Z270 & KabyLake OC

[BitMaster]

Now that my new system seems to have reached some stable settings I will share with you what I found out so far. Mind you, this is the first downloadable Bios 0701 that Asus has for this board. It came with 0505 that would not boot with all 4 modules inserted so I had to flash it right at the assembly point. I am confident newer Bios releases will even open the envelope further, more tolerant on hardware and better overall performance and reliability.

 

The system as is would not prime95 out-of-the-box. That was astonishing !

 

It needed an 0.050V UP on vCore to pass p95 for 1h. This due to the fact that Asus asks you if you want the "ALL CORE ENHANCEMENT" activated as well when you click XMP. I always said yes and that caused p95 to fail. The reason is, by Intel settings, the 7700k has 4.2 base clock, Turboboost 4.4 on ALL cores and 4.5 on ONE core. Asus then applies Turboboost 4.5 to ALL cores when you say YES when prompted. Since p95 uses AVX the power draw jumps SIGNIFICANTLY when you come to small FFT's, small enough to fit in L1 Cache, that's when things get hairy, hot and lots of juice runs. For example, on 5G and all maxed, the jump is from 130 Watt at 320k length to 150 Watt when doing 8 or 12k FFT. Still, it either does it or not and I have come back to p95 despite I have written otherwise before. My main reason is that p95 solely tortures the CPU and RAM but leaves the GPU out. RealBench is a nice "after all has been set" benchmark to finally combine or "cook" your separate ingredients ( CPU RAM GPU..all overclocked seperately while the other 2 stay stock settings ) together, to one useful things called GAMING RIG :D

 

Overclocking each component while the other 2 stay calm is one thing, getting them running in harmony together is a more time consuming job but still fun. I like doing this despite that I am sure, for DCS the stock clocked 7700k is plenty. There will not be much difference in those last 500 MHz unless you have a Über-GPU that could make use of it.

 

Anyway, just for the fun & records, here is what I found out for the CPU:

 

When you have an Asus board with said XMP All Core Enhancement, check if you pass p95 the way it is on Auto-ALL. Mine didn't. UP the Offset by 0.050 and try again. Mine then worked.

 

To countercheck, use XMP but say NO to the other, leave vCore on Asus Auto and p95 it, it should work as your CPU now only clocks to 4.4 when priming, not 4.5 anymore.

 

When that is done and corrected, UP the vCore Offset to 0.100 which should result in a total of 1.350V or close to it. Save & Apply --> UP the Multiplier to 48 for ALL CORES and see if it runs stable. Leave ALL THE REST on Auto, dont touch LLC or Phase Control, it is not needed as I found out on my board. You can enable them on various levels but it never helped the system to stay stable when it had problems being stable, those things got solved by pure vCore and LOWERING the LLC and PhaseControls as I had them also engaged in all levels during the tests. It showed that it stays most stable as of now without any LLC change ( DEFAULT is LEVEL 2, on my board )

and Phase Control also on Asus Optimized, not Extreme. Asus Auto-5G-Profile applies a hefty portion of those two settings but it would not prime stable with it ! Up the vCore and take the other 2 out and it ran stable.

 

So, let's assume you reach 4.8 and it wont clip out when you hit apply with a BSOD.

Run a p95 right away, see if it holds it, watch your temps. That setting should NOT cause severe temps btw. You should end somewhere in the 70°C range peak.

If prime95 fails, either disable XMP and try 2133/2400 standard and 4.8 or UP the vCore.

Up it only by 0.010V, should total at 1.360V then, which is already a lot for 4.8G.

 

Mine runs 4.8 with 0.060 added total, 1.31V. Base vCore with XMP-All-Core is 1.250V on this board/bios.

 

So, assuming it holds p95 for an hour you can start LOWERING the vCore WHILE you prime.

Hold each setting for a few minutes, then take another 0.010V off, until it drops the first core !

 

When p95 starts dropping cores, UP by 1 or 2 steps and prime it for an hour now.

If all goes well, you have All Cores @ 4.8, only added a little Voltage and have XMP running as well.

 

IF, your Combo is troublesome, mine was sometimes too, failed each and every test at certain stages, you have to leave XMP alone and concentrate on the CPU only while the RAM hums along with 2133 or 2400, whatever your JDEC is. That would exclude the RAM as a point of failure.

 

Just some XMP is stable enough to do it in 1 step and safe time. Anyway, if you get into trouble, clock the RAM back and do the CPU alone.

 

Once 4.8 is stable, which most if not every 7700k should reach, you can try to find the Voltage for 5G. I would not even consider 4.9G as that would not he where you wanna go I guess. Only if 5G just wont work, you can check if 4.9 is max or if 4.8 already was the highest achievable.

 

All reviews I have read about 7700k could boot the OS with 5G, so that should not be the problem. The question is will it run stable at temps you can accept or will it fail, no matter how much you ramp up the vCore.

 

So, assuming 4.8 was reached with no more than 0.100V added as Offset, you could try 5G like that and hope you have the Golden Die or UP the vCore right away by 0.050V to a total of 1.350V ( again, you have been there before ). Good ones run at 5G with LESS, so you have good chances of having a stable 5G there, if not, up the vCore by 0.010V, as before, prime it.

NOW WATCH TEMPS ! I got easily into the low 90's C when I tested a lot. THAT you cannot accept but for a very short time. I guess the highest temps one should have in p95 at max OC is mid 80's with no core peaking the 90's. Seems high but p95 AVX is beyond daily use, if it holds this, it'll hold anything you could possibly throw at it...CPU wise. RealBench is totally different thing.

 

During all my XMP, the VSAA ( CPU system agent voltage ) , default 1.050V, was raised again considerably, like on the Z170 M9 MSI Board I had, up to 1.4V on 3733MHz and beyond.

 

I tried to lower it but then p95 dropped cores occasionally which gave me a hard time to find the root cause cause I had forgotten about that lowered VSAA that i safed in the profile.

 

After I put that back to Auto and let it go UP in auto mode when on XMP and higher settings it primed again when all the rest was also properly set.

 

 

This is written down in 30min but as you can see, each p95 run takes at least an hour, many of the more critical settings took a while to fail in p95. If your vCore is far to low, cores drop fast and you soon know what to do, just UP it, but if cores drop after 25min or 40 min it is usually not mainly the vCore, maby eonly 0.005V then, it is usually your IMC or PCH, or your RAM, which is related to the IMC Voltage. This is the tricky part.

 

Now I had trouble to get 5G and XMP together, they did like each other by default when priming.

 

Turned out that I had to set the Voltage myself, not Auto-OC with Asus-5G profile for my combo.

It boots, it games, etc. on that profile, it even will pass RealBench, but it will fail p95 in seconds.

I upped my vCore by manually setting the Offset, LOWERED the LLC to default ( Level 2 ) and Phase to Asus Optimized default...and OLA..it kept stable 5G/3600 XMP for a few 1h sessions.

 

Once I had that stable I went on to the RAM and see where they would go to. The CPU as said would even boot cold with 5.2 but temps are outside of this world. I have to delid that candidate to get there. Not sure yet if I 3D-print the tool and do it...I am tempted I have to admit as I have vCore headroom :D

 

I went to Bios, clicked 3733 F10 and GO :D BOOTED

Did the whole 1h testrun and RB, on STOCK CPU, the CPU was in default then. I highly recommend you dont OC the CPU then 1st place. wait till you find the highest it goes and then drag the CPU up to that point as well. It is bound to give you trouble if you keep the CPU at 5G while you try 4000MHz.

 

All tests passed I went back to Bios, 4000MHZ , changed CL to 17-17-17-38 , F10 & GO --> BOOTED, man I smiled :D

 

It loaded OS without any glitch but failed ALL cores in less than 1 second...ahh I had choosen the 5G and forget I had the DRAM at 1.35 still as for XMP and 3733, for 4000 I wanted 1.4 right away, maybe more. So I dialed in 1.4V vDIMM and ..SAME. a bit better, didnt stall right away, maybe took 2 seconds :) Dialed i 1.45....better...dialed in 1.6 !! not better.

 

OK, they dont like 4G and I dont like any vDIMM near 1.4 or even above.

 

I went back to Bios, tried 3866 but didnt get that stable with 5G. It works for hours with the CPU at stock settings but as soon as I 5G it the RAM becomes troublesome in p95. Not that I think it will not work BUT with 3733 I can keep my really low CL16 setting and only need to UP the vDimm by 0.015V to 1.365V, which is still OK for the CPU's IMC and the RAM. VSAA needs 1.375V as well.

I am unsure if I keep the 3733 setting. The toll is still fairly high, The VSAA is hitting the 1.4V when priming, for 133MHz more I am not killing my IMC. If I cant lower it and keep it stable I will evert to XMP and leave it there, 3600 is not bad at all, given the lower Volts it needs.

 

Yeah, I aalso added the Cooler that came with them, they now have the Vengeance Air Flow Cooler on top, 6cm fan, spinning up to 7k rpm. It's a quiet fan, I was not able to hear it in 3.5k rpm mode, when it goes to 7k all the other fans go UP too and cover that noise. It's a gimmick, I would have not bought it if it wasnt included.

 

While I type on the Mac the rig is doing another stability run, 5G/3733MHz 1.376vCore manual ( it still applies a curve, there is no manual Voltage anymore in this Bios as on older Boards, it is always a curve an no flat line, like 1.376V --------. The system also applies more Volts than you dial in, add a good portion to what you set is what it is actually fed, according to all meters there are. DRAM at 1.4058V and VSA is 1.3750V ( default is 1.050 !! ). It seems like it is sometimes dropping a core, not in every testrun, if any of those two, vDIMM and VSA, are at 1.365 and 1.2500V for VSA. I dont think both are needed that high, but I need a reproducable stable plattform 1st before I can lower any Voltage and see if it remains stable for longer runs then, 2-8h needed then I guess to find weak spots.

 

It still primes, this 1h run seems to work, if it does another one for a few hours I will start to lower VSA first, as that value is like in the reds, vIMM at 1.4 is rather uncritical, just not nice. It passed afew runs with 1.365V but now that seems to fail every now and then.

 

 

Bottom line:

 

5G on water, yes, any furhter needs a delid and relid procedure of the CPU. Getting there, need my pal 1st to setup his brand new 3D-printer. We have not rpinted ANYTHING yet and I wont print the tool as a numero uno and put it in the vice, HELL NO !:book:

 

XMP yes, any further settings come at a price. I hope that new Bios' will improve RAM overclocking.

 

I also hope future Bios' will address the "I wont prime out of the box" Voltage issue with XMP-ALL-CORE-ENHANCEMENT, that is rather nasty for beginners and people who dont even know what a Bios is for and how to get there. Out of the box it MUST work. I dont see any room for interpretation there tbh, but things can happen and it is the 1st Bios available.

 

The board is VERY predictable. My past expereince with 400-500€ boards was far worse, actually horrible down to the bone. This board doesnt have fancy Qcode, IT DOESNT NEED ONE ! It boots the way it comes, it fixes itself when you overdo it with OC, I NEVER NEEDED TO PULL ANY RAM OUT to get it booting other then when I put it together and had to falsh the factory Bios to acccept my XMP. That was the only time I pulled 2 of them. Very very good job Asus.

The most you have to do is disconnect the 230V plug for 30 sec and plug it back in...Overclocking failed...Press F1.....very nice.

 

 

It still primes, 30min now, hottest core 84°C on 5G/3733...acceptable...but its winter and its NOT HOT in my room, its rather chilli...yeah...I am boosting OC with an open window ! Hehe

 

I guess when Summer comes I need to dowclock to 4.8 or delid it. In summer with 28°C in the room the 5G will remain a thing for cold days and nights but not 24/7.

 

Will I use it at 5G ?

 

Yes :D

 

Will I kill it sooner therefor ?

 

Probably !

 

 

 

Bit ;)

[BitMaster]

Things I forgot to mention:

 

The RAM alone was stable at 3866MHz as long as the CPU wasnt overclocked.

When you the started to OC the CPU the IMC would not hold the high setting no matter what I tried till now.

 

 

The core that drops on my CPU is usually #3 , numbered 0-7. If another one drops, it's usually #4, that I could manage to stop by just raising Vcore till 5.0 where that stops to work.

 

The problem is not the individual component but when you oc them together because the IMC will behave differently when you overclock high. That is the real tricky part, find your sweetspot somehwere there. Actually 3733 might even work at 4.8 with CL16 and stock vDIMM at 1.35...never tried that I have to admit. Sounds like a plan cause the last p95 from above dropped #3 again....after like 40minutes !

 

That tells me 2 things: A 15min doesnt tell you much when finetuning B Its gotta be the IMC making things complicated, from what I think it is, that integrated memory controller is the tricky part. It is of such a kind, that it takes quiet a time to provoke an error in p95. I cant even tell you which part it is, both alone do each setting without dropping cores.

 

I am now doing another 4.8/3600 calibration run to see it's still reprodcable stable at modest 4.8 and stock XMP. If that succeeds I will go 3733 and see if the IMC goes higher with less oc on the CPU. It does 3866 at stock CPU, that in mind.

[The LT]

Wait, so 7700K need that delidding/relidding crap again? I kinda thought Intel learned the lesson with Haswell.

[Andrei]

Wait, so 7700K need that delidding/relidding crap again? I kinda thought Intel learned the lesson with Haswell.

 

They have learned the lession, and continued applying crappy thermal interface further down the line. Whether a person fries his CPU due to overheating or damages via de-lidding attempt - intel wins anyway as new CPU has to be bought :)

[BitMaster]

The weaker part is still the IMC, that you cant fix with brute force.

 

The 5G is not stable if I raise the RAM to XMP or even further !

 

It is stable enough to prime for 1-2h but may drop a core on every other run, not always.

 

I would have to delid it, apply some 0.015V more and that may work, but right now that would cook my KabyCrap :D

 

4.8 is rock solid with XMP and temps are A LOT lower, due to Voltage being considerably lower.

 

 

I tried RAM as low as 2666MHz and it would still drop core 3 or 4 after 20-45min...sometimes.

 

With 2133/2400 it holds it...must be the IMC then I assume.

 

 

I dont think I will delid it, the CPU is not "golden" enough to reach 5.1 or 5.2 after the beheading. If it was any better I would consider it.

 

 

Maybe, with a new Bios things change, till then I settle 4800/3600 @ 1.25V adaptive +0.075 Offset ( total 1.3250V ).

 

I was able to lower VSA to 1.25 manual setting and the board stays stable, there is no need to leave it on Auto and feed 1.425V thru VSA, this is insane !

 

 

IPC is ~ 2900 +/- a few and its pretty fast, I dont miss the 200MHz to 5G tbh. That is fun only :D

 

 

 

I testflew the Ka-50 on Kaukasus, lowest was 70ish and if not there it was 120-140fps, OFFLINE.

 

Havent really tested DCS yet due to SF kicking my lics if my PC tilts...aka RAM in and out or such. YOu simply cxannot use DCS if you work your PC hard, StarForce will punish you !

Edited February 17, 2017 by BitMaster

[PureEvil]

Whether a person fries his CPU due to overheating or damages via de-lidding attempt - intel wins anyway as new CPU has to be bought :)

 

https://click.intel.com/tuningplan/ - covers overheating.

nothing covers damage caused by delidding.

 

Haswell-E has soldered IHS for those who don't dare delid.

[BitMaster]

I dunno what ruined my 1st 6700k but it died after I worked the MSI M9 and Asus VIII Extreme board and the CPU hard :D

 

Intel replaced the CPU without any questions, that I have to say. Took about a week and I had a new 6700k, boxed !

 

Delidding...think twice and check if your CPU is worth it. Mine needs to show it can run 5.0 without trouble with a new Bios at RAM speed higher than JDEC2133. If the IMC wont hold, I dont see a reason to do it as with 4.8 my temps are still all green, even AVXed & primed for hours. Just did another 5h run without error.

 

 

Thing is, this darn Bios is brand new and the 1st of it's kind. I dont expect stability as with Bios #10 or so...

 

 

Still testing and priming.... quiet evenings with some Rum and stuff :D

 

 

Let it smoke baby.....:D

[tintifaxl]

I dunno what ruined my 1st 6700k but it died after I worked the MSI M9 and Asus VIII Extreme board and the CPU hard :D

 

Intel replaced the CPU without any questions, that I have to say. Took about a week and I had a new 6700k, boxed !

 

Delidding...think twice and check if your CPU is worth it. Mine needs to show it can run 5.0 without trouble with a new Bios at RAM speed higher than JDEC2133. If the IMC wont hold, I dont see a reason to do it as with 4.8 my temps are still all green, even AVXed & primed for hours. Just did another 5h run without error.

 

 

Thing is, this darn Bios is brand new and the 1st of it's kind. I dont expect stability as with Bios #10 or so...

 

 

Still testing and priming.... quiet evenings with some Rum and stuff :D

 

 

Let it smoke baby.....:D

 

FYI info: my board works flawlessly. I don't overclock, because performance is good enough for me atm.

[BitMaster]

FYI info: my board works flawlessly. I don't overclock, because performance is good enough for me atm.

 

It's not because of the performance, that is plenty with stock speeds for any task I have, even gaming. I also dont see much gain in DCS, it's for pure fun :D

 

Sure, it's trial & error, lots of frustration one could argue if a p95 fails after so and so many minutes...and still you think...ahh there is one more dial that I can turn...MAYBE..

 

That MAYBE and the search for the unknown drives us overclockers, it's the battle vs. the beast.

 

 

You set yourself a goal and get there, very simple.

 

My goal is stable 5G "WITH" XMP....can't be that hard...just a matter of Voltage & Cooling Asus says...so lets go and find the setting :D

 

 

WHat I am dealing with now is VCCIO, raisng that to 1.30v seems to help 5G/XMP...testing right now.

 

 

Dont ask me why I OC at a sunday noon...it's FUN :D

[BitMaster]

stable for 30min with temps at the edge of insanity :D

 

once it shows to remain stable...it might be that it doesnt really need the 1.4000 vCore but hopefully only 1.375vCore.

 

Once you change 1 dial, all others are in question again, maybe they dont need to be that high if VCCIO solved the XMP stability issue.

 

 

Yeah...161.403Watt peak OMG

[BitMaster]

Maybe I take the cooler block off and re-TIM that thing with a different brand.

 

I used Noctua this time and not my usual Ceramic-2 compound. I may should do that to hopefully cut 1-2 degrees. I also think I may have put too little Noctua on it..that might explain some temp nightmares as well. My 6700k was WAY cooler, way way cooler.

[BitMaster]

5.0Ghz with reasonable temps is easily achieved with 4790K and Noctua NH-D15.

 

I wouldnt say "easy" but it's in reach for some of the DC's but not all.

 

 

Meanwhile, I got mine stable at 5G-3866 :D

 

 

Thaiphoon.exe helped me to find the die that is on my Corsair RAM, turned out to be Samsung "B", the best OC there is according to some overclockers forums.

 

There is a table of settings available...and what can I say..it boots till 4133MHz into OS.

 

Just any setting higher 3866 means you have to increase Latency by 1-3 cycles and/or raise Voltage even further, that I dont want as I am already a bit up on VCCIA and SA Voltage, both 1.3000V has proven to be stable for hours now, RAM at 1.4035v as well, CPU @ 1.375v was also enough. 1.4vCore was not needed, what helped a lot was VCCIO to 1.3..and suddenly everything worked. As I then found out that a lower vCore will also be stable I was able to lower temps enough to call it 24/7 worthy, which I would not really be at 1.4vCore...unless delidded.

 

I may delid that thing soon. Now that I tasted the potential and know my way around the KabyLake die and my RAM things make fun again.

 

 

It took me 2 weeks to find the settings it needed to run stable. there are dozens of ways to get to 5GHz or more but only some pass the hour long stress tests repeatingly. That I learned the hard way.

 

 

It is considerably harder to stabilize the KabyLake at high RAM frequencies than compared to stock JDEC settings and 2133MHz 1.2v. VCCIO has solved that problem, NOT vCore.

 

 

Also, might be the early bios, when I change the dials a lot, it will at a certain point fail all tests until you do a cold reboot. SOmehow, I got the feeling the Bios and Nuovoton chip dont like each other a lot when you rotate the dials a lot :D Nothing you cant fix, but you have to find out first that it is like that or you do many tests that say nothing and cannot be reproduced as the dials you look at have a different value than what you think. Thats the only way I can explain to me, as when I cold reboot the same settings qualify again. Early Bios problems I guess.

 

 

I have done multiple RealBench 1h and Aida64 Reliabilty Test for 1h also, it runs :D

 

Finally

 

 

I call a day and will settle here for a while until I delid it and go for 5.X, wherever this may lead too.

 

 

IPC is 2930 in Passmark and Memory score is beyond max, 4216 :D tilting the graph HEHE

 

 

I still wanna see if 4000 or 4133 at a slower latency will gain some speed or if it's contraproductive to go beyond 3866, which still holds the same CL as XMP-3600, CL16.

 

 

 

Now, I will get back to DCS :joystick:

 

*will post some screenshots soon

[BitMaster]

This is a 2h Aida64 Stability Test. Right before that it passed 1h RealBench.

 

*highest core temp was 84°C on those 2 runs

[trigen]

This is a 2h Aida64 Stability Test. Right before that it passed 1h RealBench.

 

*highest core temp was 84°C on those 2 runs

 

Running about the same as mine 5ghz @1.375 80ish C on water. Really should delid it. For prime i gotta set -2x on AVX so it runs at 4.8ghz at 5 it burns to a crisp instantly

[BitMaster]

Yep, at 1.375v its a "whaaaaaat NINTYTWOPLUS" if you use p95_v28.10_build1

 

Use p96 v26.6 for the smoother variant. Much more realistic and no AVX.

 

There is plenty AVX in realbench to test that

 

 

Gesendet von iPhone mit Tapatalk

 

edit* highest temp after 1h p95_v26.6 is 69°C vs 91°C on p95_v28.10

 

to be fair, Volts are lowered to 1.320V adaptive, no offset...for now

Edited February 22, 2017 by BitMaster

[BitMaster]

Btw... Intelburntest.exe also works great.

Set it to max and wait for bsod. Lol

 

Linpack sits underneath

 

 

Gesendet von iPhone mit Tapatalk

[BitMaster]

Also, the ultimative 30€ Question:

 

Either one sells for 30, DieLidMate2 or Intel CPU insurance ( covers oc and overheating, but no delidding!!)

 

Which one ?

 

 

Gesendet von iPhone mit Tapatalk

[BitMaster]

This I found just yesterday, hit F5 and tried it their way.....not bad I have to say.

 

The article explains some topics in detail that make it easier to turn the dials once you know what they exactly do, what consequences + and - has etc.. its a MUST READ I assume:

 

https://rog.asus.com/articles/guides/the-kaby-lake-overclocking-guide/

[BitMaster]

Running about the same as mine 5ghz @1.375 80ish C on water. Really should delid it. For prime i gotta set -2x on AVX so it runs at 4.8ghz at 5 it burns to a crisp instantly

 

I have been there too but apperently the chip offers more ways to achieve stable 5G then just pure vCore.

 

 

The last week I just tried lowering Volts again after achieving a stable state that I am happy with.

 

My die behaves as follows with Volts, LLC/Phase's, VCCIO ( RAM Signal Voltage, very crucial thing ! ), VCCSA ( talks to PCH and other Peripheral devices ) and a little bit of vDIMM.

 

Findings:

 

First off, the RAM would not pass any serious test at any CPU speed at XMP or higher, the vDIMM needed 1.3728v as lowest to not fail p95-AVX and Aida64 Stresstest 1h each.

 

Aida64 in particular is very good in tweaking Volts to correct Values for my approach of tuning. It pays if it just fails the test and not freeze or BSOD. prime95 is also very good but AIDA64 errors when p95-avx runs for an hour, Aida64 is more picky on RAM and finds those weaknesses a lot faster.

 

Second. VCCSA is 1.35...life with it ! On fast RAM that is like this, you may lower it once done and check if it remains stable, I have left it there as anything below 1.3 wont work or wont even post the Bios, same with VCCIO, to low and it wont post, too high and your signal noise screws up the show. VCCIO is very critical to have the correct value, just not only "high enuff". Too high fails badly, worse than a high vCore or vDIMM, that wont neccessarily void a test, just produce stupid heat.

 

 

Till 4.8, things are very easy...no LCC needed, no stronger Phase control, no 0.01 IA AC/DC LL settings, etc... very straight forward and bullet proof.

vCore is 1.30 VCCIO 1.225 SA 1.35 dimm1.3738 rest auto

 

at 4.9 to remain stable for hours p95-avx it needed Phase140% ( before 100% ) and PhaseControl Extreme ( before Optimized ), LLC remains at default value 2. rest as 4.8 settings

 

5.0 needs a lot more tweaking, it was not that easy to go from 4.9 to 5 as it was from 4.8 to 4.9. Totally different approach and it took 2 weeks with daily testing on what would do it finally, maybe compromising and AVX offset.

 

Well, it turned out, I dont need AVX offset at 5G at all.

 

My previous and early 5G runs were with higher Volts to remain stable, but for the move to 5.0 the actual thing it needed was to move LLC2 to LLC6 and VCCIO to 1.30, vCore 1.330

 

That is stable in Handbrake real world AVX for 4h now, its running while I type, doing 38GB of videos from mp4 to mkv, temps are low 70's°C. It passes AIDA64 and p95-non-avx for hours and does the p95-AVX but needs higher Voltage for that, which finally would need an offset of 1 to 49x as with 4.9 it still does full p95-avx and doesnt get critical temps.

 

 

The thing is, with high phase control and load line calibration as well for 5G I was able to clock just as high, but with 0.05-0.1v almost lower in vCore, resulting in a lot less heat overall and thus more stability in the longer runs, say longer than 1h.

 

 

So maybe, if you havent tried that out, employ a higher Phase control, maybe also add LLC and see if your CPU will remain stable at lower vCore.

PowerPhase Control and LLC if raised does also produce more heat, but I could lower the vCore further than I needed to add with the other two, so I saved same °C bottom line.

[BitMaster]

BTW, I flew DCS with 5.1, no crash no BSOD, so for light loads, that 5G setting also allows gaming at 5.1. Just be aware that it is on the edge.

 

 

The RAM...for now I gave up on 3866, not because of the RAM but because of the VCCIO and IMC. Every paper I read says stay at 3600 max, any higher and high OC causes headaches, lots of time investment and maybe wont pay.

 

I have done dozens of runs at 3866 and 4.8-5G, at 5G things get even more tricky than with XMP3600 alone. I am not sacrificing latency over bandwidth at this point and Volts are already high enough on the VCC-IO Signal Controller, any higher would damage it over time, and the modules I guess as well.

 

For gaming, it passed all games at 5.1-3866, easy, games dont tax a system like AVX or IBT does, the watts being drawn on any game is peanuts compared to a full AVX load with 93% RAM filled as well.

All those Benchmarks are peanuts too, they will never show a real hidden weakness, CineBench, UBM, 3dMark...Kindergarten...you run those last to score it, but not to check it. It passes all those with 5.1-3866...just dont try Aida64 or p95-avx

or IBT....the worst I have is an alpha version from Hardocp forum....that one will BSOD it if it fails...but OMG...didnt know you can top p95 loadwise and wattwise, but this thing does, called Statuscore.

 

Which game does this ? NONE !

 

But honestly, I dont see much difference between 4.8-5.1. Just Watts and temps climb in a very steep curve :P

Edited February 28, 2017 by BitMaster

[BitMaster]

For Andrej,

 

I have made some screenshots from my AI suite, you have to have your Bios at Auto Adaptive as well, if your Bios is set to another Voltage Control Mode the AI Suite's TurboV may not show "Adaptive" volt control options.

 

I have NOT DELIDDED (yet)

 

Temps: My temps in anything put p95-AVX are modest; linpack ok too, handbrake etc

 

Just p95-avx after 20min will cause 90+ to 100°C WCS but not drop a core.

That needs a delid and -10 to -20°C less peak to be 1h testproof I guess.

 

Try prime95 version 26.6b3, your temps should not reach 90°C at smallest FFT's.

If they do, scale down Volts or MHz.

 

 

Yeah, that Heatkiller-IV waterblock is a true heatkiller :)