a more refined Skylake that can run better at higher clocks (ignoring iGPU stuff for the moment).

I said nothing of the kind. I said the Kaby is being pushed harder than the Skylake, both at stock and overclock, so it's not too surprising that it runs hotter, especially with what looks like an over-aggressive voltage profile. We saw similar things between Haswell and Devil's Canyon. Compare the 4770K to the 4790K. Intel marked the DC chip with a 4W higher TDP because the stock clock was now 4.0 GHz compared to the 4770K's 3.9 GHz Turbo Boost clock. The subtle refinements simply allowed a better power management that enabled them to set more chips to a stock clock on auto control that normally would've taken manual intervention to reach.

Also, notice the heat increase between the two. Kaby only went up 6° while Skylake went up almost 20°. Again, this in in line with a poor stock voltage control.

Yes, as you said, there are a variety of other factors at play here, not the least of which is simply the silicon lottery. No one is saying "well there you have it, Kaby Lake runs hotter and more power hungry than Skylake" (my toothepaste remark was obviously tongue-in-cheek). But the sample shown in this preview is still an odd and somewhat disconcerting result.

And I fully agree with that, so I'm not sure what you're getting at. As I said, "it raises some concerns."

 

[L33der of PCMR]

Soooo The only good think about Kaby Lake is it will reduce Skylake prices so i can move off my i3 6100?

 

[ultranoobcannon]

As a Sandy Bridge 2500k owner who is looking to upgrade in 2017, I have to say that this Kaby Lake preview was fairly underwhelming and discouraging. While I did not expect a huge leap over Sky Lake, I was hoping that Kaby would at least have greater over clock potential than its predecessor.

Perhaps this was a bad chip or a poor mother board that ran the benchmarks; however, I am not very excited about Kaby Lake right now.

 

[cloudonsky]

Please run HEVC 10 bits benchmark. Kaby Lake should do it in hardware.

 

[cloudonsky]

Please run HEVC 10 bits benchmark. Kaby Lake should do it in hardware.

 

[ledhead11]

As more than a few have noticed with Nvidia: Watch out for those updates! Some really help w/ known issues while others simply make the newer ideas more shoe-horned into mass consumption. It's not always a bad thing but it does make the 'older' tech less functional. I give us 2 or less(sorry post edit: years) until suddenly our 3 or 4th gen CPU's seem to have drastically less performance gaps compared to these very same kaby's.

Slightly off topic but not by much, about a month ago a leak showed a 6ghz OC spec. I'd be interested in something credibly to go with!

 

[anbello262]

I am actually quite amazed by this launch. It's everything that I expected it to be!

Of course it makes no sense to upgrade from skylake. That was completely obvious, and has always been. Just like upgrading from haswell to devils canyon, or Ivy Bridge to Haswell. It has always been around 7.5% improvement. And the fact that they can keep it up is surprising on itself, in my opinion.

But it is great to see that the good OC headroom we had in Skylake was improved even further, and that the chips can now perform to their full potential. Now we (might) have 4.8 Skylake chips. Who wouldn't want that?

For anyone building a new system, this is great news!


Also, sorry if I sound agressive. Not my intention. I'm just trying to be enfatic, and trying to point out what I truly believe has always been known or expected.

 

[dimonakid]

Thats what u get when FINFETing.
Compress the floww and u'll get these temps and voltage.
SO......... STILL NO 5GZ......

 

[sarinaide]

I am actually quite amazed by this launch. It's everything that I expected it to be!

Of course it makes no sense to upgrade from skylake. That was completely obvious, and has always been. Just like upgrading from haswell to devils canyon, or Ivy Bridge to Haswell. It has always been around 7.5% improvement. And the fact that they can keep it up is surprising on itself, in my opinion.

But it is great to see that the good OC headroom we had in Skylake was improved even further, and that the chips can now perform to their full potential. Now we (might) have 4.8 Skylake chips. Who wouldn't want that?

For anyone building a new system, this is great news!


Also, sorry if I sound agressive. Not my intention. I'm just trying to be enfatic, and trying to point out what I truly believe has always been known or expected.

80% of the desktop market doesn't need overclocking parts as 80% is OEM based systems for companies etc, add in the added costs of having a safe PSU, higher end board to OC and third party cooling makes OC as a target market a potential lose scenario.

Out the box this part runs hotter and uses more power, that already is bad in a market that doesn't require that.

The ball is in AMD's court now, if Zen delivers big it may be a hammer and chisel to Intel, and with their arches already in development for Cannonlake, they won't be able to just remedy that. The annoying thing is that Zen is more secretly guarded than top secret state dosiers, and AMD have been very strict or gestapo like on anyone that dares to leak anything. I think there will be a major shake up though for eth ebetter.