Retail 7700K Not Up to 5GHz - 3600MHz

HoffY said:
anyone able to do a few sketches up with dimensions for the delider? Shipping costs are a bit rough to down under! So i'll make one! :)
Click to expand...

Google it. Ppl have posted numerous design so I bet you will find something.
 
Is there a reason you don't use Prime95 anymore? I know I'm not even able to run my i7 7700k at 4.5 with HT without it having a a thread die :( I managed to play games at 5.0 ghz for 10 hours though without any craziness..

I've been out of the OC game since Ivy Bridge, and while my new system is way faster, I really feel let down if P95 is still a good test

****Just for clarification these are rounding errors, my system isn't crashing. Could it be the Ram?
 
Last edited:
Bahanime said:
So basically a more realistic OC for KL is ~4.8 to 4.9Ghz...

And when you compare that to Skylake typical OC, we're gaining like 100mhz extra.

Or in the grand scheme of multiple Ghz clocks, that's like a 2% clock speed advantage.

Oh boy. How exciting.
Click to expand...

But, but ... how else are you going to 4K Netflix and chill?
 
Xero717 said:
Is there a reason you don't use Prime95 anymore? I know I'm not even able to run my i7 7700k at 4.5 with HT without it having a a thread die :( I managed to play games at 5.0 ghz for 10 hours though without any craziness..

I've been out of the OC game since Ivy Bridge, and while my new system is way faster, I really feel let down if P95 is still a good test

****Just for clarification these are rounding errors, my system isn't crashing. Could it be the Ram?
Click to expand...
Basically is was burning up our CPUs. But keep in mind, across systems, those CPUs would run literally under stress tests for months at a time cumulatively. The AVX instruction set has become so specific, it was time to let it go. HandBrake is my goto first app now days to see if a system is even somewhat stable. And you can usually find out in about 20 minutes or not.
 
Meeho said:
HandBrake specifically or RealBench? What stability test procedure (app(s) and duration) would you find as THE test for 99.99% stability confirmation. I've had SB chips BSOD after >16 hours of Prime that could run RealBench without problems and I have a hard time considering them stable at that point.
Click to expand...
We use both. I have been running RealBench for 24 hours after HandBrake tests. I would suggest that Prime95 is no longer a client to tell us "real-world" stability. If it is for you, I am good with that.
 
Meeho said:
It's more of a "if it can run it stock it must be able to run it OCd" thing.
Click to expand...
Exactly. We kind of ran into the same thing on GPUs years ago, around the 480 timeframe. Yeah, testing programs can be built to break these chips. But what is the point of that once it gets to such a level? I try to think of much of what we do around here in a "real world" focus in terms of stress.
 
Thanks, I think I'll try both of those tonight before throwing in the towel. I did notice temps were insanely hot during prime 95 even with an h115i and max 1.38v (core 1 throttled..). Couple times I even canned the test myself.
 
Sharps97 said:
I've found the best program for beating the snot out of your CPU is to run some folding. That tends to trip your system pretty quickly given the extreme stresses it puts on the entire system.
Click to expand...

Yeah folding is pretty up their among the top contenders.
folding at home (CPU only) 65-67
Prime95 small fft 67-71
Prime95 Blend 67-72
Intel Linpack 71-74


But their is another problem with just running a CPU load on a potential gaming machines. it the heat from the GPU and the effect from the power stress is not taking into effect when just running a CPU stress test.
Asus realbench at least does some kind of low level GPU stress testing (not near the levels of a game though)

Also I'm unsure realbench actually verifies calculations which is another issues . you can have a stable system that churn sout miscalculations and slowly corrupt the data you are working with. i dont believe folding at home verifies the results are correct Prime95 and Linpack does.
I've had machines run games fine but could not compress correctly. and the archive would have CRC's errors all over it. but some people don't see the difference between working correctly and not crashing.
And I've seen handbrake complete through an unstable computer. and the results was just some weird delta frames behavior so 2 scenes had some weird garbage on it.. again not crashing is not the same as working correctly


Around 3 years ago i did a test of CPU stress testing software across AMD and Intel different generations. and prime95 OCCT and IBT ( no avx) came out ahead with a big distance down to other software to find instability.
IBT would win on CPU's without AVX. and prime95 blend would take the crow on machines with AVX
sadly it was only after the test i realized IBT was using a linpack without AVX support :(

I clearly remember SuperPI running 17+hours on a system that would crash in less than 5 mints with prime/IBT

of top of my head the test candidates list:

Aida64
Passmark burnin
SuperPI (multiple instances)
CPU-burn
Hot-cpu
BurnMMX
BurnK7
BurnP6
OCCT own
Prime95 blend
Priime95 small
+ some other it was 21 or 21 in total

if they ran stable for 3hours they would be declared as "did not find instability)
and i would lower the voltage/increase the oc a bit and run them all until all of them would crash withing the giving 3 hours frame and thereby easily see across all the runs which ones detect errors the first.
took some day though to run through all the trials on all the system though :cry:

Linpack & Prime95 + furmark for me every time
But thats because my system often is runninh at 100% load levels for weeks in a row.
 
  • Like
Reactions: Meeho
like this
I'm just going by what it reports:
kDMT8Sh.jpg


I was a firm believer in >24h Prime stability, but I do see the validity of more real world tests after Haswell and/or new Prime versions.
 
Last edited:
So much for the vaunted process improvements.

Disappointing compared to :
  • Q6600 B3 -> G0
  • i7 920 C0 -> D0
  • 4770K -> 4790K
 
Again, I say I'm not impressed with Kaby Lake. I got 4.8 on my 4790k at 1.3v, 2 and a half years ago, and it was still running at that until I lost my motherboard on 12/26. Oh, and that was without delidding the CPU.
 
There is an assumption being made that if someone and or a group of people invest in and maintain a somewhat reputable PC hardware review site that they automatically know how to overclock properly. I know this is not true.

I love Hardocp, I support and defend this site more than any other. Hell, I'm even a Texan at heart just like Kyle and Steve. While I live in Kansas City, I am back in Dallas almost monthly to visit friends and family. I have family all over Arlington, Garland, Mesquite, Irving, Denton, Dallas, Ft. Worth etc. So I'm definitely a fan, I think we are even close in age. 48 here and I know just as many aggie jokes as those two I can promise you that.

Not to step on any toes but it's very possible the people not hitting 5.0ghz just don't know how to. And I also disagree, 4.7ghz is still awfully close to 5ghz. 4.7 is close to 5 anything period.

I just dropped $2,276 on a new Kaby i5 dual gtx 1080 system and hit 5.0ghz using the Asus Hero 9. Lean mean gaming machine. I may keep it I may not. I do believe that with Asus that 80% of people out there will get 5.0ghz. I know this is true because using their auto overclocking 3way software I can get 4.7ghz stable but can't get that on my own. I'm ether retarded or missing something. Seems Asus does have some secret sauce.

BTW, for the audiophiles on hardocp. The Asus Hero 9 has a built in DAC and it sounds pretty damn amazing if you have top-shelf speakers.
 
I'm still on pins and needles for the results after a delid and coat of liquid metal. I know thermals aren't really the issue with the 5ghz ceiling, but it'll be interesting watching you guys fry a couple of the bastards finding out what voltage they WILL work at.
 
4.8Ghz is pretty darn close to 5Ghz, IMO. I know hitting that magic # makes everyone feel all warm & fuzzy, but the little few percent margin from 4.8Ghz to 5Ghz makes very little real world difference. And since "we" (people around here) at Hard OCP always focus on real world performance, overclocks, etc... not hypothetical performance benchmarks.

Example:
4.75 = 95%
4.8 = 96%
4.9 = 98% of 5Ghz

So maybe he's looking at it as how many will get 95% of 5Ghz (4.75Ghz) or better.
 
SixFootDuo said:
There is an assumption being made that if someone and or a group of people invest in and maintain a somewhat reputable PC hardware review site that they automatically know how to overclock properly. I know this is not true.

I love Hardocp, I support and defend this site more than any other. Hell, I'm even a Texan at heart just like Kyle and Steve. While I live in Kansas City, I am back in Dallas almost monthly to visit friends and family. I have family all over Arlington, Garland, Mesquite, Irving, Denton, Dallas, Ft. Worth etc. So I'm definitely a fan, I think we are even close in age. 48 here and I know just as many aggie jokes as those two I can promise you that.

Not to step on any toes but it's very possible the people not hitting 5.0ghz just don't know how to. And I also disagree, 4.7ghz is still awfully close to 5ghz. 4.7 is close to 5 anything period.

I just dropped $2,276 on a new Kaby i5 dual gtx 1080 system and hit 5.0ghz using the Asus Hero 9. Lean mean gaming machine. I may keep it I may not. I do believe that with Asus that 80% of people out there will get 5.0ghz. I know this is true because using their auto overclocking 3way software I can get 4.7ghz stable but can't get that on my own. I'm ether retarded or missing something. Seems Asus does have some secret sauce.

BTW, for the audiophiles on hardocp. The Asus Hero 9 has a built in DAC and it sounds pretty damn amazing if you have top-shelf speakers.
Click to expand...

Yes, possible but I find it very improbable given the experience of [H] and every other publication mined above.

Not to mention the comment made by Francois seems to heavily imply that it shouldn't be difficult.
 
Last edited:
Pretty funny when the criteria of 'stable' varies and talks of programs burning up a CPU/Motherboard - either way I enjoy the [H]'s overclocking methods because they keep the OEMs honest even if the old girl PRime95 isn't used. I still go with Prime95 myself because I want to see what would happen in a almost true 100% loaded scenario (Since 100% CPU reported load can be different things depending on WHAT the load is) -- as this is what also happens when Windows takes a complete shit across all cores and pegs them all out to 100%. Another thing that one might want to take into consideration is the load on the integrated memory controller (IMC) - the more RAM, the more load, which means more heat/power drawn at the CPU socket that ends up above TDP. I have a system that will be stable at 1066MHz at 4GB, but when cranked up to 16GB the stable bus speed plummets to 570MHz under Prime95 - CPU TDP was affected also and the HSF air flow was modified to suite (TDP capacity increased by ~30%). Overclocking these days is anything but an art - it is almost a science (i.e. it's helpful when you know electrical, fluids, and heat transfer theory with an engineering background).

I am sure Kaby Lake can pull of 5.0Ghz, but under what conditions? sacrifices? and what limits? These constraints need to be found- and will these constraints be acceptable to the enthusiast community is another question. If AMD has done their homework for Ryzen especially with power consumption vs performance, and communicated proper design requirements to motherboard/power supply OEMs; Intel will be in for a serious ass whopping this quarter and beyond - unlike the AM2/AM3 days with burned out motherboard VRMs.
 
SixFootDuo said:
BTW, for the audiophiles on hardocp. The Asus Hero 9 has a built in DAC and it sounds pretty damn amazing if you have top-shelf speakers.
Click to expand...

Every consumer motherboard has a built in DAC. :)

C-rizzle said:
4.8Ghz is pretty darn close to 5Ghz, IMO. I know hitting that magic # makes everyone feel all warm & fuzzy, but the little few percent margin from 4.8Ghz to 5Ghz makes very little real world difference. And since "we" (people around here) at Hard OCP always focus on real world performance, overclocks, etc... not hypothetical performance benchmarks.

Example:
4.75 = 95%
4.8 = 96%
4.9 = 98% of 5Ghz

So maybe he's looking at it as how many will get 95% of 5Ghz (4.75Ghz) or better.
Click to expand...

We've been pretty close to 5GHz since Pentium IV. Please, no more pussyfooting around. We had enough of "great overclockers" and "5GHz easy" promises in both CPUs and GPUs the past years.
 
When a girl asks for 12", saying that 11.75" is only 2% off ain't gonna fly.
Just sayin' ;)
 
  • Like
Reactions: Meeho
like this
Insula Gilliganis said:
Long time buddy of [H] Gordon Mah Ung wrote this..



Given that statement and Franky Ped's 'tude, I would guess that at least half of those "review" 7700Ks should have hit 5.0GHz or higher. So let's see how "overclockable" those 7700K that Intel did give out to the hardware sites really were.. processors given to reviewers (a couple of sites did buy their own retail 7700Ks) who probably have more than a passing clue on how to overclock a CPU to it's maximum stable GHz!! This is what I found going through the first 3 or so pages of a Google search for "7700K review".. not a total or complete search but I wanted just enough of a sample to start drawing some vague conclusion about 7700K ability to hit 5.0.

Eurogamer.net - 4.8GHz (used both a Asus Z270 Maximus 9 Code & MSI Z270 Gaming M5)
Hexus.net - 4.85GHz (Asus Z170 Pro)
Trustedreviews.com - 4.9 (Asus Z170-Deluxe)
Eteknix.com - 5.1 (ASUS Maximus VII Ranger)
PCPer.com - 5.1 (ASUS Maximus IX Code)
Guru3d.com - 5.0 (ASUS Maximus IX Formula)
Anandtech.com - 4.9 (ASRock Z270 Extreme4 & MSI Z270 Gaming M7)
Kitguru.net - 4.8 (MSI Z270 Gaming Pro Carbon)
Hothardware.com - 4.7 (ASUS Maximus IX Hero)
Legitreviews.com - 5.1 (Gigabyte Aorus Z270X-Gaming 5)
Tomshardware.com - 4.8 (Gigabyte’s GA-Z170X-Ultra)
Gamespot.com - 4.8 (Gigabyte GA-170X-Gaming 7)
Techspot.com - 4.9 (Asrock Z270 Gaming K6)
Gadgets.ndtv.com - 4.8 (Asus Maximus IX Hero)
Techreport.com - 4.8 (Gigabyte Aorus Z270X-Gaming 5)
Tweaktown.com - 5.0 (four different Z270 motherboards that weren't specified - all hit 5.0)
Overclockers.com - 5.0 (MSI Z270 XPower Titanium Gaming)
Gamersnexus.net - 5.1 (Gigabyte Aorus Gaming 7)
PCgamesn.com - 4.9 (Asus TUF Z270 Mk1)
Computershopper.com - 4.8 (unspecified Z270)
Techpowerup.com - 4.9 (ASUS Maximus IX Code Intel Z270)
Overclockersclub.com - 4.95 (MSI Z270 SLI Plus)
Tomshardware.co.uk - 5.0 (MSI Gaming M7)
Bit-tech.net - 5.0 (MSI Z270 Gaming Pro Carbon)

Of the 24 listed reviews, 9 made it to 5.0 or above. If we add in Kyle's 1 in 4 making 5.0, that means 10 out of 28 attempts made it to 5.0 or above, or 36%. Gordon's "5GHz or got very close" statement seems true as all but 1 review got at least "very close" (I don't consider 4.7 "very close").

Of the "known" motherboard manufacturers listed above, MSI had 3 hit 5.0+, ASUS 3, and Gigabyte 2.. if done by percentage, MSI was 3/7 (43%) successful, ASUS was 3/10 (30% <-- NOT 80%), Gigabyte went 2/5 (40%) and Asrock was goose egg for 2.

Wonder if Franky Ped of Intel would still be bitching about drawing "conclusions" based on 28 attempts??
Click to expand...


Lets do some statistics based on the binomial distribution.

with a sample size of 28 and 10 successes, using the binomial distribution, this would predict that the actual success rate of all chips on the market, at a 95% confidence interval is between 18.6% and 55.9%.

So, we are 95% sure, that somewhere between 18.6% and 55.9% of all 7700K's will hit 5Ghz.

Given this data, an absolute best case predition is that just over half of all retail chips hit 5Ghz, and a worst case prediction is 18%

I can probably tighten that statistical range a little bit using a tolerance interval k-factor calculation based on variable data, rather than pass/fail data, but that will have to wait until tomorrow, as this is a pain in the butt to do manually, and I have Minitab to assist with calculations like these at work.
 
I've been looking for a reason to upgrade, and I ain't seeing it.

I have a 3930k that is completely stable on a 6x6mm heat pipe cooler at 4.7GHz; it will recover from sleep, everything.

I just bought a water rig to see if it will hit 5GHz, Stably.

I doubt it, but we'll see. :)
 
Meeho said:
Every consumer motherboard has a built in DAC. :)
We've been pretty close to 5GHz since Pentium IV. Please, no more pussyfooting around. We had enough of "great overclockers" and "5GHz easy" promises in both CPUs and GPUs the past years.
Click to expand...

Given how many process shrinks and generations it's been since the P4, I'm really starting to believe the 5ghz ceiling is perhaps a physical limitation of silicon more than anything else at this point. Sure, we can get close and beat it with some exceptional cooling and magical smoke trickery... but all these recent CPUs, like the P4, seem to have trouble STAYING at or above 5Ghz for a while. (Numerous reports of gradually failing overclock settings requiring people to back off to regain stability) I expect this to be a bigger issue as time goes on with the smaller processes.
 
Seems like early reviewers did get golden samples as usual. Sweclockers for example had 4 samples and 3 out of 4 did 5GHz 1.32v stable. :p

Still you seem to have gotten a bit bad luck, I'm watching a forum thread with large majority doing 5GHz stable from between 1.31 to 1.37v.
 
RPGWiZaRD said:
Seems like early reviewers did get golden samples as usual. Sweclockers for example had 4 samples and 3 out of 4 did 5GHz 1.32v stable. :p

Still you seem to have gotten a bit bad luck, I'm watching a forum thread with large majority doing 5GHz stable from between 1.31 to 1.37v.
Click to expand...

I don't think most are truly stable in that it could go days without bs. Only because people don't want to spend the time.

And everyone else should give Francois a shout out that his thermal solution sucks ass!
If a DIY'er can drop temps 20 degrees with a few minutes and 2cents worth of thermal paste, maybe Francois should go back to engineering school.
 
Zarathustra[H] said:
Lets do some statistics based on the binomial distribution.

with a sample size of 28 and 10 successes, using the binomial distribution, this would predict that the actual success rate of all chips on the market, at a 95% confidence interval is between 18.6% and 55.9%.

So, we are 95% sure, that somewhere between 18.6% and 55.9% of all 7700K's will hit 5Ghz.

Given this data, an absolute best case predition is that just over half of all retail chips hit 5Ghz, and a worst case prediction is 18%

I can probably tighten that statistical range a little bit using a tolerance interval k-factor calculation based on variable data, rather than pass/fail data, but that will have to wait until tomorrow, as this is a pain in the butt to do manually, and I have Minitab to assist with calculations like these at work.
Click to expand...


Well, to follow up on this, I just punched the numbers into Minitab here using the table of data provided by Insula Gilliganis, and added in Kyle's 4 test results (5.0, 4.9, 4.8, 4.9)

First, some descriptive statistics:
Code: 
Descriptive Statistics

 N  N*     Mean     StDev  Median  Minimum  Maximum  Skewness   Kurtosis
28   0  4.91429  0.112099     4.9      4.7      5.1  0.208896  -0.852299

Now the bummer. This data is not normal at all, which makes these calculations trickier:

upload_2017-1-20_12-2-40.png


And it doesn't match any other distributions or transform into anything that can be analyzed either...

Code: 
Goodness of Fit Test

Distribution                 AD       P  LRT P
Normal                    0.943   0.015
Box-Cox Transformation    0.928   0.016
Lognormal                 0.932   0.015
3-Parameter Lognormal     0.937       *  0.588
Exponential              12.296  <0.003
2-Parameter Exponential   3.274  <0.010  0.000
Weibull                   1.191  <0.010
3-Parameter Weibull       0.905   0.017  0.029
Smallest Extreme Value    1.222  <0.010
Largest Extreme Value     0.926   0.017
Gamma                     0.974   0.016
3-Parameter Gamma         1.238       *  1.000
Logistic                  0.964   0.007
Loglogistic               0.958   0.007
3-Parameter Loglogistic   0.946       *  0.528

Across the board, the P-values are way too tiny. Not even a Johnson transformation can make this data analyzable.

That being said, if we disregard these precautions, and just assume that the only reason this is not normal data is due to the rigid 100mhz increments of our data, and analyze it as if it were normal data anyway, with a 95% confidence level, if you go out and buy a 7700K, the max overclock could fall anywhere between 4.6Ghz if extremely unlucky, and 5.2Ghz if extremely lucky, with 35.7% of chips predicted to hit 5.0 or above.

Though, as mentioned above, this analysis is on shaky ground due to the data being non-normal, not matching any other distribution we can analyze, and not able to be transformed, so take it with a grain of salt.
 
You must log in or register to reply here.
Back
Top