Tuesday, July 3rd 2018

Core i7-8700K Overclocked to 7.34 GHz (3c/3t) on Z270 Chipset, Bags SuperPi Record

German professional overclocker Dancop got the Intel Core i7-8700K processor to work on an ASUS ROG Maximus IX Apex (Z270) motherboard, something that's not supposed to work. CPU-Z screenshots seem to confirm this unholy union between the 6-core "Coffee Lake" processor and a 200-series chipset motherboard, using a custom "0084" BIOS dated 11th June, 2018. Dancop then proceeded to overclock the chip to 7344 MHz using extreme cooling, and 2x 8 GB (dual-channel) DDR4-4000 memory. This bench-stable build was then used to bag a SuperPi 32M world-record.

There's a rather big catch, though. Half the cores on the i7-8700K were disabled, and so was HyperThreading (not that it's relevant to SuperPi). The 3-core/3-thread chip was still bench-stable at 7344 MHz, crunching SuperPi 32M in a world-record 7.609 seconds. The clock speed was achieved by dialing up the multiplier to 73.0x, with the base-clock probably untouched, at 100.61 MHz (with the +0.61 MHz probably being variance). Supporting this clock was a blistering core voltage of 1.984 V. A liquid-nitrogen evaporator was used to tame the CPU and motherboard VRM. Find the validation in the source link below.
Source: Dancop (HWBot)
Add your own comment

75 Comments on Core i7-8700K Overclocked to 7.34 GHz (3c/3t) on Z270 Chipset, Bags SuperPi Record

#26
bug
R0H1T said:
I guess I should specify certain Linux distros, although you can also run XP in safe mode & it'd consume even less resources.
Yup "Linux" is a dangerous term to use without given context ;)
Posted on Reply
#27
kastriot
So this is like half/quarter record.. :)
Posted on Reply
#28
bug
kastriot said:
So this is like half/quarter record.. :)
No, it's a proper record.
You don't complain about turning and/or handling in a record-setting dragster, do you?
Posted on Reply
#29
Caring1
"2x 8 GB (dual-channel) DDR4-4000 memory"
Actually it shows he used dual channel DDR4-2132MHz Ram with an XMP rating of 3600MHz, and overclocked that to 4148MHz.
Posted on Reply
#30
bug
Caring1 said:
"2x 8 GB (dual-channel) DDR4-4000 memory"
Actually it shows he used dual channel DDR4-2132MHz Ram with an XMP rating of 3600MHz, and overclocked that to 4148MHz.
Which is rather interesting, since you can buy RAM that runs faster than that out of the box. SuperPi must be sensitive to a combination of bandwidth and latency, I guess.
Posted on Reply
#31
R0H1T
Caring1 said:
"2x 8 GB (dual-channel) DDR4-4000 memory"
Actually it shows he used dual channel DDR4-2132MHz Ram with an XMP rating of 3600MHz, and overclocked that to 4148MHz.
Like the other guy said, that's one hell of a record. CAS 12 at that frequency is amazing :eek:
Posted on Reply
#32
bug
R0H1T said:
Like the other guy said, that's one hell of a record. CAS 12 at that frequency is amazing :eek:
Actually, back in DDR-400 days, the best sticks could do CL1.5. At DDR4-4000, that would translate to CL15. So it's roughly the same latency. Memory used to run @2.5V+ back then, though.
Posted on Reply
#33
Dancop
Quick and dirty:
The z270 board supports 6 cores with the modification and bios. But no ht!
This cpu can run 6/12 7010mhz cinebench r15 on Maximus X Apex
For overclockers like me, the benchmark is ancient, yes, but it is by far the most difficult. Same settings like in my record, done on a normal win 10 would take at least 12 seconds more. On a normal xp around 10 and on my slipped XP with the same cpu speed but without mem tweaking at least 15 seconds more...
You don't have to love this kind of overclocking, but it's like formula 1...either u like it or u just can't understand why they do this.
Btw... This chip does 5650@1.45v cinebench r15 on aio!
Posted on Reply
#34
Prima.Vera
@Dancop
How and/or who provide you the custom bios for the board? Also, why didn't you use a Z370 board? Just curios.
Posted on Reply
#36
Bones
Nice job Dancop!

I can explain why he used XP and not all cores for the run.

First, XP is allowed by HWbot, the ruling body that sets the rules and standards for such like FIFA does for soccer or the IOC does for the olympics.

As long as the run was done according to the rules about the bench it's fine, the rules state that a Windows operating system is to be used and only Windows, the rules also say with Super PI any Windows version can be used so it's up to the bencher what to use.
CORRECTION:
Super PI can only be ran with Windows versions 7 and earlier with exceptions for Win 8 and 10 - Can be ran with those under certain circumstances but not at large.

In some cases you cannot use the later OS versions because of what's known as the RTC bug.
This causes inaccurate results with certain benchies running Win 8 or Win 10 because those use RTC as part of how the OS operates - So based on that it's not always legal to use Win 10 or Win 8. Some benchies are affected by it, some are not.

XP also gives the best results with Super PI, doesn't make sense to use an OS that will slow it down - As long as the bot says it's OK, then it's OK provided all other rules about it are followed to make it a legit run.

As for not using all cores Super PI isn't multi threaded and using addtional cores doesn't make any difference in the result like it does for Cinebench as an example so it's legal to do that if you want. The result would be the same whether running all cores or not at a given speed in MHz, by disabling cores that's not needed to run it he can clock it higher for a better result. For other benchies like WPrime the number of cores used affects the result so you'd want all cores going for that.

As for the bot being the standard, the same guy that runs it is also the author (At least in part) of CPU-Z and that's also a standard we all use just like GPU-Z is used for GPU's.

Edited for clarity of the post.
Posted on Reply
#37
Vario
@Dancop congratulations on your record! And I think it is all the more fantastic that you did it on Z270, a thumb in Intel's eye for killing Coffeelake support for the older chipsets.
Posted on Reply
#38
dj-electric
Dancop said:
Quick and dirty:
The z270 board supports 6 cores with the modification and bios. But no ht!
This cpu can run 6/12 7010mhz cinebench r15 on Maximus X Apex
For overclockers like me, the benchmark is ancient, yes, but it is by far the most difficult. Same settings like in my record, done on a normal win 10 would take at least 12 seconds more. On a normal xp around 10 and on my slipped XP with the same cpu speed but without mem tweaking at least 15 seconds more...
You don't have to love this kind of overclocking, but it's like formula 1...either u like it or u just can't understand why they do this.
Btw... This chip does 5650@1.45v cinebench r15 on aio!
Hi, thanks for the reply.

I would like to know more about what types of CFL configs actually work on Z270. So you're saying that the 8700K is actually capable of working correctly in a 6C\6C config in the board for general purpose computing?
Posted on Reply
#40
Reeves81x
7.609 second or 4 minutes 7.609 seconds.. ahhh. lol TPU rather major typo.

ExV6k said:
Is it running only 3C/3T for overclocking purposes, or is it a platform limit?
I highly suspect that it was a core limit he set in his custom bios. who knows what else he disabled on the silicon to achieve this... so not legit.
Posted on Reply
#41
Bones
Reeves81x said:
7.609 second or 4 minutes 7.609 seconds.. ahhh. lol TPU rather major typo.


I highly suspect that it was a core limit he set in his custom bios. who knows what else he disabled on the silicon to achieve this... so not legit.
As long as he followed the rules for Super PI it is legit.

Disabling cores is allowed under the rules for Super PI because the number of cores used does not affect the computational output/result of the bench.
Posted on Reply
#42
Dancop
I can run the same with 6/6 but since there's no need, better disable it...otherwise, there's nothing diabled, only the benchmark was set to core 3
Posted on Reply
#44
bug
Reeves81x said:
7.609 second or 4 minutes 7.609 seconds.. ahhh. lol TPU rather major typo.
Probably top scores being in the 4 minutes something ballpark, the 4 minutes are supposed to be a given? :P
Posted on Reply
#45
cadaveca
My name is Dave
Dancop said:
I can run the same with 6/6 but since there's no need, better disable it...otherwise, there's nothing diabled, only the benchmark was set to core 3
Can you explain how this translate into real-world uses? This is the sore part of these demonstrations... seeing how record benching in turn helps "normal" users.

You got killer CPU and pretty decent rams for this and 4 min Pi is crazy. what was CPU temp while benchmark is running? Is it delidded? Many questions.

ShockG said:
The reason is that's the benchmark he can run at that frequency and the one he chose to tune for. There isn't another CPU in existence that can give this performance or run at this clock speeds.
A gem of a CPU. He could have run 2 cores as well.
I can bet you nobody on this forum could get this result even at the same clocks, simply because it's more than just cpu clock.
It's plenty of DRAM tuning, which (from the dram reviews) is beyond not only the people doing the DRAM reviews but most certainly the audience (as gathered from what gets written here).
Either way, glad he is pushing the limits of what is possible with silicon.
I mean, benchmarking is nice, and it's rewarding to see increases in your scores when you benchmark, but for games its about as useful as a wet sock. Providing the bridge that actually shows us how this translates would do wonders, and the idea of "BIOS Tuning" working when automatic rules are garbage and so many complain about it makes many wonder why... the number of posts I have seen in the past few days on here about problems with ASUS BIOS might make people question why time is invested in this, and not normal usage stability... like great, ASUS board helped benchmarks. But 24/7 usage having fleeting stability raises the question as to how useful doing these things might be. Fix the problems that people encounter every day, rather than the ones that are seen once in a while, would go a long way. I saw this because I used to like ASUS, but I'm not the only one not impressed with how things are with their products lately, and seeing demos like this doesn't help fix those problems. That leaves people with "Great, but I still have this problem" in their mind... and kills brand loyalty.
Posted on Reply
#46
Bruno Vieira
dj-electric said:
Hi, thanks for the reply.

I would like to know more about what types of CFL configs actually work on Z270. So you're saying that the 8700K is actually capable of working correctly in a 6C\6C config in the board for general purpose computing?
CFL on 100/200 series chipset is stable by now. He could also run the 8700k on other modded bios for full HT support. I've an 8700k (6c/12t) on my Asus z270-e 5.1Ghz 1.32v for 3 months now. The HW mods are very simple, no solder and you must have an SPI to flash the modded bios.
Posted on Reply
#47
Dancop
Delidded for sure, Thermal Grizzly Kryonaut used on die and IHs and pot temp has been -196°C
Posted on Reply
#48
dj-electric
Bruno Vieira said:
CFL on 100/200 series chipset is stable by now. He could also run the 8700k on other modded bios for full HT support. I've an 8700k (6c/12t) on my Asus z270-e 5.1Ghz 1.32v for 3 months now. The HW mods are very simple, no solder and you must have an SPI to flash the modded bios.
I have a Z270-k board and a 8700K to try and do this... wukk dig more into it
Posted on Reply
#50
cadaveca
My name is Dave
Dancop said:
Delidded for sure, Thermal Grizzly Kryonaut used on die and IHs and pot temp has been -196°C
Thanks! Would be great to see a video of getting everything ready... maybe a series of videos... getting OS prepped. Getting board and CPU and mems and VGA prepped. BIOS set-up. Mods. Then the benchmarks... a full capture of every step... every single one... would be great to see. Even when things go wrong... we like drama! Think back to Macci's videos back in x850 days and beyond... but with more info. XOC is missing this sort of personality that isn't so tied into competition... it would be great.
Posted on Reply
Add your own comment