Anybody got a Skylake HQ/HK CPU yet? I want guinea pig-- I mean testers.

ok..... so bear with me!

just going off of this setup above (matching settings and bench conditions), ive been running through prime on and off for 3 or 4 hours now just nudging the settings up (without core throttle), i cant seem to break my TDP..?? maybe a BIOS limitation my end

so, i cant recreate the above results with my laptop currently..... about 3.6ghz average on all 4 cores constantly is all i can muster right now, temps are pretty good..... lots of headroom, plenty of voltage headroom too!!

early days ...... (what do i need to unlock this thing!!)

 

ok....... i think ive cracked it!!!!

4ghz AVG constantly on 4 cores for nearly 5 mins, no throttling / Prime torture test..... temps got very scary, but with better cooling IT IS POSSIBLE!!!!!!, the image actually cut off the package TDP but it was only around 52w..??? thats EXCELLENT is it not..??



watch this space ..... 4.2ghz here i come!!

 

and heres your wPrime run......

solid 4ghz!!!! on all 4 cores / 1024, 8 thread run

at 50%


at 90%


completed score....


notice the unwavering red line ......lol, love this machine!!!

 

erm........ IT DOES!!

 

Fantastic! One other person I believe showed this to be true, so it means likely the 6820HK can hold its clocks should one be capable of cooling it... with the right BIOS setup. As for skunk however...

Nice. I can see package TDP there, though that's so ridiculously low. Can I ask you to have throttlestop open for me while doing the same Wprime benchmark and another P95?

What I want to see is maybe 4 or 5 minutes into P95, and another 90% shot of Wprime. The reason I want throttlestop is to make sure you're using C0 states properly, and it's not somehow using C7 states. Your low TDP and relatively low temperatures seem to not add up with your speed and the voltage the CPU likely needs for that clockspeed. That way I'll also see your voltage. Since you just posted this, I'm assuming you should be ready to go right away =D.

If you just simply somehow got lucky and are using a very small amount of voltage and thus your temps are low (meaning you got a lucky chip) then more power to you good sir. But this information will be great.

After that, I just need to find someone with a 6700HQ that can force his CPU to pass the 45W limit in a stress test while remaining cool, and maybe a 6820HQ/6920HQ user to see if those chips behave any differently at all, and I'll be able to update my guide accordingly. This is good information!

 

right..... here we go ultima.

50%


90%


what a cracking chip!!!, super low volts

 

Well @skunk1 you seem to have both a golden chip and a bad paste job! Your core 3 and 4 temps are 10c lower consistently than your core 1 and 2 temps! You might want to repaste there.

Aside from that, I hope you never have to change that machine out, you're going to cry to lose that chip if you do.

 

Indeed looks like his 6820HK is one of the better ones... Repaste Skunk and let us know.. Get some good Gelid GC Extreme paste... The new AW's suck when it comes to repasting... Takes like 5-10mins in the old ones, these new ones take 1-2 hrs lol..

 

cheers all........ the little AW has really surprised me today!!

certainly not even going to attempt a repaste, its being used by my sons for batman arkham knight currently .... trying to keep em off it today has been a nightmare!!

pull this thing apart and run the risk of bricking it...i might as well leave home.

 

I've got a 6920HQ, actually. Turns out it's not locked! With the right BIOS, of course.

 

Really now? I'd like to see your testing if you please, and to see if you can indeed overclock it =D.

 

You already have! Unless you mean using something besides Prime95.

 

good things come to those who wait acquiring some good data here ultima, looking forward to the update of your guide.

Sent from my Nexus 5 using Tapatalk

 

I don't remember that, actually. You haven't left any posts in this thread about it, and I'm bad with names. Wprime seems to be useful though. If you don't mind doing over your P95 test if your CPU is unlocked (in TDP) that'd be great. If you can overclock it too, that'd be even better to know.

 

Right here. Come on, that was like a few hours ago.

 

Ohh. you never stated your CPU, I assumed 6820HK.

Yeah I didn't look at your original name, my bad. Can I see that it is a 6920HQ using throttlestop in another quick test if you don't mind? It may be tedious, but I *MUST* be certain.

 

Yep, if you look at the XTU screenie it seems the 6920HQ supports an additional 6 bins of OC.

Nice results guys. IMO would be nice to see the power draw on the 6920HQ @ 4x 4GHz running wprime 1.55 (1024M) and result, then compare to the 6820HK posted result.

 

TSBench seems a lot easier on the CPU than Prime95, since it never broke 57W.

Spoiler

Spoiler

 

Well I didn't really need tsbench, I just wanted TS open to see the name and clocks etc, in the same P95, but okay.

Interesting, you get to overclock to 4GHz as a hard limit, but the HK is unlocked. Interesting.

I wonder if the 6820HQ is unlocked? Glad to see the 6920HQ, at least in your machine, has no TDP limit and can be overclocked some. What notebook is that?

 

It's the GT80S-6QF!

 

Oh, I didn't expect a MSI to do it. Now I have a better understanding of your temps too. Well enjoy your notebook then!

 

Nice results @skunk1 I was promised 4.1 and perhaps 4.2Ghz Oc on all 4. Can you test your stable 4.1Ghz on all 4 cores or even 4.2Ghz?

 

I think the difference is in the voltages. Though he didn't show Wprime on the 6920HQ, he did show Prime95, which ran somewhere around 1.25v on average. The 6820K though, during Wprime, was hovering between 1.12v and 1.145v, which is a sharp decrease.

But yes, I'd like to see Wprime as well.

Skylake in desktops too though, seems to have way too much voltage by default. A friend with a 6700K in a P770DM-G has something like 1.05v for 4GHz perfectly stable (a -160mV offset on default voltage), drawing less power in TSBench than I do at 3.8GHz and 1.0854v on my 4800MQ.

Maybe @Katiecat might find adding more of a negative voltage offset might be stable for 4GHz on that chip.

If Skylake would just do what we tell it to do (not bumping voltage randomly when idle, actually listening to manual and static voltages when set, etc) it'd be pretty close to perfect. The 6700HQ however, needs to jump off a cliff. Especially since the suggested tray price of the 6820HQ is the same, and it's faster, with more L3 cache. Might even have small OC bins, since as we just found out the 6920HQ does.

 

lol...... give me a chance, hopefully today ill be able to pin down 4.1ghz

here's a tiny run i tried this morning, needs more voltage me thinks ...... it blue screen'd shortly after the pic, ill get it stable!!

 

could i just ask...... as XTU is still new to me (and had it not been for KatieKat ), there would of been no OC as i had no clue where to start with voltage settings...

so..... when fine tuning this what sliders should i be looking at...????

 

I had stability issues at -100 under load, so I backed it off a decent amount to be conservative.
Also, it seems like the voltage fluctuates way more than it should... earlier, a WP run was dropping to 1.15V and 61W briefly before going back up, but the clocks were stable at 4GHz the entire time.

Spoiler

Spoiler

I'm not an experienced overclocker either, but from what I remember from physics, if you have stability issues, you need to bump up your voltage or drop your clocks, and if you're thermal/current throttling/overheating, you need to drop your voltage and/or clocks.

 

thanks Katie...... but XTU is nothing like BIOS work im used too... it seems that were actually giving the CPU a negative voltage ..?(i.e. actually lowering VCORE with the voltage offset..?? seems strange not to be adding.(or have i got something wrong)

also how did you come to your 105A on the ICCMax..?? or the 200 on the turbo boost..... was it just trial and error, or is there a sneaky webpage somewhere explaining all these voltages...

maybe Ultima can give us a breakdown on what area of the CPU the sliders actually alter...

 

Magic!
Also, those current values were the defaults.

As for undervolting, the reason why you have to add voltage to hold higher clocks is because the voltage can't transition quickly enough at higher clocks to reach the logic 1 threshold, so by increasing the voltage, it hits that point sooner. If you're not running at a fast enough frequency for this to be a problem, there's actually no benefit (that I'm aware of) of higher voltages. In fact, it's a huge problem because the power goes up with the square of the voltage and linearly with the frequency, and current goes up linearly with voltage and frequency, so you're going to heat up your CPU more and hit the throttling limits much faster.

(again, from what I remember from physics!)

 

lol.... maybe im not doing mine any favors by running your settings then, the 200 turbo boost is a looooong way off my default clock of 90.... but it works i suppose with similar temps to yours.....

gonna start messing around with XTU soon, few more coffee's ..... see if we can get some 4.1-4.2ghz lovin!!!

 

Well......... that didnt take very long, following the above posts from katie i decided not to overvolt anything, but move the IccMax from 100 to 105
left the turbo boost at default as well as the offset...

look....

4.1ghz WTF is going on!!, where is this thing pulling its power from..the ether ??... temps are pretty reasonable too. these new machines from AW are superb!!

 

A question for you. During those benches why does the 6820HK show a difference in pkg and core power of less than 3W while the 6920HQ shows 13W?

As for the DT chips I often wonder if a lot of the mobile chips are DT rejects.

@skunk1 a tiny bit of throttling as the ratio is reading 40.8 and not 41.0 but still a very nice result.

 

Yeah i saw that, but throttlestop actually shows it being above with 4100.17.... but its pretty good!!

i think ultimately the heat dissipation is gonna be the problem, i could create a pad with 4 Noctua 3000rpm monsters chucking 280cfm each through the vents but the heat from the chip wont be able to transfer to the block inside the chassis fast enough..... so i reckon that will be the limiting factor.

 

@skunk1 @Katiecat

heres some quick rules to follow in xtu:

- max out all sliders that have units in seconds, watts and amperes. these are basically the limits u set for your cpu to properly stretch its legs in terms of the max wattage its allowed to draw the amperage its allowed to use and the length of time its allowed to use its turbo. no worries, as these are maximum limits, thus your cpu will only take what it needs in order to reach the set multiplier clocks! but this way ull avoid any power limit and tdp throttling

with that set as a base, do the following for proper overclocks:

- set all your multipliers to the same value, so u only have to deal with one multi and one value of clocks across the board, makes life easier and tweaking faster

- set your voltage to static instead of adaptive, that way ull have less voltage fluctuations and more control over core voltage. just beware that u set a voltage right away that u know or suspect to be stable in order to be applied when u do the switch. additional reason for static vs. adapative: u can actually reach lower stable voltages for a given multi than with adaptive, giving u more oc thermal headroom!

- now comes the fun part: at a given multi, lower the core voltage in 5 mV steps until u start getting instabilities. then up the voltage in 5mV steps again until it passed all of your stability tests! if u havent hit the thermal limit yet, proceed to the next higher multi for all four cores with the voltage u used so far as a starting point for stability testing

to give u a proper example with numbers:

- max out all sliders pertaining to wattage, amperage, turbo length in seconds
- set all multis to 4ghz and change core voltage from adaptive to static, setting 1050 mV
- lower the core voltage in 5mV steps until ur machine crashes (lets say it does so at 990mV)
- now test 995mV for stability:

is it boot stable? does the voltage stick at a reboot?
if not, up the voltage by another 5mV and boot test again
if yes, use the occt tool (google) in its current version with large data set on all cores for 20 min with thermal limit set to 96C
if its unstable, up the voltage by another 5mV and test again with occt
if its stable, and your havent met ur 96C target yet, proceed to the next higher multiplier on all cores and repeat this procedure by starting with the previously stable voltage base. only now u dont dial back the voltage anymore at the start, u directly proceed to boot testing!

- lets say uve gotten 1025mV stable for 4.0ghz, save those settings in an xtu profile to keep for later usage

- up the multis by one to 4.1 ghz and proceed stability testing with 1025mV as a starting point

- at some point ull hit the set thermal limit of 96C in occt. that doesnt mean u have to stop here! u can now just switch to the intel xtu stress test for 20 min for stability, thattl give you about 3-4 additional multis of thermal headroom to test your cpu in

rules of thumb:

- anything stable for 20 min large data set occt is rockstable for everyday usage, that also goes for temps

- anything stable for 20 min xtu stress test is good for gaming and benchmarking, but not for everyday usage or marathon gaming sessions of 10 hours or so

- be aware that those stability tests are just for load scenarios. so it could be that ur tested voltages will on rare occasions crash at idle or maybe light loads during browsing. if thats the case, dont panic, just up the voltage another 5mV and then ure definitely 100% good to go just keep that in mind and youll be fine!

cheers


Sent from my Nexus 5 using Tapatalk

 

ok...... here it is!! (after many blue screens)

4.2ghz (or just shy of), slight IccMax adjustment.... little dash of VCORE.

 

good job there

dont forget to max out those turbo boost power / turbo boost short power / processor core IccMax sliders to give your cpu more room to stretch its legs!

 

Yeah, your chip is probably nearer to average is all.

Yes, Skylake does this in stress tests and the like from everything I've witnessed. It happens to the desktop chips in desktop boards as well. In short, don't worry about it unless you BSOD, I guess.

Yes, you're giving a negative voltage modifier. Hence "offset" (displacement from normal) of a negative voltage modifier. When you offset negatively, you're reducing Vcore. When you offset positively, you're bumping Vcore. It's basic math =D. Voltage + (+/- voltage) = new voltage. x + (-y) = x - y. Like that.

ICCMax is the core current limit, I.E. how many amps it is allowed to draw. For these tests, you should not need to pass around 85A or so, but I always find it's a good idea to set my limits to 256A and leave it, as (at least with haswell) 256A kills ALL current limit throttle as long as a user is using a proper EC (embedded controller). Sometimes a buggy EC firmware might cause random throttle when the CPU should otherwise be working fine, but it seems your ECs are working just fine, so don't worry about this.

200W turbo boost is simply how much power (watts) the CPU can draw. You're pulling in the 50W-60W range now, which is fine, but people usually leave this to the limit of their power brick's "excess power". For example, I leave mine to 100W, because each GPU runs on average around 105W and the brick can supply about 380W from the wall for short bursts, so 105W + 105W + 100W = 310W, +20W for the system on average, and if the system needs burst power, it'll be fine. You can tailor it to your brick and CPU as well if you want, though you have a golden chip it seems with such low power draw.

What you did is return voltage to stock, I.E. remove an undervolt. You didn't overvolt anything in the first place =D. The current limit probably didn't do nearly as much; from what I know a CPU won't crash or BSOD from lack of current limit (at least well into the range of 50A); it'll just throttle. Your returning of voltage is likely what did the OC stability.

Throttlestop doesn't use a lot of power in TSBench for those tests. You should try Prime95 and Wprime again with the overclocks; though you're likely to overheat if those are your TSBench temps. I bet your power draw will be a decent bit higher since your voltage and clockspeed has been bumped.

Also, I guarantee you that you just have an exceptional CPU, rather than the AW machines being "good". I've seen more than enough users with problems to know that the machines are not that good. Consider yourself very lucky.

Honestly, I think he just has a very special chip. His system is even using the iGPU for Optimus, being an Alienware, and the 6920HQ in the GT80 has its iGPU disabled, so those values in theory should be switched. But I don't understand a lot of skylake at this time. The chips do not seem to LISTEN to things, and exhibit a lot of weird behaviour.

Truthfully, it should be the opposite; desktop chips should be reject mobile chips that couldn't get temps/tdp/voltage low enough to be useful. But who the hell knows right now, I wouldn't put it past BGAtel to focus on desktops rather than mobiles. In fact, most of the 6700K chips I've seen run their 4GHz stock settings with less voltage than the 6820HK chips for mobiles, which are generally somewhat unstable approaching the 4GHz mark and usually use a lot more voltage. Your theory could be right, for skylake in particular.

It's that his bclk is above 100 (looks like he manually set it there) so the multiplier needs to be somewhat under the number he's trying to attain, as the CPU is basically focusing on a flat speed number. My friend has a similar problem on her 6700K; it'll run 100.20MHz bclk (by default) and multiply at say... 39.84, and end up with just shy of 4GHz (3991.9MHz) or something.

Coolaboratory Liquid Ultra.

Now, can you keep prime95 and Wprime running at that clockspeed? WITH throttlestop up (just monitoring) in the screenshot? And XTU monitoring too, of course.

 

nope.... youre right.

cant hold 4.2 on the big benches, temp soars past 95c within a few seconds ... power limit throttle kicks in and drops clocks!! GRRR the dreaded throttle has returned to haunt me.... dont think theres any way around this one without a decent repaste.. (and some dry ice in direct contact) i could probably game @ 4.2ghz but it wont torture run!!

oh well......

 

I already gave you the answer to your temps above.

But yes, I know the difference between TSBench and those heavier stress tests.

Now, did you say "POWER" limit throttle happened? Not "thermal" limit?

 

yep... ''power throttle limit'' basically doing the same as previously, i think the initial hit of power to the CPU creates such a build up it just cant be pulled away quick enough, chip breaks thermal limit for a millisecond and it throttles, not by much though... seems to hover around 4ghz until stress stops then pops back up....

its gotta be the heat burst.....

 

This power limit throttle means you can't draw over X TDP. What is the limit it gets stuck to? 55W? 57W? If you're seeing power limit throttle, it has NOTHING to do with heat. Thermal throttling is heat, and nothing else. Seems that while you have a golden chip your motherboard doesn't want to let you use it... if you can't figure it out I may have to give MSI the nudge here.

 

i can test the 6700HQ for you. I have some experience overclocking my 3720QM with XTU but i miss the unlocked sliders now . Max watts i ever saw the 6700 consume was 43W. Sits around 35W at full load and around 75-80C

 

If you can't pass 45W in Prime95 or Wprime, could you try overvolting a little and seeing if you can force it? If it overheats, then don't worry about it. From what I've seen so far is that the 6700HQ is so slow that, coupled with Skylake's lower (than haswell/broadwell) power draw, that it won't hit the 45W limit in any kind of scenario unless forced. But I'd like to know for certain.

 

at the start of prima it was around 35 ish, topped at 47W later on. Running at around 45W average.

Spoiler

ill try overvolting it a bit, temps were still fine so i have some headroom there luckaly. While gaming the temps were a lot higher tho


edit: this was too easy i bumped up the vcore by 70mV. The drawn power immediately shot up by a lot

Spoiler

around 50W's drawn. I hope this helps you.

 

If you can hold the 50W for about 5 minutes (without overheating of course) that'd be perfect. I'd need to see the whole XTU graph with CPU util at 100%, and of course with Throttlestop and XTU and prime95 in the shot too. Once that's done, it'd be a great help, thanks very much.

 

so....... after a night of reboots i can run 4.2ghz without throttle while gaming (black ops 3 / batman AK).... but i cant torture run it, so ultimately its not 100% prime stable but it is game and browser stable... thatll do me.

'power limit throttle'

how do we stop it ultima..???, would the 330w brick help..???

 

i tried again yesterday with +75mV. I was running XTU stress test and Prime95 (large FFTs simultaniously for around 10 minutes and saw watts flucuate the whole time from 40-50W. And i would run at 3.1GHz constantly. This did not make any sense to me.

I decided to stop and try something else and very interesting things happend. Decided to run large ffts Prime95 only. For the duration of the turbo watts periode (around 90 seconds or so) the cpu drew 52-51W stable without fluctuating and after that the watts immediately got stable and settled at 45W! Activating the TDP throttle button in Throttlestop. These are exactly the results i expected to get. However the CPU was not able to maintain the 3.1GHz while throtteling and would clock down to 2.9 or 2.8GHz.

Spoiler

Started Xtu stress test again to check what would happend and as before the watts would start fluctuating again and the clockspeed would go up again.

Spoiler

When i put back the voltage to normal cpu would remain at contant power draw of around 41Ws running at 3.1GHz again.


So it seems for ultimate torture it seems better to run Prime lager FFTs only. Hope those results help you.

this is exactly what Skylake was designed for running at turbo watts for small periods of time but not for extended time like in Prime95.

I believe the power limit throttleing is something intel has locked and you cannot change. It is defently not your PSU running out of watts. How many volts are you running at and have you tried dropping your voltage?

 

by volts do you mean vcore..??? if so i cant really get a definitive answer, cpuz show anything between 0.988v - 1.28v

what would you all class as a BIG jump in volts..... if i upped the vcore by 100.00mv on the offset, is that excessive for skylake..??

tbh...... i just think its too much @ 4.2 on such a small die, as you say it works perfectly in bursts, but sustained clocks may not of been what they had in mind with their ''Dynamically Overclocked'' cpu's

still ive managed 4.2ghz on all cores as long as the load is not continuous ...... which is pretty good for a chip designed to run 3.2ghz on all cores..

 

Yes volts is vcore. it depends on your own chip what voltage it can take. Higher clock speeds need higher voltage usualy. However the stock voltage is usualy a bit on the high (safe) side. But since a higher voltage also results in higher watts consumed its not better to just add voltage. Too low voltage will result in an unstable system or crash. It would be best to run your CPU at the least volts it can take while being stable. So decreasing your voltage will result in less TDP throttling and therefor higher clocks.

So if i were you id try to run at a lower voltage and see what happens. 100mV off set is quite a lot so you might still have a lot of room left to lower the voltage. I would drop 10mV at a time and run prime for a bit and see if its stable. If so drop another 10.

Its easy to get a non stable overclock thats not the fun. Its the fun to optimze everything so you can get the highest stable clocks as possible.

 

I'm interested in actual performance (productivity) achieved with these stock or overclocked Skylake (mobile) platforms - does anyone here have an executive overview for me?

Especially compared to SB, IB, Haswell, or Broadwell based platforms...

 

i have an ivy bridge machine. I have ran a lot of bench marks on both these machines. Which ones are you interested in?

My 3720QM at 4ghz beats the 6700HQ in almost all of them except Cinebench i believe