How to Overclock the Alienware 18 and Haswell CPU (or actually have it run full stock Turbo Speed)

yeah, lets open up these AW's and throw a pot on them, see how they run when they're subzero. haha... damn that's a good idea... hmmm...

 

If it throttles under dry ice I would blame the motherboard

 

In the 4 months that I have had my stock machine I've never been able to get more than within 20% of the promised 4.3 GHz with any benchmark that I've tried for any period of time. Does such a single condition really exist in a system as delivered and has anyone actually reproduced it?

This is a significant departure from the previous overclocked Alienware, which I still have, or any other overclocked system that I've purchased which demonstrated target performance on a regular basis and degraded in a predictable manner. I agree that Dell and Intel don't make any promises for overclockers that wish to go beyond what the system can do as delivered, but this is a case where customers paid $500 for a 4.3 GHz performance option that was not delivered. Is it not?

 

Well if they had any sense they would have looked to see if the machine can hit 4.3ghz even for an instant, if not, then yes that would leave them open.

 

Figured that I had quite a warm room, right now towards winter I didnt dare to aim towards 4.3Ghz.

I overclocked to 4.1Ghz, tried one game so far, fans went on in overdrive for a few minutes when it went up to 80 degrees. It cooled down quite quickly again. What worries me is that it wasnt even that demanding a game. Will try a session of BF4 later to see how it holds up

Thanks Mr. Fox for the guide!

 

My CPU goes up to 90C+ dissipating 70W of power in benchmarks. I wonder if it's normal. There must be some way to increase the heat transfer. Maybe the copper plate isn't perfectly flat, maybe I need to sand it and polish it. Maybe I applied too much paste, but with very little paste the temperature is not even across the cores.

Gosh I am getting paranoid.

 

Yeah, Haswell sucks... plain and simple. You can thank Intel for that rather than blame Alienware. You can run it at 4.3GHz on all cores, and higher, as long as you can keep it cool enough and use the right settings.





For most it seems 4.1GHz is more realistic... They probably do need to stop advertising it because it does not and cannot run at 4.3GHz with default settings. In fact, it can't even run at default turbo clock ratios with default BIOS settings.

 

this is starting to get frustrating, as all we can do is sit around and wait - for nothing in the end...

i think i read some statement conc. repasting of the cpu (i think it was written by mr. fox). did you see some benefit? i'm thinking about using my icd24, but atm to me it seems pretty pointless, as the problems are not a little bit too hot, but just smokin' hot on oc, which will not be fixed by some repaste-job ;-(

 

Oh I noticed the fan speed cap now. The max CPU fan RPM is 4300 RPM, but unless you force it using HWiNFO it will only spin up to 3800 RPM at most. I am not sure why they limit it that way, at 3800 RPM it's noisy enough to be noticeable, I don't really care if it's 500 RPM faster. I am sure most users want the CPU to stay cooler.

 

yeah, but when you force it using hwinfo, the gpu fans just shut off.. =/ same here, 1. i want the fans to constantly spin at 2000, don't care about the noise, and 2. i want the to spin faster earlier, i don't want to see the cpu burning at 90+ degrees and then after a few minutes the fan slowly turning on... -.-

if it was possible to have hwinfo working as well with the gpu fans, problem (for me) would be solved, but according to my compatriot svl7 the ec on this board is a big binary-blob making it a real mess... -.-

 

Yeah Mr. Fox told me that the GPU fans would stop spinning if you control the CPU fan speed manually. I set the CPU fan RPM to max just for CPU benchmarks. As long as you don't load up the GPU they won't overheat. I set the global power option to adaptive in Nvidia CP so the GPU clock speed can ramp down.

I repasted the CPU again today, with much less paste. Temp dropped by a few Celsius, but I still notice a hot core. One core will run much cooler than the rest. Overall the pattern is consistent across repasting, although the overall temps are slightly lower. I checked the surface of copper plate and notice quite a few defects on it, which makes it not smooth.

I also noticed that screw 2 and 4 exerts more tension than screw 1 and 3, maybe it is intended design. I tried pressing down on the heatsink while running CPU benchmark, and I don't really see any improvement in terms of temperature. I have also put some 0.5mm thermal pad on the 4 chokes/inductor above the CPU. Previously I was using 1.5mm pad and that prevented proper contact with the CPU die, causing high temps.

The next step would be to use different grits of sandpaper to polish the copper plate and make it thinner. That should improve the heat transfer somewhat. Maybe in December I will try this. I will need some sort of hard glass surface, but with the screws in the way it won't be easy, unless I want to break the plastic washers holding the screws in.

I also observed that the heatpipes for the GPUs are much hotter than the heatpipes for the CPU. I touched them while 3Dmark11 was running. It's what I call direct contact, and they were hot.

If someone has a infrared thermometer, you could try measuring the temps of the CPU heatpipes from the left to right. There are 3 of them. I suspect towards the left side of the CPU die is where the iGPU is, since that heatpipe is not as hot.

It seems that the bigger die size on the GPU helps in dissipating the heat, even though it uses more power than the CPU. Maybe some of the power goes to the GDDR5 vRAM, and the heat is more spread out.

 

I read that Intel might move the integrated voltage regulator back onto the mobo. There's no telling yet if Intel will bifurcate the desktop CPUs and laptop CPUs lineup, but if they do move the VR off chip, it's quite likely that the current HM87 mainboard won't be able to support the next gen Broadwell CPU, even though the pins are compatible.

And that means, we are stuck.

 

I would be totally shocked if it would support Broadwell anyhow. It would be a foolish business move for Intel to make new CPU architecture work on older chipsets because they and their business partners will lose money on simple CPU upgrades without corresponding new system and motherboard purchases. Plus, there's no real guarantee Broadwell won't be worse than Haswell and not worth upgrading to. I know a lot of folks thought I was crazy for being skeptical about Haswell, but the trend is toward producing more mediocre garbage and my skepticism turned out to be justified. It was not rocket science or insider knowledge... just common sense that smaller and "more efficient" (code word for compromised) is going to turn out to be a piece of crap. It usually does and I expect Broadwell to be more of a big to do about nothing.

Just stop and think about it. How much hype would they generate if they said, "Hey everybody, check out our awesome new less-powerful SoC processors that allow us to mass produce cheaper computers! They throttle severely, overheat uncontrollably, but are capable of being down-clocked to run slow enough to not melt an Ultrabook. They don't overclock well, but at lower clock speeds they are more powerful than our previous processors running at an equally slow clock speed."

Let's hear the fat lady's tune first. No need to get too excited at this point. It's always nicer to be pleasantly surprised instead of disappointed by the outcome.

 

It is a new process which should bring about inherent improvements even if architecturally they are not helping things.

I am sceptical about broadwell and tuning potential though.

 

Just tried with some BF4 at 4.1Ghz for about an hour, worked just great, might just be me but it felt like it ran so much more smoothly. Fans where quite high all the time, but definitely fine.

Next goal is to put new bios on both 780s and see if I can improve them some as well

 

i used modified vbios from svl7, and to my surprise the 780m s perform better even on default settings (svl7 told me they will, but i was still amazed ;-) )

what were the max temps on your cpu, did you log them? crysis 3 i.e. can crank up the heat quite well -.-

 

Unfortionately I run BF4 in 64-bit mode, Apparently Afterburner does not show the temperatures and such then so I couldnt see what they went up to.

During X-Com Enemy Within(Unknown) they all bounced up to 80C and down to 60. So I'm guessing they went up around the 80s constantly

 

Well, you can use EVGA Precision X, MSI Afterburner, NVIDIA Inspector graphs or logging, or HWiNFO64 logging to see what happened. It won't be on-screen display in real time, but if there were any unusual drops, temperature problems, etc., etc., you would be able to see what the behavior was.

 

that would be interesting. i have it running at 4ghz 24/7 atm, but even if i only play i.e. fear 3 for like 20 minutes the cpu heat rockets up to ~89 degrees. it's rock stable though (4ghz is not really a lot of course). hello fantable update by dellienware where are thou?

 

My little 19 inch 1440x900 2nd display is great for having monitoring apps open on the side

Gpu-z is usually my favorite tool.

 

If you have that option, it's great watching monitoring tools on a second display.

If not, even this will work: http://forum.notebookreview.com/ali...are-pc-system-monitoring-program-android.html

 

So many issues with this new gen. Glad I have a warranty on my r2 till Nov 2016, she's amazing because she just.. works.

Sent from my SM-N900T using Tapatalk

 

Now to think of it, since the OEM Haswell is slower and hotter than the pre-production samples, I wonder would the QS chips run faster and cooler.

Like this one here: New Intel Core i7 4930MX QS 57W 3 0 GHz 3 9 GHz 8MB Cache Haswell | eBay

Too bad I don't have the cash to give it a try. Even if I have it's rather risky. But there's no doubt that the pre-production samples sent for reviews run easily 15C colder than the production samples ( Retail Versions of Intel Core i).

 

Trying something like that could be the ticket. So far so one here has struck it lucky in the silicon lottery.

 

Posted this the other day in the benchmark thread...

 

What's the core voltage for 1111MHz on the GPU?

How do you prevent the machine from shutting down from pulling too much power, apart from the TS toggle?

 

If I remember correctly, that was 1.1V. It's impossible to keep the machine from turning off except for running a 3.2GHz ThrottleStop profile for Test 1 through Test 4, then flipping it up to 4.5GHz for only the Physics and Combined Tests. It makes it about 2 seconds into Test 1 at 4.5GHz. This is necessary with the M18xR2 as well, with 1.1V on 680M and 780M using a single AC adapter.

I noticed when I repasted the CPU that the copper plate on the heat sink had a very dark rectangle around the perimeter of the die contact area from the heat. I've never seen that on my M18xR2 heat sinks before.

 

What I did was TS toggle only near the end of Test 1, after the FPS drop. It starts out with 110 fps+, and when it drops I toggled it if not it will shut off.

I also have the rectangle shape around the copper, although mine is not that obvious on all 4 corners. It's only at two of the diagonal corners. I am thinking of lapping the copper but I will need to get a few grits of sand paper. I will also need some small piece of glass.

The thing I noticed was, if I toggle it to 3.2GHz at the start of test 1, the FPS drops rather low very quickly.

 

Set it to 3.2GHz before clicking the orange button to launch 3DMark11. Let it stay at 3.2GHz until the first appearance of the loading screen for the Physics test, and only flip it to your max overclock at this point and you'll probably get a better overall score. At 3.2GHz and 1111/1500 Test 1 begins at 123 FPS and ends at 90+ FPS. Even though it may not shut down on you after Test 1 completes, at 4.1GHz, the power depletion causes a reduction in performance throughout Tests 1 through 4. At 4.3GHz and higher it will not complete Tests 1, 2, 3 or 4 without shutting off with the 780M cards overclocked/over-volted that high along with the CPU.

For 4.5GHz I used 1.400V Core Voltage with a +.125mV offset in XTU. I used 120.000A Core Current Limit and 180.000A Processor Current Limit, 10 seconds for Time Window and 200W each for the Long and Short Duration Power Limits. I had the Cache Ratio at 46x, default voltage on Cache and Core #1 at 46x, the rest at 45x. Be sure to set ThrottleStop accordingly or it will override what you set in XTU.

These settings are similar to, if not slightly higher, than the 3920XM requires for a stable 4.5GHz overclock. The 3920XM seldom goes over 100W and the 4930MX easily draws 130W at this clock speed. (I think the peak was around 138W.) While it might be efficient running at stock speeds, when overclocked to this level is not "efficient" in comparison to Sandy or Ivy by any stretch of the imagination. This is basically the same power demand (130W) as the 4930K Ivy-E desktop CPU. It's a MASSIVE power hog, which is a serious problem with a meager 330W AC adapter and some sort of power limiter on the Haswell motherboard that prevents the dual 330W AC adapter from providing any benefit.

 

Well I think the 4930K overclocked to 4GHz+ will eat a bit more than 130W

We need a 450W PSU.

 

The PSU's are a massive limiting factor. I'm even thinking 330 isn't even enough for gaming come the day a CPU and GPU OC is needed. The thing is even a tiny OC brings these beast to the limit.

If Dell just made a bigger or dual PSU mod life would be alot easier.

 

It's not Dell that makes it. It's Delta or Flextronic or some other OEM. For Dell to commission them to make a special 450W PSU, it'd be rather costly. Unless people from the Sager side push for it too, higher wattage PSU is not coming anytime soon.

 

We don't need Dell to make a better PSU for the 18. Having 660W of AC power available yields no benefit whatsoever for the 18 because Alienware put something new or different on the motherboard that causes the system to turn off at a certain load point without tripping the AC adapter. The dual AC adapter setup that works flawlessly on the M18xR2 functions exactly the same as a single 330W AC adapter on the 18. So, first things first, we need them to need to either remove or increase the capacity of whatever kind of power limiter crap device they added to the 18 motherboard. I'm sure it was done with good intentions, but it is an unnecessary protective measure and a huge impediment to achieving extraordinary performance. We get that hurdle out of the way, then we can handle the PSU part without any assistance.

It seems like all new tech (hardware, firmware and OS) made a hard left turn in 2013 and the whole industry has taken on a highly unethical " you get exactly what we want you to have... period" type of approach. Having everything locked down or crippled sucks. I kind of half way expect this aberrant approach from Micro$lop Mafia and the crApple Gestapo, but imposing artificial hardware limitations and firmware lock-downs is over-the-top unacceptable. This makes me not want to buy anything new. It feels like renting instead of owning. If new is not better and not faster, and it is extremely difficult or impossible to mod it, then there is absolutely no reason to consider buying it. If you can get better performance through an upgrade or buy used stuff for less money, then it's foolish to spend more money on something that is crippled just because it's a new and shiny object.

 

this is awesome! especially the temps! when i did my 14.9k run with 3DMark11, i had the cores at 100degrees, BUT with the laptop standing underneath an open window with ~5 degrees cold air coming in. and this was with icd24 applied already...

 

check this example: this is after playing 20 minutes of crysis 3, the settings are a mix of "high" and "very high"..



EDIT: this is not to like, this sucks! ;-)))))

 

Yeesh I get that with 6 cores stress testing at 4.3ghz.

 

I find that setting Core Voltage Mode and Processor Cache Voltage Mode from Adaptive to Static brought about tremendous improvement to both stability and temperature. For instance, both my processor and cache frequencies do not fluctuate at all during XTU benchmark. For reference my core and cache are at 1.16V and 1.17V respectively for 41/41/41/41.

 

In adaptive mode the lowest voltage you can offset it to is 1.167V. I believe this is a problem with the BIOS rather than the CPU. I have seen Asus mobo that can go lower in adaptive mode.

Downside with static mode is that the idle temp is a bit higher than adaptive mode.

With Haswell you don't really want to run any synthetic benchmark with adaptive mode. Or so I read from Googling around. Certain benchmark that uses AVX will make the CPU pull much higher voltage than the max voltage you set in adaptive mode.

 

interesting! hm, just for a try, could you post me a screenshot of your settings in xtu for this?

 

One doesn't really need 1.16V at 4.1GHz. I run 4.1GHz @ 1.13V. It's stable in a few benchmarks. I am still testing the power settings.

View attachment 105726

 

Initially I found that 1.165 was stable for 4.2GHz but not 1.16. I ended up keeping things at 4.1GHz because with 4.2 I can get thermal shutdowns with Intel Burn Test, which doesn't quite cut it as "stable" for me.. From what I've read so far individual chips are showing a large amount of variance so it's best to explore optimal settings by oneself.

BTW what's up with the 82W power limit? Afraid to melt your desk?

 

82W is all that it needs to run at full speed.

 

I guess you're right that during normal usage the CPU will never need to draw more than 82W.. But then again Prime or IBT easily goes up to something like 95W even at 4.1GHz..

It's a good strategy though, limiting yourself to a set thermal target allows you to get up to crazy clocks for most applications. Even though it might throttle during CPU stress testing.. how does that hurt anyone

 

Oh how do you keep the CPU cool dissipating 95W in AVX optimized Prime95's small FFT test?

What's your power setting then? How much is long duration boost power, how much is short duration boost power?

 

thx for your input i'm sorry for all the noobish questions, as i'm still reading into intel xtu (there's some settings which i don't really understand what exactly they contribute to the whole image ;-) what i DO know though, is how to repaste in a propper way, so it's not like i smear e tablespoon full of t.p. on the die or forgot to apply it or anything like that haha

so i tested the following settings and did a 3DMark11 run (gpus on default clocks). the max temp on a cpu core was 89 degrees C.

if i do the stresstest within intel xtu for 3 minutes, the max temp is 98 degrees.

EDIT: if i use 1.13V on the core and do a 5 min stress test with the core multis at 39 i get a max temp of 96 degrees C.

EDIT2: ThrottleStop is used to avoid throttling down of the cpu.



what i am trying to find out is, if i'm just dealing with the fantable problem and haswell with the higher heat generation (then for the moment it's ok for me and i hope for dellienware to release some updated fantables..) or if there's a hint of anything not being like it's supposed to.

 

I use HWiNFO64 to spin up the CPU fan to its max RPM before I run any stress test on the CPU. The fan table is slow to react, it's a problem, the bigger problem is the heat from the CPU. And it's not total power usage that contributes to the heat, it's just localized heat due to high heat density.

Each CPU will run differently so you shouldn't have to copy my XTU. Take it as a guideline and find your own settings.

 

well i'm not really concerned about the benchmark heat, i'm just trying to find a setting, which makes a game playable at a certain performance (i.e. crysis 3 on default cpu settings, call it "the dell standard" - is poorly playable, while with the cpu holding it's speed it's already a different story). kind of like a game-profile where i can play a game with a nice performance without having to worry about the cpu hovering around the lower 90s all the time.
yeah for sure, i was acutally just wondering what other people use and wanted to see how my cpu behaves using those settings

well, the haswell heat behavior isn't too much fun, and especially with not being able to use hwinfo for the fantableoverride, it's more of a hasslewell ;-)

 

Maxed out.

They're just the limit... use everything else to determine the other stuff..

 

Have you been able to hit P15k in 3Dmark11 with your settings?

Setting max power would just result in thermal shutdown on mine. I prefer to use power to throttle the CPU rather than using current.

 

Sadly no. I get 15k graphics, 10k physics and 9k combined for P13k+.. That is with 4.1GHz, a slight GPU OC+undervolt (941/2650/0.987 which actually improves the score for me).