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

I was asking Mr. Woodzstack, who currently hold the highest XTU benchmark score.

I'd imagine he will have no problem working some of his magic on the GPUs too.

 

I once hit 14000, I know my cpu was like 10k, and the physics 9k, and like 15,000 on the graphics... but so many damn settings, and the fact it was weeks ago - I can not recall much of what I was using or doing.

Im finding the 1.2v vbios for the graphics cards waaaay to high, same as the 1v vbios' and now I am looking for something around 0.91625v .... as for cpu... if you can make it not spike too quickly, and you always HAVE ENOUGH POWER TO HOLD THE CLOCKS, it will throttle normally.. staying around 99-101 degree's celcius.... it crashes from not enough power. OR too large a vDroop.. meaning too much current being instantly drawn, causing enough friction to cause it to setoff all kinds of safety features, whereas -heat wise - it hasn't reached its thermal limit. but it is ridiculous the amount of power it needs to be stable at 4.3Ghz... about 160 watts... if you want it to throttle between 4.1-4.3 then like 135-148 should be your goal... now the issue with XTU and voltages, is, the VID from XTU is paired with the IA.... meaning, while your making one thing stable, its making another thing unstable. That, if I had to guess, is intel's fault, and is whomever either designed the hm87 chipset or programmed 4.2 version of XTU.

I've gotten stable 4.3Ghz clocks without using anything but the bios, and still broke 100k MIPS which is about the same as 960+ on the XTU scores in their bench. that was just from using the Flex override. BUT - I had the 780m's removed. not enough power to have all those in there, and only use bios.
but 4.1Ghz, using just bios, and a 0.9v vbios for the 780's works great in videogames for me... i'd say its about the best balances, wisper quiet, stable for over a week with 10+ games loaded and revolving /alt-tabbed...and no issues resuming them...

hope this helps..

it could also be, I have one of those so called higher performing cpu's too, I don't know about variations between them or not. If it means anything, I've been trying to sell this laptop, but only scammers seem interested, LOL.. haven't trie don this website/NBR though, just local craigslist and kijiji etc..

no one seems interested. Haven't had a single issue with it too...

anyways, ask me what you will, ill get to it as I can when I can..

 

Yup, the power demands of the 4930MX with a heavy overclock is totally insane. It consumes close to 50% of the available 330W of power and leaves hardly anything for the 780M SLI, much less the rest of the system to function with. It adds insult to injury that it throttles and runs too hot on top of being a severe power hog.

If you sell it are you going to grab a Clevo P570WM with 4930K and 780M SLI w/ dual AC adapter?

This is the first time ever that I have considered buying a new Clevo to be a more attractive proposition than buying a new Alienware... but only on account of Haswell's mediocrity. I still love the Alienware brand and there's a lot to sacrifice with that move... superior build quality, superior aesthetics, superior warranty, more than 45-60 minutes of battery run time... but at the end of the day performance matters most and a Haswell just ain't gonna cut it. The entire Haswell concept is built around nasty compromises instead of extreme performance--thanks a lot Intel.

It would be really nice if Alienware would release a new version of the 18 with Ivy-E 4930K, X79 chipset, dual GPU and either dual 330W or single 660W AC adapter. That would be really sweet, and worth the price difference. (The P570WM is about $1,000 cheaper, but I would gladly pay the higher price for an 18 with the same Ivy-E CPU, X79 chipset, with 780M SLI and an overly-abundant power supply.)

 

If I sell it I will get a desktop with 4930K with dual GTX 780 TI Kingpin. Or dual MSI Lightning 290X which is coming out soon.

I have since gone back to 4GHz (again). Since at 4.1GHz with the 780M SLI overclocked the system pulls too much power, I either trip the mainboard circuit breaker or the PSU circuit breaker. I have not been able to go higher than P14.8k in 3Dmark11. Also the 4930MX isn't that stable at 4.1GHz, unless I raise the core voltage quite a bit. That in turn creates quite of bit of heat. I want to maintain 100% stability for now.

Without unlocked BIOS I don't know what settings might work, what might not. Like Mr. Fox has said, we need to set the trip point to something much higher.

So you're saying set the power to max? I get crashes from setting stuff to max. What's the processor current limit that you use?

Where do you get the 0.9V vbios for the 780M?

Are you going to try to hit P15k with a higher GPU overclock?

Have you ever ran Prime95 version 27's small FFT test? It's a class in its own in generating heat. Much higher than XTU, wPrime or anything I have ever seen. Maybe only IBT or OCCT with AVX enabled can match it.

 

nah, I don't use p95 often.

as for vbios's theres a bunch floating around, with different voltages stock, as NVidia inspector as example, doesn't seem stable in keeping the voltage set to what you set it to, on these 18's. -.95 is very stable for me, but generates too much heat still... and 0.9v doesn't get hot enough, thanks to the copperplates I threw in under the heatsink to replace shim pads/paste...

Im not really bothering with clocking this beast. It has far too many holes that make it far less enjoyable then I wish. Once it works good, I cross my fingers and pray it stays working and try an enjoy it.

This cpu would be fine, if :
-the chipset was controllable in bios.
-we could circumvent the PSU to add more power
-the fan tables would work/could be customized


any one of those options, would create workarounds that could potentially solve our issues.
The Haswell CPU does a great job at throttling, it really does, its almost brilliant, but many are not seeing its beauty, because, at higher clocks it draws sooo much heat, and soo much voltage, that anyting under load at a full OC usually will draw too much current and shut the machine off. You only THINK 1.4v is stable for 4.5Ghz... that's BARELY enough for a full 99% load.... and then, it still needs around 170-180watts. about 145 on average, but theres small spike that need the extra room to play. At that speed, and voltage, and heat, things become much more sensitive - so because XTU and the BIOS end up pairing setting for the cpu and voltage together rather then run them independently, you end up OCing something that is not meant to be overclocked and you crash... ALSO...the darn most retarded thing it, when the FANs finally do hit light speed... they draw an extra 5watts from your system.. that's usually, enough for me - to cause the crash from too much power loss, unless I disabled all ports/usb's lighting, OR took the videocards out, I use one powered port extender for my usb's on a usb3.0 hub to save the machine some power. I use a powered dvi adapter instead of HDMI for the fact 1.4HDMI draws some decent power itself... and to be fair, I even thought about shaping my nic, or mac wall gardening it to like 100k or 1m connection of shape the MTU to like 600 and half duplex, and force it to run at its lowest speeds. I always have Bluetooth and wireless capabilities disabled, as this machine can not afford to use them.

basically, DELL sold us a ferrarri without an engine. Sounds great, but it only LOOKs good. its not practical, might have all leather interiors and sporting all the best options, but it won't even hit first gear and roll out the garage because theres nothing to power it. Might as well claim insurance from an oil drum spill in the garage and watch it burn.

But heres what people can do. Wait until dells next product, pay for the "Accidental" warranty... call them up, and keep replacing your machine with the next generation in 2014 or 2015, 2016.. hopefully by then dell gets it, and understands you want what you paid for. Not the bovineturd they sold us.

 

Do you guys run IBT Linpack? If so, how many GFlops do you get?

Woodzstack I remember you saying that you did Linpack before, how did you prevent thermal trip point of 100C?

 

What's the point in measuring this if you can derive no useful benefit from it in normal use scenarios?

 

The GFlops doesn't matter, it's the stability that does. I'd like to know what's the stable clock speed for other 4930MX users. The throughput will drop as the heat builds up and after a while it will stabilize around a certain value.

How about this, run a 10 pass IBT Linpack on Stress Level High, and see what result you get.


This is what I get from the log. Running at 40x 100.12MHz @ 1.110V. 100W short power, 77W long power. Any higher will result in thermal tripping at 100C in Linpack. Ambient at 24C.
Processor: Intel(R) Core(TM) i7-4930MX CPU @ 3.00GHz
Clock Speed: 1.65 GHz
Active Physical Cores: 8
Total System Memory: 24499 MB

Stress Level: High (2048 MB)
Testing started on 12/8/2013 11:37:55 AM
Time (s) Speed (GFlops) Result
[11:38:35] 27.652 95.5629 4.753050e-002
[11:39:15] 28.457 92.8602 4.753050e-002
[11:39:56] 28.839 91.6296 4.753050e-002
[11:40:37] 28.986 91.1651 4.753050e-002
[11:41:18] 29.105 90.7923 4.753050e-002
[11:41:59] 29.184 90.5487 4.753050e-002
[11:42:40] 29.253 90.3333 4.753050e-002
[11:43:21] 29.340 90.0671 4.753050e-002
[11:44:02] 29.355 90.0206 4.753050e-002
[11:44:43] 29.393 89.9028 4.753050e-002
Testing ended on 12/8/2013 11:44:44 AM
Test Result: Success.
----------------------------

 

Courtesy of Mr. J.Dre.

http://forum.notebookreview.com/ali...re-17-4700mq-cooling-temps-2.html#post9485027

Spoiler

That's a screenshot of the TS' TPL menu on the 17 for the 4930MX. Note that the stock package power limit and package current limit is quite a bit more aggressive than on the 18, which kind of suggest that Dell/Alienware knew quite early on the 18 will have problems coping with the power draw of the CPU + SLI GPU.

The 17 has 240W ac adapter, assuming other components don't draw too much power, with 780M in the 17, there's about 240-120 = 120W of power reserve for the CPU. On the 18 with dual 780M SLI, the 330W ac adapter leaves only 330-2*120 = 90W of power for the CPU. That might explain why they limit the package current limit to 55A on the 18. I don't remember exactly, but I believe the CPU power limits were also lower on the 18 too.

What a shame really.

 

Your observations about power deficiency are right on target. The Alienware 18 cannot adequately meet the 120-130W power demand of a substantially overclocked 4930MX and 780M SLI with a 330W AC adapter and whatever kind of power limiting device exists on the motherboard. The only reason 780M SLI works so fantastic on the M18xR2 is because the 3rd Generation XM CPU draws 20-40W less under heavily overclocked conditions, and a dual AC adapter works for the M18xR2 because there is nothing limiting the amount of power the motherboard is capable of drawing (or the limit is set much higher if there is such a power limiter).

This is all speculation on our part... there's no way to know what kind of thought and planning went into the overall system engineering. However, I tend to think the opposite it true about this being orchestrated because that does not even seem to be logical. I doubt they built it with mistakes like this on purpose. It looks more like customers (true performance enthusiasts) have brought problems to light that they hadn't adequately planned for. And, let's not forget that hindsight is always 20/20 no matter who you are.

The introduction of Secure Flash and locked NVRAM is a chic industry trend, and I suspect they never anticipated getting such vehement pushback and actually losing business in this customer niche over not being able to flash an unlocked BIOS. I think the borked CPU power settings are a result of mistakes and the problems with fan tables may be symptoms of decisions that were made by people more worried about the casual gamers that whine about fan noise than concern for the kind of enthusiast customer that buys an Alienware 18. There are lots of folks that pitch a big fit about fan noise and huge power demands, but those are not hardcore performance enthusiasts... they are people that are willing to make compromises and buy things like the Alienware 14, XPS, Latitude and Inspiron machines rather than the dual-GPU/XM powered leviathan.

CPUs that run at 2.5-3.0GHz and fans that seldom run are fine and dandy for ordinary consumer-grade machines used for running office applications and web browsing. The people that are in charge of design and have final decision-making authority for a machine like the 18 need to be hardcore overclockers that brutalize, tweak the daylights out of these machines and set new benchmark world records in secret before it ever gets seen by any consumers. If it can't beat last years machine in public benchmark records in the hands of an engineer, then it should not go into production. These should be employees with decision-making power that are more concerned about achieving industry-leading benchmark scores and maintaining Alienware's reputation for producing high-performance beasts rather than concerns about meeting power consumption, energy efficiency and low noise level targets because the kind of people that buy the maxed out 18 specs consider all of those to be secondary to industry-leading performance, or they consider them to be downright unimportant.

 

As you go to 1.125v I have seen the TDP on each card register to 160-170W which would each a 330W brick by themselves.

 

ahheeemmmmmm...... No need to discriminate on the 14" series. 18" for hardcore gaming and 14" for travel with family. Best of both worlds and you are still entitled to bish about fan tables

 

Mr. Fox or anyone else, something is not clear for me.

If I set the Power current limit to 95A with XTU on my 4900mq will this stick on reboot?

If not, is there a solution?

I use Linux and therefore can't use XTU or anything that I know.

I don't understand how Alienware can't figure out how to simply set the proper current limits (core/power) properly.

Also, Chris from Dell (community forum) doesn't seem to be up to date on this issue which leads me to believe that there is no A04 in the short term.

 

For me (4930MX) as of BIOS A03, everything set in XTU seems to stick indefinitely but not after any crashes. Just avoid CPU thermal shutdown and crashes due to GPU OC.

I believe that the XTU settings are applied at the BIOS level so they may stick even after booting into Linux. My Linux is on the "work" laptop so can't test it out for you unfortunately.

 

That sounds good!

If it sticks booting in Windows it most probably will stick in Linux.

You vould use a live image like Ubuntu and boot into Linux, however you will need something like "i7z" ( i7z - A better i7 (and now i3, i5) reporting tool for Linux - Google Project Hosting) to correctly report the frequencies.

I use "stress" as a cpu burner.

I will receive my Alienware 18 4900mq tomorrow or Friday so I will test all this and report here.

I am really pissed at this ugly situation.

-

 

Mr. Fox is away for a while and should be back after a few days. XTU's Core Current Limit has been unlocked since A02. Processor Current Limit will stick if you use Legacy Boot. If you're on UEFI you will need ThrottleStop 6.10 to make it stick. That you will need to get it from UncleWebb. Alternatively, you can use Windows 8 fast boot and it will stick, but won't persist upon reboot. So basically the 2 ways are to either stick to Legacy Boot, or use TS 6.10.

I have no experience with Linux on Alienware so can't comment on that. CPU will throttle on the 18 because of power limitation of the 330W ac adapter. With GPU and CPU overclocked, there's not much juice left. The onboard circuit breaker will trip in benchmarks or stress tests if you OC the CPU and the GPU too much.

 

Ok, so people who uses Linux like me.

I found a temporary solution until BIOS is fixed.

The Power Current Limit register is 0x601.

Solution

Use msr-tools and kernel module msr.

rdmsr 0x601

0x101414000001b8

Only the first (16-bits) is interesting.

So, 0x01b8. Now The first 3-bits are for something else (I won't go in details here).

So, 0x01b8 shifted 3-bits = 0x37 = 55 (55A)

To set your own current (i.e. 95A):

wrmsr 0x601 0x101414000002F8

This will set the power current limit to 95A.

-

I will find and set all other interesting msr registers like the cache multiplier.

At least I am now somewhat happy...

 

Great information! Thank you for sharing. Any chance you can share some screen shots showing what you have described for those of us that are not well versed in Linux?

Do your settings get permanently committed to the MSR registers, or do you have to manually reset them after reboot or a cold boot? If the settings can be permanently written using a Linux Live distro booting from USB or DVD that would be pretty sweet.

 

Setting the MSRs are not permanent. You need to add a service script to your boot process.

I will report more settings like CPU Multipliers and soon.

I might also provide a nice script.

-

It would be easy to play in the BIOS flash but I would recommend against it, unless we know the exact storage protocol used, i.e.: is there a checksum or hashing on the data.

If a change breaks this protocol then you can brick your computer rather easily.

-

 

Finally got my Liquid Ultra, and found the time to lap the copper plate.

I ran XTU benchmark once, before repaste and after repaste. Lapping is no fun, now my hands are all stained with copper. Applying Liquid Ultra is even worse, it doesn't really stick to the die as they say. You have to spread it gingerly. I accidentally got some stain on the tiled floor, and there was no way to scrub it off except with the metal brush that came with the kit. So yeah it stains everything.

A tube of Liquid Ultra has 15ml. And how much did I use this time? 10ml. So 5ml left. Luckily I have another tube that's coming from FrozenCPU. I probably will stick to using ICD on the GPU. Only on the CPU will I use the liquid metal TIM. It's too much hassle, and the die is too close to the SMD resistors on the 4930MX (I had to put some non conductive TIM on top of it just to prevent the liquid metal from touching those).







Temperature before and after.

Spoiler

On the other hand, there was 14C drop on the hottest core. Anyhow I'd still expect the CPU to overheat running the latest Linpack with AVX2. The new Intel Linpack with AVX2, can push close to 200 GFlops on the Haswell CPU, and there's where the Haswell will beat the Ivybridge or anything before resoundingly. Problem is, AVX2 stress test makes the CPU runs hotter than everything and anything out there.

I shall go run some benchmarks and see the durability and consistency of this liquid metal TIM.

 

Does your laptop just sit on a desktop with no cooling pad? You can check your CPU temps in games like BF4 and see if you can push the CPU to the desired 4.3Ghz mark without touching 90C. Of course in Intel XTU and other heavy CPU tests the temps will rise way quicker

Sent from my Nexus 4 using Tapatalk 4

 

It sits on a cooling pad, which doesn't do much. The airflow doesn't matter that much, it's the heatsink that's getting saturated from the heat coming off the CPU.

I don't use games to check temps. I use benchmark and stress test. I have raised my benchmark of stability to the rather brutal Linpack recently.

All in all I have finally reached my conclusion, that Haswell runs hot like a . Heat aside, then limited BIOS options and locked down BIOS doesn't help either. Adaptive mode voltage doesn't work for voltage less than 1.167V and I guess that's probably due to the BIOS too. It's one hurdle after another.

I suspect I have a weak core 2 (or core 1, if you're computer science person and love to start from 0). Intel doesn't use common centroid layout on their cores, and that has negative effects where heat isn't distributed evenly on the lateral plane. That hot core is probably sandwiched between the other cores. If I am not mistaken, the coldest core will be beside the GPU, with the hottest core 3rd furthest away from the GPU.

 

I really want to like the Alienware 18 but with all these shortcomings. For the enthusiast the Alienware 18 is a step backwards and should stick to M18x r1 or r2 with 3940xm and 780m sli. The Alienware 18 with 4930mx and 780m sli is way to expensive to justify the performance you would get from a m18x r2 with XM series chip and 780m sli. Until Alienware releases some sort of fix for this issue in way of unlocked bios and proper fan tables. This is a huge fail and Alienware 18 r2 will probably fix all the shortcomings. Sucks to be the guinea pig sometimes especially if you are spending $3k-$4k on a system.

 

Well...if you buy a R2 and then get 780M SLI, that dual GPU will cost you easily another $1000-1600, depending on the deal that you get.

There's no easy fix for this. Having fan tables that ramp up the fans faster will help, but won't fix the core issues.

What they need to do, on Dell's side, is to improve the CPU cooling. There easily space to fit a 10cm or even 12cm wide CPU heatsink in there, but Dell only uses 8cm wide copper heatsink. The GPU aluminum heatsink are both 9cm wide, but because they are at the side, the heatsink kind of slope off. Overall though, the heat dissipation area should be bigger on the GPU heatsinks.

Perhaps it's time to consider having quad pipe heatsink, or thicker/bigger heatpipes. Dell could easily do what MSI did, by linking the CPU heatsink and the GPU heatsink with some pipes.

 

I really don't think there is a legitimate fix for anything running on Haswell if you are an enthusiast that is passionate about overclocking and want to achieve some fantastic benchmark scores. The processor architecture just sucks no matter how you slice it. It's fine for Ultrabooks and ordinary laptops running at slow to average clock speeds, but for enthusiasts Haswell is a real turd. Unless Intel has some kind of major epiphany and pulls their head out, I don't expect the next generation of mobile processors to be any better. For the most serious enthusiasts, the M18xR2 might be the last of the Mohicans. We can thank all of our "friends" that are goo-goo-gah-gah about ultraportable trash and the tree-hugging god of energy conservation worshipers for creating a demand for such overwhelming mediocrity. We can also thank Intel for abandoning their mobile enthusiast clientele. Why they bothered putting the "X" in 4930MX is a mystery. It feels like they just made a conscious decision to lie to get more money from people that expected the X actually meant something special like it did before Haswell.

I hope that I am proven wrong in time, but I fear the days of having a wicked gaming laptop are going to be numbered. The OEMs don't give a rat's butt about us any more. Think about it... the kids are excited about iPads, XBOX One and PS4, handheld gaming devices and Honda Civics with LED lights on the rocker panels. They think an overclocked Snapdragon processor is something to be excited about. They have no real concept of what extreme performance really is and they are tickled to the point of tinkling in their shorts with utter crap. They make stupid comments like "780M SLI is overkill" and some actually believe it. (Some are certainly using criticism as a means of covering for their jealously.) So called "gaming laptops" made by Alienware, MSI, Asus, Clevo and Lenovo with a single half-baked GPU that is less powerful than last year's 680M (in some cases even less potent than 580M or 6990M) are flying off the shelves and people are actually excited about them, LOL... go figure. They are too preoccupied with the the chatter about an imaginary elegant robe to recognize that the emperor is actually running around buck naked.

 

Yeah I've heard 780m SLI is overkill for a gaming laptop many times. Then the same people whine 880m isn't going to be powerful enough step up.. Huh what!? Lol

 

At least the 17 gets some love from Dell, they have A07 BIOS now. We only have A03...

 

What is it three months and counting since the last BIOS update? The next update better have some serious improvements.

 

I can't find throttle stop 6.1?

 

It's technically still Beta. You will have to PM UncleWebb, the person who makes TS, to get the link from him.

 

Please post some screen shots

Does seem that way

No problems with adaptive (dynamic) voltage on my GE60 so probably a BIOS issue. Kevin should hopefully have some Haswell voltage settings in TS, perhaps after the New Year. Static (Fixed) seems to give some strange results in that VID remains rock solid so perhaps the Intel core voltage measuring doesn't really show the true picture.

Anyone played with the bclk straps, 1.25, 1.67... ?

 

Nope, it's easy to do the repaste. Don't even bother with anything except for IC Diamond or the Liquid Ultra stuff that kh90123 just used or you will be wasting your time and money. None of the other pastes do a good enough job. Repasting will help, but until Alienware releases a BIOS/EC update with the fan tables fixed you're going to have a machine that runs much hotter than necessary and with degraded performance. Haswell is already runs way too hot without any help, and the engineers making the fans not work right to try to cut down on noise exacerbates it something fierce. The people that made the decision to do this obviously were not paying close enough attention to what a screwed up product Haswell is. Otherwise, they would have recognized the effort to make the fans too quiet was going to create a far more serious problem that having audible fans. This is what happens when people work in a siloed process and focus on one small piece of a product or project instead of paying attention to the big picture.

 

Pre repaste result.

----------------------------
IntelBurnTest v2.54
Created by AgentGOD
----------------------------

Processor: Intel(R) Core(TM) i7-4930MX CPU @ 3.00GHz
Clock Speed: 0.20 GHz
Active Physical Cores: 8
Total System Memory: 24499 MB

Stress Level: High (2048 MB)
Testing started on 12/8/2013 10:21:20 AM
Time (s) Speed (GFlops) Result
[10:22:00] 27.611 95.7074 4.753050e-002
[10:22:40] 28.377 93.1238 4.753050e-002
[10:23:20] 28.674 92.1598 4.753050e-002
[10:24:01] 28.861 91.5624 4.753050e-002
[10:24:42] 28.813 91.7148 4.753050e-002
[10:25:23] 28.838 91.6329 4.753050e-002
[10:26:03] 28.852 91.5899 4.753050e-002
[10:26:44] 28.846 91.6101 4.753050e-002
[10:27:25] 28.779 91.8225 4.753050e-002
[10:28:05] 28.752 91.9086 4.753050e-002
Testing ended on 12/8/2013 10:28:05 AM
Test Result: Success.
----------------------------

Post repaste result.
----------------------------
IntelBurnTest v2.54
Created by AgentGOD
----------------------------

Processor: Intel(R) Core(TM) i7-4930MX CPU @ 3.00GHz
Clock Speed: 3.58 GHz
Active Physical Cores: 8
Total System Memory: 24499 MB

Stress Level: High (2048 MB)
Testing started on 12/21/2013 2:51:57 PM
Time (s) Speed (GFlops) Result
[14:52:37] 26.820 98.5280 4.753050e-002
[14:53:16] 27.591 95.7747 4.753050e-002
[14:53:56] 27.849 94.8896 4.753050e-002
[14:54:36] 28.083 94.0965 4.753050e-002
Testing ended on 12/21/2013 2:54:45 PM
Test Result: Stopped by user.
----------------------------

I stopped the test before the last run, since the result is stabilizing at around 94 GFlops. I think it might drop to 93 GFlops. Before repaste it would drop to 89 GFlops. Maybe if I find the time I will run the AVX2 Linpack.

Beyond lapping and using liquid metal TIM, I am not sure if there's anything else that I can do. So before repasting, there was no way I could finish a Cinebench run at 4.2GHz, but after repaste it did. Then again I have a conundrum with this CPU. To run the CPU at higher clock speed so I can get better benchmark scores, I will need to apply a higher core voltage. But that is not good for the Linpack stress test, as I will actually get less GFlops there due to higher voltage, under the same power constraint.

If adaptive voltage works for voltage less than 1.167V (I agree with you Dufus, it does seem to be a limitation exclusive to the Alienwares), then I can test out more settings.

I have instability issues with BCLK settings, so I left it alone.

Repasting with Liquid Ultra will produce good result, and so far so good, no degradation yet. You have to be very careful with the liquid metal though, it's like melted solder. It will stain every surface that it touches. And it corrodes aluminum. Apart from the copper plate, copper pipes and copper fins, the grey-ish metal on the heatsink is probably aluminum.

Overall I lowered my load temp from like 90+ to 85-ish. I could possibly clock it at 4.1GHz but I am leaving it at 4GHz for now. So 100MHz more only.

I used to use Shin Etsu X23-7783D, and although it's good, the temperature will increase after 1 or 2 weeks. I will observe and see if this liquid metal hold up over time.

 

That's why I am not planning to use it yet. I will be observing how it does for you before I decide whether or not to give it a whirl. I wish you could overclock your 18 a little more so we can understand how it handles more aggressive clock speeds and more frequent exposure to extreme temperatures.

If it holds up to the kind of heavy overclocking that I like to do it would be pretty sweet, but the thought of the amount of risk and hassle is keeping me from getting excited about using it instead of IC Diamond. If it has to be redone due to degradation, it will be a lot more trouble (and much more expensive) than it is using IC Diamond. I generally get about 90 days of effective cooling with 4.5 to 4.7GHz benchmarking out of IC Diamond. I typically get anywhere from 2 days to 2 weeks out of other brands of thermal pastes with that kind of overclocking until the paste has broken down to the point that it is no longer effective.

 

The gains from overclocking any further for me is just too small, for the risk of instability that I get. If I have time I will run some XTU/Cinebench/wPrime. My ambient temp is rather high too, and I have to account for that.

To put things in perspective, before repaste it was impossible for me to run a wPrime/Cinebench run at 4.2GHz, but after repaste it's possible.

Someone else did some testing on his 3770K here, with quite a few TIM. NH-T1 is on par with the liquid ultra, but after a while it suffers from pump-out effect, so the performance would degrade. I'd suppose that most paste would suffer similar effect, except those very thick paste, like the ICD. They mentioned on their website that ICD7 won't suffer the same effect.

Bare-die testing: A delidded 3770k, an H100, and 9 different TIMs - Page 3 - AnandTech Forums

Max GPU temp on mine is 88C, on the test 1 of 3Dmark11. I am waiting for the ICD7 to come in, then I will repaste the GPU and see how it goes. I will remove those pesky plastic washers too, hopefully that helps.

 

Thanks for the results kh90123.

If you can please try Linx. If you load it to to desktop, go to task manager and set affinity to 1,3,5,7 for it and then select 4 threads in Linx. You should get consistent results, providing there is no throttling, showing optimized fp performance. Check CPU freq, power and temp while running.

Short runs can be used to test the CPU package itself. For example, if the overall thermal resistance between cores and case is 0.5C/W then running 100W will net a 50C rise between ambient and cores. So if ambient were 30C in the laptop then you already have 80C before cooling and would never be able to get the temp below this without using sub-ambient cooling (chilled water, dry ice, LN2...). This is just an example, I'd expect better than 0.5C/W.

Longer runs will test the cooling, how long it takes to heat up from the initial CPU temperature differential. Be aware that the internal temp of the laptop will likely rise too, compounding things.

If you are able to plot temperature then you should see a sharp rise to start (CPU internal thermal resistance) and then a slower rise (cooling). Used to be able to do this with PECI but not sure where it's accessed on the newer platforms.

 

Alright thanks! PM where NBR? or?

 

Will run the Linx later. How do I set affinity in Windows 8?

I can't run 100W, I think I can do 77W, or about there.

Yes I do notice a sharp rise in temperature, to about 80C, then it slowly goes up from there on.

He's on NBR too. Go PM him.

 

Until they fix the fan tables so they respond quickly to temperature changes, cooling is going to be an unnecessarily greater challenge than it needs to be no matter how good and effective the thermal paste is. They absolutely have to get more air flowing through the heat sink radiators, and sooner, before the 18 cooling system is going to achieve its full potential. The delayed reaction of the fans and the 500 RPM reduction imposed on the fan speed capacity is a severe hindrance. Placing greater importance on reducing fan noise versus cooling and performance improvement reflects a disconnect and a lack of understanding about what is most important to the kind of customer than buys an 18 with maxed out hardware specs. The people that actually want quieter fans that seldom run and are satisfied with less performance can always stick with A03.

 

Setting affinity on W8

 

Yeah well, it's only a problem in benchmark where both GPU and CPU are stressed at the same time, like 3Dmark, for other benchmark or CPU stress test, I always make sure to use HWiNFO64 to spin up the fan to its max RPM. Otherwise it'd always overheat.

I don't mind quiet machine, but stopping the GPU fan when the machine is idle heats up the chassis quite a bit. Why not let the fan spin all the time, at low RPM? You really have to put your ears near the vents to hear anything.

And between 3700RPM or 4200/4300RPM, I really hear not much difference in terms of noise. It's noticeable, it can be noisy to some, but I don't run my GPU and CPU at max power/load all the time. My point is, why go the extra length to try to make it quiet, but in the end it doesn't make it much quieter anyway...

It's what I'd call, inefficient engineering.

 

those are awesome temps!

i'm awating some liquid metal pro and what i don't get (because i'm not native english speaking), what do you mean with "lapping the copper plate". i thought liquid metal can be applied on copper coolers, and the coolerplate of the alienware 18 touching the cpu die is copper, right? so did you put something on the plate, before you applied l.m.? also, what did you put on the resistors next to the die? just some tiny amount of some non conductive thermal paste?
sorry for the questions, just wanted to make sure before thrashing the whole system

cheers from basel

 

Liquid metal pro is tougher to remove later on then liquid ultra, so I didn't bother with that one. Lapping the copper plate means I used sandpaper to sand down the surface imperfection on the copper surface touching the CPU die.

The copper plate is copper, but the silvery grey-ish metal holding the plate and the pipes is aluminum. With other thermal paste even if you get those on the aluminum it'll be fine, but with gallium alloy liquid metal (most of the liquid metal TIM are gallium alloys), it will attack the aluminum.

There's a video on Youtube demonstrating the effect of gallium on aluminum. It makes aluminum like cookies. They crumble.


I covered up the SMD resistors (I guess they are resistors, could be caps, I am not familiar with SMD devices yet) with some non conductive thermal paste. I just use the Shin Etsu paste and put some there.

Too bad even with 10C lower temps in wPrime/Cinebench, I can still get the CPU to overheat if I run some Linpack/Linx.

Thanks. I will report back later today or tomorrow.

 

yeah, liquid metal ultra i meant i'm getting, i was just reading something about the pro and it was probaly still in my mind ;-)

ah i see, well, i will get some really thin sandpaper (can't remember the english word for it, with "thin" i mean with really small particles..).
holy crap, now you scared me! did you put anything on the silver-greyish base to cover it, or just made sure to use as little l.m. in order for it not to touch it?

damn, just watched the video again, this is scarry ^^

sorry, one more nooby question: did you sand it using some kind of sand block with the paper around or did you put the paper flat on the table and then made the cooler go over it? just asking since i've never sanded down a cooler block before.

i got those infos from here: How do I properly lap a heatsink/ waterblock? | techPowerUp

 

No to both questions. The surface touching the CPU is copper. I expect to need to lap the copper surface next time I have to repaste.

If you're not comfortable doing it, then don't do it. I took away the washers holding those 4 screws in. The right way would be to rub the surface on the sandpaper which lies flat on a piece of glass. I just used my hand to hold the sandpaper while rubbing it.

 

yeah i know this, i just wasn't sure looking at your pictures if you did something to the aluminum base as well. i also don't have in mind anymore how much space is left between the long side of the die and the aluminum base. i will remember again when i get there ;-)

well, i'm actually really comfirtable with pretty much everything that has to do with electrics, electronics and as well handwork, i just never sanded down a copoer coolerplate (if you sand down anything else, mostly 0.5 mm more or less doesn't relly make a difference, while here it does).
i wasn't sure if the lay-down version is even possible to do with this cooler. if the aluminum is slightly higher than the copper it won't work, if the 4 corners, where the 4 screws go through the aluminum block are
higher than the rest, same story.
thx for your feedback!

 

Lay down way of lapping would not be possible due to the 4 protruding corners where the screws go in, so you have to either use your finger or some tiny piece of glass. I sand it to remove the surface imperfection and the oxide layer of the copper, since it's metal I think less than 0.05mm are sanded away.

 

You don't want to remove much since you will start to reduce the pressure of the heatsink.

 

My heatsink pressure is higher than usual anyway, since I have added some washers to the backplate.

 

Yes but that pressure is good, you don't want to reverse it.