Intel's i7-7700K Kaby Lake Overclocked and Reviewed at Tom's!

[Drak3]

3 hours ago, MageTank said:

This is gonna be fun. How am I wrong now?

I'm curious about this as well.

[Drak3]

2 minutes ago, MageTank said:

I like how people seem to think that you can just fix binning by throwing a decent loop at it. "Any CPU can hit 5ghz if you cool it hard enough". Sure, in some cases, that might be true, but then it begs the question, is it 24/7 stable while doing so? Let's face it. Overclocking is very black and white. You are stable until you are unstable, that is the most simplistic truth of it. What might be stable today, may be unstable tomorrow. I know people with different overclock profiles for the different seasons year round, lol.

 

If proper cooling was all it took to make 5ghz stable on the vast majority of CPU's, more people would be doing so. We don't see that because the lottery is real. No loop will make a chip that is incapable of 5ghz, do 5ghz, at least not without the risk of severe degradation.

 

Almost reminds me of the people that claim to be running 24/7 5ghz without a delid. 

The full truth is that CPU stability isn't black and white. There is alot of grey area to, and only black. There is completely unstable, and then a range of stability that doesn't reach 100%.

Also, proper cooling is what it takes to push chips like the 6700K to 5GHz while being relatively stable. That means investing quite a bit of time and money into a custom water loop or phase change cooling. Not many people do it, which is why we don't see many people with 5GHz 6700Ks. With Skylake, hitting that 5GHz on the 6700K without a loop with a price tag that rivals the rest of the system is winning the lottery.

[Drak3]

24 minutes ago, Lays said:

of course it's not lol

 

On skylake it is, for stability at least, booting into windows for a screenshot is easy sure, but stability at 5 Ghz on Skylake is damn near impossible. 

 

The OCN skylake OC thread has 92 samples last I checked, here's how %'s work right now in terms of clock speeds on the CPU's that have been stability tested.

 

Without an actual link, I don't exactly trust that those numbers are taking into account cooling, power delivery, and testing methods.

92 samples out of thousands also doesn't lead me to believe that their results are overly reliable for a conclusion over an entire generation of processors.

[MageTank]

1 minute ago, Drak3 said:

The full truth is that CPU stability isn't black and white. There is alot of grey area to, and only black. There is completely unstable, and then a range of stability that doesn't reach 100%.

Also, proper cooling is what it takes to push chips like the 6700K to 5GHz while being relatively stable. That means investing quite a bit of time and money into a custom water loop or phase change cooling. Not many people do it, which is why we don't see many people with 5GHz 6700Ks. With Skylake, hitting that 5GHz on the 6700K without a loop with a price tag that rivals the rest of the system is winning the lottery.

Stable isn't a word that deserves a prefix or suffix. You either are, or you are not. Yes, specific workloads cause instability more often than others, but it does not change the fact that in the end, you were unstable. That is the problem with the current world of overclocking. We have a bunch of people running around with pseudostable overclocks and try to pass them off as if they are "24/7 stable". The stability is always black and white, it's the users definition of stability that seems to differ per person.

 

As for what it takes to get 5ghz, yes, proper cooling is a must, but it's not the only aspect. Binning is a huge part. I have yet to see a loop that magically makes a bad bin turn good. I also have friends with delidded 4790k's and 6700k's that, even on custom loops with very high end hardware, fail to hit 5ghz even with what I would consider extremely amazing temps. @Lays has been through the binning process, and can likely attest to this as well.

 

Another aspect of CPU stability that people often overlook is memory. Overclocking ram (or even loading highly clocked XMP's) add heat to the total CPU package. Your CPU overclock might be stable, but once you add memory into the mix, that stability will be compromised a large percentage of the time. People tend to blame the ram itself in this situation, but that is simply not the case most of the time. It's why I never take ram reviews on Newegg seriously anymore, people simply do not understand what they are talking about. 

[Drak3]

1 minute ago, MageTank said:

Stable isn't a word that deserves a prefix or suffix. You either are, or you are not. Yes, specific workloads cause instability more often than others, but it does not change the fact that in the end, you were unstable. That is the problem with the current world of overclocking. We have a bunch of people running around with pseudostable overclocks and try to pass them off as if they are "24/7 stable". The stability is always black and white, it's the users definition of stability that seems to differ per person.

Stability isn't black or white, never was, never will be. Stability is a relative term, with the only absolute being unstable. If a chip can fail one time out of 1,000, and then work normally for 1,000 more times, it's still not 24/7 stable.

[MageTank]

Just now, Drak3 said:

Stability isn't black or white, never was, never will be. Stability is a relative term, with the only absolute being unstable. If a chip can fail one time out of 1,000, and then work normally for 1,000 more times, it's still not 24/7 stable.

That is by definition, black and white, lol. You are stable until you are unstable. I agree entirely with that statement. 

 

I wish I could convey my disdain for the modern way of overclocking without sounding like an old man complaining about kids being on his lawn. My point is, if people fail a stress test, they would rather switch to an easier test, instead of compromising slightly on performance for a better shot at relative stability. This leads to a ton of misinformation down the road, like "This specific application is broken, it crashes my PC" when it's actually stressing parts of their CPU that they have yet to deem stable with their OC.

[Drak3]

1 minute ago, MageTank said:

That is by definition, black and white, lol. You are stable until you are unstable. I agree entirely with that statement. 

 

I wish I could convey my disdain for the modern way of overclocking without sounding like an old man complaining about kids being on his lawn. My point is, if people fail a stress test, they would rather switch to an easier test, instead of compromising slightly on performance for a better shot at relative stability. This leads to a ton of misinformation down the road, like "This specific application is broken, it crashes my PC" when it's actually stressing parts of their CPU that they have yet to deem stable with their OC.

No, that's showing that nothing is 100% stable. Intel and AMD know that, and they base the clock speeds of their chips on what will be relatively reliable for the intended use cases.

 

I agree, there are a shit ton of stupid people that don't know how to stress test a CPU, but they also don't know how much of an issue heat is. Many people think that any AiO, and some of them think that an air cooler, is good enough to achieve high OCs.

[Lays]

25 minutes ago, Drak3 said:

Without an actual link, I don't exactly trust that those numbers are taking into account cooling, power delivery, and testing methods.

92 samples out of thousands also doesn't lead me to believe that their results are overly reliable for a conclusion over an entire generation of processors.

http://www.overclock.net/t/1570313/skylake-overclocking-guide-with-statistics

 

As of right now, I believe it's the largest database of publicly available / publicly submitted OC results in a thread.  HWBOT.org is the only other area one could see clocks, but that's just for benchmarks, not for stability tests.

 

I have only seen one other user on this forum since 6700k has been released that has said he has a 24/7 5 Ghz 6700k, but claims he's on an AIO without a Delid, and says he only needs 1.39v and it "crashes sometimes in games" yet claims he ran prime95 for 8 hours & realbench for 16 or something like that.  I myself have never had a CPU out of all ~10 I've owned in the past few years that couldn't run a game at an OC I could run prime95 at.

[MageTank]

Just now, Drak3 said:

No, that's showing that nothing is 100% stable. Intel and AMD know that, and they base the clock speeds of their chips on what will be relatively reliable for the intended use cases.

 

I agree, there are a shit ton of stupid people that don't know how to stress test a CPU, but they also don't know how much of an issue heat is. Many people think that any AiO, and some of them think that an air cooler, is good enough to achieve high OCs.

I never claimed that something is 100% stable. My point is, you are stable until you are unstable. Black (being stable) and white (being unstable) is my point. There is no in between if we are using the word stable. You either are, or are not. I might be confusing what black and white mean in a adjectival sense, but the point I am trying to convey remains the same. 

 

That's not to say I condemn people for having overclock profiles that are stable when gaming, but fail during Linpack or GROMACS folding, I simply disagree with their advertisement of "100% stable". 

 

1 minute ago, Lays said:

http://www.overclock.net/t/1570313/skylake-overclocking-guide-with-statistics

 

As of right now, I believe it's the largest database of publicly available / publicly submitted OC results in a thread.  HWBOT.org is the only other area one could see clocks, but that's just for benchmarks, not for stability tests.

 

I have only seen one other user on this forum since 6700k has been released that has said he has a 24/7 5 Ghz 6700k, but claims he's on an AIO without a Delid, and says he only needs 1.39v and it "crashes sometimes in games" yet claims he ran prime95 for 8 hours & realbench for 16 or something like that.  I myself have never had a CPU out of all ~10 I've owned in the past few years that couldn't run a game at an OC I could run prime95 at.

"I was stable on Prime95" also means very little to me, especially when they use non-AVX2 versions. Stable in what regard? 48k? God knows without a delid and using an AiO, you won't pass small FFT's, lol. 512k-4096k? Congrats, your ram, IMC and IO lanes passed, but it tells me very little about vcore. 1344k-2688k for vcore, as you are already aware. 8 hours of a standard blend tells you nothing these days. 

[8uhbbhu8]

And here I am with my 4790K perfectly fine no issues or reason to upgrade for at least another year or so

[Simon771]

When will Intel learn to solder K chips already ...

If they can do it with X99 (2011) chips, why can't they use the same for unlocked i5 and i7 ffs

[Lays]

22 minutes ago, Simon771 said:

When will Intel learn to solder K chips already ...

If they can do it with X99 (2011) chips, why can't they use the same for unlocked i5 and i7 ffs

There's a huge writeup done by der8auer ( a professional overclocker and electrician-ish engineer that deals with the in's and out's of the business )

 

http://overclocking.guide/the-truth-about-cpu-soldering/

[patrickjp93]

27 minutes ago, Simon771 said:

When will Intel learn to solder K chips already ...

If they can do it with X99 (2011) chips, why can't they use the same for unlocked i5 and i7 ffs

AMD doesn't either below a certain die size. The problem is if the dies are small enough, the soldering process can actually melt portions and make them unusable.

[Lays]

1 minute ago, patrickjp93 said:

AMD doesn't either below a certain die size. The problem is if the dies are small enough, the soldering process can actually melt portions and make it unusable.

And apparently due to the tiny die, when the solder / cpu goes through thermal cycles, it can cause cracks and further damage.

 

People think it's just Intel being cheap, but there's so much more to it from the looks of it.

[Simon771]

9 minutes ago, Lays said:

There's a huge writeup done by der8auer ( a professional overclocker and electrician-ish engineer that deals with the in's and out's of the business )

 

http://overclocking.guide/the-truth-about-cpu-soldering/

 

7 minutes ago, patrickjp93 said:

AMD doesn't either below a certain die size. The problem is if the dies are small enough, the soldering process can actually melt portions and make them unusable.

 

6 minutes ago, Lays said:

And apparently due to the tiny die, when the solder / cpu goes through thermal cycles, it can cause cracks and further damage.

 

People think it's just Intel being cheap, but there's so much more to it from the looks of it.

 

Yeah I do know it's not as simple as it sounds.

But company like Intel ... they are more than capable of figuring out the way to solder a bit smaller dies also.

[Lays]

2 minutes ago, Simon771 said:

 

 

 

Yeah I do know it's not as simple as it sounds.

But company like Intel ... they are more than capable of figuring out the way to solder a bit smaller dies also.

u rite they should break the laws of physics and spend billions of dollars on it so we can have 10c lower temps, thus making their end sales numbers to us higher, (our $340 i7's would be more money, giving more people shit to bitch about)

 

thumbs up buddy

[patrickjp93]

1 minute ago, Simon771 said:

 

 

 

Yeah I do know it's not as simple as it sounds.

But company like Intel ... they are more than capable of figuring out the way to solder a bit smaller dies also.

No they aren't. Some things in engineering cannot be done. IBM never figured it out, Intel hasn't, GloFo hasn't, TSMC hasn't, so I don't know why you think it's reasonable to harp on Intel for it. Intel's at the head of the industry. If it could do so, it would.

[Simon771]

2 minutes ago, Lays said:

u rite they should break the laws of physics and spend billions of dollars on it so we can have 10c lower temps

 

thumbs up buddy

So how can they solder i7 6800/6850/6900/6950 CPUs?

[Lays]

Just now, Simon771 said:

So how can they solder i7 6800/6850/6900/6950 CPUs?

because the heatspreader and die are huge in comparison, haven't you seen the die on an x99 cpu? it's fucking huge lol

[Simon771]

2 minutes ago, Lays said:

because the heatspreader and die are huge in comparison, haven't you seen the die on an x99 cpu? it's fucking huge lol

Well I never had any x99 CPU so I can't say.

They could use same die size on other K chips also, so they can solder it right?

Or would that cost too much, and that's also reason why i7 6800k and better CPUs are so expencive?

 

Just trying to understand how that work.

[patrickjp93]

2 minutes ago, Simon771 said:

So how can they solder i7 6800/6850/6900/6950 CPUs?

Because the 10-core die is 240 mm sq, not 144.  When the die is that large heat dissipates much more easily, AND the greater size and number of anchor points to the package reduce the strain on the corners of the die when solder heats and cools (expansion and contraction can cause the smaller dies to break).

[patrickjp93]

1 minute ago, Simon771 said:

Well I never had any x99 CPU so I can't say.

They could use same die size on other K chips also, so they can solder it right?

Or would that cost too much, and that's also reason why i7 6800k and better CPUs are so expencive?

 

Just trying to understand how that work.

The smaller the die, the less wasted area on a wafer, and the higher the yields. Financially it would be stupid for Intel to make the HEDT processors mainstream starting from 6 cores. We haven't even remotely used up the power of our quad cores if consumer software was made correctly.

[Lays]

4 minutes ago, Simon771 said:

Well I never had any x99 CPU so I can't say.

They could use same die size on other K chips also, so they can solder it right?

Or would that cost too much, and that's also reason why i7 6800k and better CPUs are so expencive?

 

Just trying to understand how that work.

Here's a pic of x99 chip next to consumer stuff:

 

 

5960x according to google:

 

 

 

 

6700k:

 

 

 

The die is huge on x99 because of all the shit they put on the cpu, like all the pch lanes, all the extra cache, all the extra cores, etc.  Mainstream consumer CPU's don't get those extra things, and if they did, there wouldn't be much reason to have x99.

 

When the die is so small, there's more concentrated heating and cooling on that solder and die, the heat can't dissipate as fast, and the cooling down process will happen more rapidly, thus causing the thermal "shock" if you will, similar to how a glass can shatter if it's hot and it comes in contact with freezing water.

When the die is big, the stress isn't as much on it when it heats and cools, because there's more surface area and more contact.

 

[Simon771]

6 minutes ago, patrickjp93 said:

Because the 10-core die is 240 mm sq, not 144.  When the die is that large heat dissipates much more easily, AND the greater size and number of anchor points to the package reduce the strain on the corners of the die when solder heats and cools (expansion and contraction can cause the smaller dies to break).

 

4 minutes ago, patrickjp93 said:

The smaller the die, the less wasted area on a wafer, and the higher the yields. Financially it would be stupid for Intel to make the HEDT processors mainstream starting from 6 cores. We haven't even remotely used up the power of our quad cores if consumer software was made correctly.

 

3 minutes ago, Lays said:

The die is huge on x99 because of all the shit they put on the cpu, like all the pch lanes, all the extra cache, all the extra cores, etc.  Mainstream consumer CPU's don't get those extra things, and if they did, there wouldn't be much reason to have x99.

 

When the die is so small, there's more concentrated heating and cooling on that solder and die, the heat can't dissipate as fast, and the cooling down process will happen more rapidly, thus causing the thermal "shock" if you will, similar to how a glass can shatter if it's hot and it comes in contact with freezing water.

When the die is big, the stress isn't as much on it when it heats and cools, because there's more surface area and more contact.

 

 

 

Well i read that article from der8auer. I also remembered that he is selling some delidding tool for CPUs ... I knew I heard that name before.

Thanks for all the explanation ... I now understand that i7 6700k and other 4 core CPUs can't be soldered because CPU die is too small.

But what I noticed while reading that article ... liquid metal performs about the same as soldering. Is there any reason why Intel wouldn't use that? I know it's not going to work with sub zero temperatures, but that's the same with TIM they are using right now, isn't it?

I'm realy tempted to delid my i7 4790k and put liquid metal on CPU die, and normal TIM around the CPU die (to prevent liquid metal from touching other small thingies around there). But I just don't want to crack my 350$ worth CPU just yet lol

Then again, 15°C lower temps would dbe great.

 

I promise I won't bash Intel about soldering i7 6700K anymore, but I will still complain why are they not using liquid metal, unless you can also convince me that there is some legit reason for that also

[patrickjp93]

4 minutes ago, Simon771 said:

 

 

 

 

Well i read that article from der8auer. I also remembered that he is selling some delidding tool for CPUs ... I knew I heard that name before.

Thanks for all the explanation ... I now understand that i7 6700k and other 4 core CPUs can't be soldered because CPU die is too small.

But what I noticed while reading that article ... liquid metal performs about the same as soldering. Is there any reason why Intel wouldn't use that? I know it's not going to work with sub zero temperatures, but that's the same with TIM they are using right now, isn't it?

I'm realy tempted to delid my i7 4790k and put liquid metal on CPU die, and normal TIM around the CPU die (to prevent liquid metal from touching other small thingies around there). But I just don't want to crack my 350$ worth CPU just yet lol

Then again, 15°C lower temps would dbe great.

 

I promise I won't bash Intel about soldering i7 6700K anymore, but I will still complain why are they not using liquid metal, unless you can also convince me that there is some legit reason for that also

Because Intel isn't allowed to. Liquid metal TIMs contain mercury or gallium which Intel legally is not allowed to use anymore.