This guy does a good job of describing what is not so great about Haswell processors.
<iframe width='853' height="480" src="//www.youtube.com/embed/FqdLpzwSDFo?rel=0" frameborder='0' allowfullscreen=""></iframe>
-
-
Meaker@Sager Company Representative
At least we don't have to worry about cheap paste between the IHS and chip.
Mr. Fox likes this. -
-
Meaker@Sager Company Representative
Well they can but that can be quite risky and intel should have at least used a decent paste if not solder :/
The real men would keep the lid off and manually adjust a heatsink to go on top. -
Back to the topic of the 4930MX overclocking. Sometime ago, it was the core current limit which couldn't be adjusted, but the processor current limit can be adjusted.
After having 2 mainboard replacement, and installation of Win 8.1, it seems that now I can adjust the core current limit, but the processor current limit would bounce back to 55A everytime upon restart. Does anyone has the same experience? -
You typically have to set Processor Current Limit at every boot. It has always been this way for me.
-
Ok I have managed to reproduce the no POST. With the core voltage at stock 1.2V, trying setting a -5mV dynamic cpu voltage offset in XTU. That should give you a no POST.
Also, XTU is kinda wonky in Win 8/8.1.
Processor current limit in XTU is actually the package current limit in ThrottleStop. Even so, after inputting a value in TS and applying it there, upon reboot it will still get reset back to 55A.
Dell please fix. -
Well this time it took about 25 minutes to revive it. Sometimes I have to make it beep without the CMOS battery in, sometimes with it in, until it probably causes some CMOS error, and starts to beep 5x. Then all is good from there on.
I think I should time myself on this. How fast can one revive an AW 18 that doesn't POST. People can try it with their AW 14 and AW 17 too. It makes good sports. -
Sent from my SPH-L900 using Tapatalk -
One has to wonder, its been about 5 months. Dell are seriously goofing off. Maybe whatever the solution is, its not an easy one.
kh90123 likes this. -
Haswell runs hot and sips power at high overclock. I have mentioned it a few times, at 3GHz it eats 35W+ of power under load, but at 4GHz (33% increase in clock speed), it eats 75W+ of power under load. That's about double.
Dell probably didn't anticipate that overclocking on Haswell is a bit different from how it was compared to Ivybridge or Sandybridge, due to the integrated voltage regulator. They think that raising the voltage and power will be enough, but no, you can keep the voltage at about constant, maybe just a tad more, but you will need to give the CPU a heck lot of current to support the clock speed.
The default settings, what we said buggy/underpowered, is just Dell's choice to keep the GPU and CPU both at 75C.
The fan table that has a hefty delay doesn't help if we start pushing the CPU. But to Dell, since the stock settings limit the CPU and GPU at 75C it's never a problem to them. They won't know until they start pushing it. Well to give credit to them, they probably know how hot Haswell gets, and with 75C max temp target in mind, it just decides how much power they can pump into the CPU. In this case, to get 75C under load you put in 35-40W of power. Dell's settings exactly there, and they lock the TDC at 55A.
It's a conscious design choice, so I believe Dell won't do much after this. After all, raising the TDP and unlocking the Thermal Design Current (TDC, aka Processor Current Limit, aka Package Current Limit) will definitely lead to more failures, and 4930MX are expensive, Dell isn't going to risk having to replace those if some user breaks it under warranty. Most of the time though, it's due to user error, rather than component failure.
But what I dislike is how Dell presume they know better than us. I get to decide how I want to run the CPU. I know the 4930MX can run at 4GHz under load even with ambient temp that's close to 80F. I get to control the fan table, and how loud I want the machine to be. Clevo machines currently provide their users with those choices, even though the stocked BIOS is somewhat lousy, but once you flash it with unlocked BIOS all is good.
At the end of the day, I am sorely disappointed. -
Even under the ideal scenario where everything has been fixed that needs to be fixed, at the end of the day it's still Haswell. It is more than obvious from what you can see posted on the web that Haswell has become a major disappointment to enthusiasts everywhere... from laptops to DIY desktop builds, it's a problem. The best solution is probably to not build or buy anything that uses a Haswell processor. It appears to be the WindowME of CPU world. Intel really stepped on it with this product and that is absolutely not Dell's fault. The computer manufacturers are victims just as we are. They have to listen to unhappy customers more than Intel does.
Intel took one giant step forward by offering all SATA 3 6.0GB/s ports, and two larger steps backward with screwed up power management and overheating. -
Meaker@Sager Company Representative
Haswell is fair enough for new system builders but does nothing for ivy bridge users really, owners should treat it as such.
-
Edit:
Seems like Core Current Limit wasn't unlocked until A03.
Also, it's kinda weird when using Intel XTU. Sometimes there's an option for Flex VID Override that shows up in BIOS, sometimes it doesn't. Does anyone know why? -
Although Mr. Fox has a rather strong disdain towards Haswell CPU, then again I have spent $1100 on it, so I have to make the best out of it. When life gives you lemon, you make lemon juice outta it.
I was running some Cinebench 11.5 in Win 8.1, using A03 BIOS. With Core Current Limit set to 112A, and Processor Current Limit set to 85A, and Power Limit set to 70W. CPU was at 4GHz, cache freq at 4GHz, I was able to get 8.63 in Cinebench, with only 1.1V on the CPU.
I think Mr. Fox were right all along, the voltage is too high. Assuming the 4930MX is like the 4770K (they have the same stock speed, same turbo speed, same amount of cache, I think we have reasons to believe they are just the same, just that the 4930MX is highly binned), 1.2V is too high.
I am not sure how much lower I can bring it, but I will try. Seems like the key here is to abuse the efficiency and IPC of Haswell, undervolt and overcurrent at the same time. I really want to hit the 9 points in Cinebench, without AC. With every 10mV undervolt or so, I am shaving off a few watts, and that gives me lower max temp. The computer would crash in Cinebench when one of the core hits 100C.
Hopefully within next week I will get this fan: Dell Alienware M18x R2 Cooling Fan DP/N: J77H4, 0J77H4, DC28000BHF0. Once I have the higher CFM fan, I will see if I can push the CPU further. I am limited by how hot Haswell runs, as well as the high ambient temp. Preliminary testing shows that the CPU works fine down to 1.07V.
Thanks to Mr. Fox for pointing out that fan in another thread. They ship from Singapore, and it's only 3 hrs drive from there to my place.
Also another thought, running AVX heavy stress test such as Aida64 will probably stress the CPU more than Cinebench, and will make the CPU pull a lot more voltage than it did in Cinebench.
Disabling AVX for stresstest? Haswell Voltage Offset - Benchmarking, system performance - AIDA64 Discussion Forum
Edit:
Seems like the sweet spot to undervolt is at 4GHz. 4.1GHz won't run with 1.1V, since I want to keep this undervolted I can't go higher without better cooling or lower ambient temp. Max from a 4GHz 4930MX should be around 8.7 in Cinebench, with 1.055V core voltage. I get WHEA BSOD at 1.05V, so I will use 1.07V/1.075V/1.08V.
You guys can bench around, play with it and let us know. -
woodzstack Alezka Computers , Official Clevo reseller.
well, Im using stock cooling, and Im at 4.3Ghz all 4 cores. BUT - my power is set to 88 watts in the bios. nothing else really. THOUGH I do hit83-84 degrees at times, and the fans after 30 minutes at that temp finally go on, and then it never passes 70 again.
-
I found out what Dell did. Basically they just locked Processor Current Limit (ICCMAX) to the max value for a 37W processor. Go to page 99 of the Haswell datasheet from Intel, you will see ICCMAX. For 37W CPU, the max is 55A (what we have now, it's underpowering the CPU). For 47W, the max is 85A. For the 57W 4930MX, the max is 95A. I need only 85A to sustain 4GHz, so 95A max is about right. Higher than that and Intel can't guarantee long term reliability. Also from browsing through the registers, it doesn't seem like there's any form of current limitation on CPU. There's only power and voltage. So any form of current limitation comes off chip.
http://www.intel.com/content/dam/ww...obile-m-h-processor-lines-vol-1-datasheet.pdf
Actually, the FIVR (fully integrated voltage regulator) they designed, is probably rated to much higher current limit. The slides, which were dated at 2010, meaning that they actually designed it 3 yrs before the CPU went on market. Integration onto the die is another issue though. The title said 400A, but real physical limit of it could be higher even.
http://www.psma.com/sites/default/f...ully-integrated-silicon-voltage-regulator.pdf
EDIT: Each power cell that's 2.8mm^2 can carry 25A. I don't know how many cells we have on the 4930MX, or if Intel disabled any power cell. -
Thankfully, even though it has to be reset at each boot, it can be set higher than 95.000A because it takes about 123.000A to run 4.3GHz, about 170.000A for 4.6GHz. They should make the default 95.000A.
Good find... this explains why the MQ processors perform as well as the 4930MX without any changes to power settings. I am going to add this information to the opening post. They should make the default Processor Current Limit 95.000A at minimum, or 125.000A if they want customers to purchase the 4930MX. 4.0GHz isn't good enough, so it needs to be more than 95.000A. -
Well more than that and Intel don't guarantee reliability. I want a min of 4GHz at full load, since any higher and I won't be able to undervolt it as much. Also, it doesn't seem like I can run 4.1GHz with 1.1V, it has to be more. With 4 GHz, I can undervolt to 1.07V. It gives me WHEA error at 1.065V. 1.075V is how I run the C2D T9500 on my 5 yrs' old Asus F9SG.
Beyond 4GHz, if you live in tropical area without AC, the CPU might overheat. It will take quite a bit of tweaking to get it stable. Hey but that's why overclocking is fun. Everyone who's in it is in for different stuff. Breaking highest benchmark is fun, but achieving stable overclock at high room temp is also fun. It's all challenge.
I think Dell's engineer didn't take into consideration the different current requirements for each SKU of CPU. Since it's one BIOS for all CPU in the AW 18 they might have gone the easiest and safest route. After all, unless Dell does what Asus does, Dell won't know the problem and variability with Haswell. Asus bought at least 600 CPU and tested them for overclocking when developing their ROG mobo. That how they came up with the tips for 4770K overclocking: set Vcore to 1.2V, set multi to 46x, and if it boots you are above 50th percentile in terms of CPU quality.
3 Step Guide to Overclock Your i7 / i5 Haswell Platform | Overclockers
Overclocking Haswell on ASUS 8-Series Motherboards - YouTube
Too bad overclocking on mobile platform is even more niche than overclocking the desktop chip. Someone needs to provide some guide.Mr. Fox likes this. -
-
Perfect Stranger Notebook Consultant
-
Meaker@Sager Company Representative
Yes but a desktop motherboard can get away with certain things a notebook board can't. In their notebooks asus "solved" the issue by soldering in lower quads lol.
I think the engineers did not lean enough from the 3940xm and got caught out by the even more aggressive requirements of the 4930mx. -
I think Dell should unlock the Processor Current Limit and set its maximum to 95A or higher for those that have the 4930MX CPU.
Instead of controlling the CPU thermals using current, give it as much current as it needs, and control it using Power Limit (TDP). Using TDC to control it is just not as intuitive as using TDP
Actually, the HQ mobile Haswell with the built in 129MB can potentially match the 4930MX, even though it has a much lower clock speed. The eDRAM is like a huge 128MB L4 cache.Mr. Fox likes this. -
For any mobile non-Extreme to be able to match an Extreme means somebody screwed up royally. Heads will be rolling at Intel if that's actually true. It doesn't mean the non-Extreme is good... it means that somebody botched up the Extreme CPU and charged extra for a CPU that can't deliver.
While the articles linked on the Asus testing are extremely informative, the settings that work well for overclocking the 4770K do not apply to the 4930MX. This 4930MX will boot with 1.25V and 46x4, but it doesn't actually run effectively that way. I set it up that way for testing to verify that my previous overclocking testing produced better results, and it does with this 4930MX.
My 4930MX won't run but 3.9GHz with 95.000A. Look at these steppings with Core Voltage at 1.25V (which is way too much for 4.0GHz) and 112.000A Core Current Limit based on my testing. Click the images to enlarge them in a new browser window... Keep an eye on "Package Current Limit" (ThrottleStop lingo for "Processor Current Limit" in XTU).
Increasing Core Voltage creates too much heat. Haswell is already cursed with too much heat. Achieving a higher overclock with less voltage and higher amps/watts will produce somewhat cooler temps. It takes a lot of trial and error with Haswell to figure out what the best combination is, but all of it is meaningless with fans that don't run often enough or fast enough to keep it cool. Beyond 4.3GHz the 4930MX starts to lose efficiency without extraordinary cooling and it doesn't bench well unless you use something exotic like my portable AC unit. Without an unlocked BIOS to access the CPU Thermal Configuration menu, extreme overclocking is going to be difficult because the default thresholds are set too low. And, 4.3GHz does not really qualify as extreme overclocking... good, yes... extreme, no.Last edited by a moderator: May 6, 2015 -
Meaker@Sager Company Representative
The cache is of no help (see anandtech) and they cant use a dedicated gpu
Mr. Fox likes this. -
Here is an interesting thing to look at... back-to-back comparison.
Ivy Bridge vs. Haswell - 3 feet apart in the same room, same ambient temps.
CPU Clock Speed: 4.3GHz GPU Clock Speeds: 975/1400 (+125/+300) Display Driver: GeForce 331.65 WHQL Operating System: Windows 7 Ultimate x64
Alienware 18 - 4930MX | GTX 780M SLI - CPU Max Temp: 105°C Alienware M18x - 3920XM | GTX 780M SLI - CPU Max Temp: 94°C Last edited by a moderator: May 6, 2015 -
How do you get the 4930MX to run at 105C?
What about the power usage of the 4930MX vs the 3920XM in that benchmark? -
woodzstack Alezka Computers , Official Clevo reseller.
-
I found out another quirk.
Switch to Intel iGPU, by pressing Fn + F5 and reboot.
Check the Package Current Limit (Processor Current Limit in XTU) in TS, one should find that it's at 95A. On the other hand, in SLI mode the Package Current Limit is locked at 55A. It seems weird that there'd be so much power diverted to the iGPU. Assuming about 1V voltage, 40A means it's close to 40W. In general (in SLI mode, with the iGPU not active), I noticed that 80% of total power goes into the CPU cores.
Both mode should have 95A there, and in both mode we want the Package Current Limit/ICCMAX/Processor Current Limit unlocked.
So the CPU throttling is related with the GPU switching. Run your CPU benchmark while using iGPU, and compared it to when SLI is on, the difference should be there.
I will PM Alienware's L_Porras about this.
Also, it seems like switching to iGPU would make my undervolt at 1.085V unstable, maybe it's because of the iGPU pulling some power there. 1.09V seems fine but I haven't bothered to stay too long in the iGPU mode. -
woodzstack Alezka Computers , Official Clevo reseller.
you saying there is a current limited placed between the gpu's and cpu's then ?
-
woodzstack Alezka Computers , Official Clevo reseller.
well, Im in a very hot house )28 Celsius) and im second place..every friggin time I bench it wants information and crap. And it will not allow me to bump Fox out of first place without verifying and waiting.
anyways the one place I didn't have to verify things was ..well, 2nd place. lolMr. Fox likes this. -
I once had 937 but that was before I upgraded to 8.1, with 8.1 I had 929 as max. Maybe I should go make an account and bump you down one place
-
woodzstack Alezka Computers , Official Clevo reseller.
-
XTU overclocking records @ HWBOT
939. Sup. -
woodzstack Alezka Computers , Official Clevo reseller.
kh90123 likes this. -
Go go break Mr. Fox's 968. I will bench again next week when I have the new CPU fan. I am contented for now with 4GHz at 939. It's called efficient overclocking hehe.
Go run some Cinebench R11.5 with your overclock and try to hit 9 points. My highest was at 8.6-8.7, it's tough to go higher unless I clock it higher, but I want to run the core voltage low @ 1.1V. I have to raise it from like 1.08V because stability was an issue when the CPU heats up.
http://hwbot.org/hardware/processor/core_i7_4930mx/ -
woodzstack Alezka Computers , Official Clevo reseller.
well, the one time I got over 1,000 I crashed..right at trying to switch to the website. Also for somereason in my profile it only shows the earlier entries, when I got 954 .
But Im only benching at 4.3Ghz since that's what this thread is about. -
No one never says you can't go for more.
-
woodzstack Alezka Computers , Official Clevo reseller.
Intel Core i7-4930MX CPU @ 5400.8MHz - 932 XTU marks on HWBOT
LOL 5.4Ghz . . .Last edited by a moderator: May 5, 2015kh90123 likes this. -
Meaker@Sager Company Representative
Yeah lol, get the liquid nitrogen pots at the ready.
-
-
woodzstack Alezka Computers , Official Clevo reseller.
well, Im 561 now . . . but with my fiancé baking, its kind of hot in the house. Maybe tonight i'' leave the door to the balconly open near the laptop lol, and get it colder..
actually, its not heat that is causing the laptop to shutdown, its some kind of autothrottling, no matter what you do, it throttles. The faster you can get it, is by finding a speed that does not throttle but is as close to its turbo value as possible..
I was getting 510-520 ratings with 4.1 and 4.2 clocks.
I hit 6.2 Ghz and still didn't get higher then fox though, because of throttling... was getting like 930-950 at best. though Had the ram clocked at 3200..lol
If you look at Fox's temps, its 34 degree's in that screenshot. the cpu can't be 34 degree's on stock cooling past 3Ghz, and its at 4.5Ghz.
(edited a stupid comment about a freezer|)|
That THEN means - I'm first place.
Wait, until I get my TEC's delivered, been 2 months been waiting... I remember hitting 7.6Ghz back in the day with my q6600 with those. Im sure I could do more, vut at this rate, I might just sell this laptop, Kind of bored with it at the moment, better to buy something new and move onto the next project.
Anyone want a perfect aw 18, lol, you let me know. -
woodzstack Alezka Computers , Official Clevo reseller.
Mr. Fox likes this. -
Plus, you finally beat me fair and square by a couple of points so you have no reason to make excuses for my score being what it is... did you use a freezer? I can't see your temps with such a tiny screen shot. Good job, by the way. It's nice to see someone else chasing numbers hard. Although, it's a real pity that benching Haswell is such an undertaking... it's a really piece of work, and not in a good way. The throttling you mention is a real pain. The average enthusiast will never see this kind of performance from Haswell because it is so inconsistent and erratic. -
woodzstack Alezka Computers , Official Clevo reseller.
You showed your setup, before, it was a rack in a freezer, or fridge or something, wasn't that you ?
But I got my ambient down to like 17 degrees, to get what I got. You have to kind of toggle the bench on and off, to get the fans going, but once their going, usually that's a good time to start benching.
It really is like you have to get the machine hot first.
Anyways, on HBOT forum, the entry lists your type of cooling, and one of the option is air conditioning or something that could be such. The listing I remembered of your did not have it. Anyways, no need to get mad bro. You had the highest place for months.
I hit Over 990, once and something like 1004 as well, but I always crashed when selecting another window with the mouse... Im not sure how Im going to address the issue, I crashed a few times getting 968 , and 970 but this was earlier in the day. I can not seem to run anything past 4.3Ghz without very bad throttling, or crashing if it does not throttle as much. As you can in one of my benches , was at 5.4Ghz or something.. it will run higher then 4.3 only if there is literally less then 1% load.
there, uploaded it with higher res, my bad. Anyways, don't mind me feeling like such a big shot beating youIts my single moment of fame haha, never been first in anything.
Mr. Fox likes this. -
I wasn't upset with you getting a higher score, but I certainly wasn't keen on the insinuation that I cheated and should be disqualified. That was ludicrous and inaccurate, and I did find it offensive. I've been using air conditioning for benching a long time and that's legit. There are others that do the same. It's simply cold air (about 60°F), not dry ice, water cooling, or anything extreme.
I don't mind you getting a higher score. In fact, I am proud of your accomplishment and wish there were more people willing to spend as much time as both of us have trying to figure out this Haswell mess, LOL. I'm seldom first at anything either, so that makes two of us. It's all for fun anyway.Optimistic Prime likes this. -
It's not just both of you guys pushing it. I am >95% of your score at 7% lower clock.
Once the fan comes I will be able to push a little bit higher. -
kh90123 likes this.
-
I was surprised to see that someone with a 4900MQ scoring 970 points.
pygo`s XTU score: 971 marks with a Core i7 4900MQ -
I saw that also. It doesn't seem right. The 4900MQ is only running 4.1GHz. I'm wondering if that's a bug with the Windows 8 RTC.
How to Overclock the Alienware 18 and Haswell CPU (or actually have it run full stock Turbo Speed)
Discussion in 'Alienware 18 and M18x' started by Mr. Fox, Oct 15, 2013.