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

Too much pressure makes me worried that it might crack the die. After the lapping and the application of Liquid Ultra, the temps are lower. There's no way to know if no lapping + Liquid Ultra might result in even better temps. I'd say no, but you know, sometimes you can't turn back around.

 

So Mr. Dufus.

Linx in Windows 8, doesn't get locked to CPU cores with proper affinity. If I spawn 4 threads, even though I set the affinity to core 1, 3, 5, 7 (or 0, 2, 4, 6) it runs on 2 cores.

So I ran the latest AVX2 Linpack from Intel instead.

Here's a screenshot of the temps. Temps going through the roof, and CPU throttling here and there. Most of the time it runs at 3GHz. Short power is capped at 90W for 3s, and long power is 77W. Anymore and it's unstable for me. Ambient is around 24C.

To lock the affinity, I read on Anandtech forum that one has to add this line in the .bat file: start /b /affinity 55 linpack_xeon64.exe



And here's the result in the log:

Spoiler

Current date/time: Tue Dec 24 10:36:47 2013

CPU frequency: 3.939 GHz
Number of CPUs: -1
Number of cores: 1
Number of threads: 4

Parameters are set to:

Number of tests: 12
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000 30000 35000 40000
Number of trials to run : 4 4 4 4 4 2 2 2 2 1 1 1
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=4210869504, at the size=40000

=================== Timing linear equation system solver ===================

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.006 106.4030 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.005 129.6417 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.005 129.2863 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.005 127.5358 1.083189e-012 3.693953e-002 pass
2000 2000 4 0.048 112.1805 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.045 117.6114 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.045 119.3650 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.045 117.5215 4.220901e-012 3.671667e-002 pass
3000 3000 4 0.146 123.1379 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.140 128.4910 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.136 132.1898 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.134 134.1336 1.016483e-011 3.914231e-002 pass
4000 4000 4 0.298 143.1109 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.313 136.3530 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.299 142.6522 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.300 142.2331 1.906425e-011 4.155234e-002 pass
5000 5000 4 0.561 148.5537 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.602 138.3997 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.557 149.7986 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.559 149.0531 2.299338e-011 3.206242e-002 pass
10000 10000 4 4.124 161.6848 9.420734e-011 3.321846e-002 pass
10000 10000 4 4.245 157.1008 9.420734e-011 3.321846e-002 pass
15000 15000 4 14.687 153.2291 2.137378e-010 3.366406e-002 pass
15000 15000 4 14.759 152.4748 2.137378e-010 3.366406e-002 pass
20000 20000 4 34.266 155.6685 3.723286e-010 3.295924e-002 pass
20000 20000 4 34.283 155.5900 3.723286e-010 3.295924e-002 pass
25000 25000 4 67.346 154.6935 5.678848e-010 3.229357e-002 pass
25000 25000 4 67.415 154.5337 5.678848e-010 3.229357e-002 pass
30000 30000 4 113.866 158.0963 7.893267e-010 3.111534e-002 pass
35000 35000 4 180.772 158.1314 1.216992e-009 3.532743e-002 pass

Not sure why it didn't update anymore after 180s. Maybe I will re-run it in the future, but for now I still have the issue with the hot core #1. I wonder if it's the washers that I added that's causing uneven pressure. When my spare tube of Liquid Ultra comes I will open it up and remove the washers and test again.

 

First of all, thanks for running the tests.

Could be a couple of reasons for this, non-linear APICID or core parking. You could check APICID with the attached.

View attachment 106600

Your 10000 results are pretty close to my 4700MQ (~160GFLOPS) when I'm running all cores at 3.4GHz. So either some throttling taking place, other than temperature I would think but than can be checked by just running a few 10000 runs and not the previous runs. Or some other program is stealing CPU resources.

After your 10000 run, results go down instead of up, probably due to temperature throttling. FWIW I run 3.4GHz at ~1.000V but anything appreciably more than 10000 will result in thermal limits being exceeded. Just the one heatpipe for the CPU on this entry level laptop so not unexpected. Ambient is around 26C, package power hits about 86W and core primary plane power about 78W, if those power figures can be believed. :/

 

My voltage is at 1.115V, so quite a bit higher than yours. For every other benchmark the CPU runs at 4GHz, so I need the extra voltage. But because adaptive mode doesn't work, when the CPU throttles the voltage stays static. If adaptive mode works properly with negative core voltage offset, then the core voltage should drop when the CPU throttles. Too bad Dell doesn't seem too interested in fixing the issue. I have pretty much lost any hope that Dell will fix the stuff. They don't even know clearly what's wrong.

That's the conundrum I have. I know the CPU will throttle under Linpack, and for better GFlops I should lower the core voltage. But running the CPU at 3.4GHz will mean that other benchmark will suffer a lower performance. I will remove the washers that I put on the backplate, and see if that will help improve temps. It's either uneven die pressure or I am unlucky to have a weak core. It runs almost 10C higher than the next hottest core.

Until Dell releases a proper BIOS, there's no point for me fiddling with the CPU.

My package power is capped at 77W.

 

It wasn't my intention for you to run at 3.4 but to get an idea of how Haswell is and your results help there. Thanks, not many takers with this bench, maybe because the results can be disappointing. Linpack is a benchmark, not real world usage so don't take the extra heat too much to heart.

Strange with adaptive voltage. Can it be properly run from XTU? Change to adaptive with offset and set static to 0.

:hi2: Merry Christmas to you and all. :hi2:

 

Adaptive voltage works, only if the core voltage is set to 1.167V and higher. For example, I could set the core voltage to default 1.2V, leave the voltage mode in adaptive, then apply a -100mV voltage offset, and the lowest it will go is 1.167V. If I use adaptive mode, and run the CPU at 1.167V, I get worse results in benchmark, just due to the increased heat from the higher core voltage.

Since I only need 1.115V (or less) to run the CPU at 4GHz, I can't use adaptive mode. I have seen users of Asus mobo having no problem undervolting further than that, so I am assuming that this is another limitation with the BIOS on the AW 18. Shame that we can't flash unlocked BIOS onto it. I am curious to see if there's any other options that I can tweak.

I have had problems with hot core #1, I suspect it could be related to the washer mod that I did some time ago. Then again, it might just be the CPU.

This is a situation where I have tried rather hard to improve the cooling and reduce the heat, but there's just so much limitation and lack of support from Dell's side that it's disappointing. Do you know, those lazy engineers just set the Package Current Limit to 55A for 4900MQ and 4930MX CPU. In its stock mode, if one puts any load on the CPU and it will go down to 3GHz, and below that even.

 

I have used adaptive down to 1.03v in another brand. Odd.

 

In your Sager eh?

Well I have suspected a long time ago that it's due to the BIOS. Somehow it just feels like they just hurriedly put the BIOS together without too much testing. Or maybe they found out too late that Haswell runs hotter and slower than expected.

EDIT:

I tested the adaptive voltage again, and now it seems with the A03 BIOS negative offset will work. It's just that the negative offset starts applying from 1.167V.

To make it clear, like my setting right now:

Spoiler

Note that even though I set the core voltage to 1.115V, with adaptive mode, it doesn't go to 1.115V. Instead the lowest it will go is 1.167V. To make the core voltage go to 1.115V, I have to apply a -50mV Dynamic CPU Voltage Offset.

Additional note:
The adaptive mode has some wonky behavior. If I run XTU benchmark, even with the negative CPU voltage offset, it tends to go back to 1.167V when the load sets in. In short, the performance is worse with adaptive mode than with static voltage...

I think I need to take a break and go play some games. So back to static voltage for now.

 

I have seen as much as 130W draw from the 4930MX and it doesn't seem to be locked. The only problems I am finding with that is uncontrollable temps and Haswell's signature throttling issues. I wonder if your enabling of Secure Boot does something with the CPU power settings that we are not identifying. I have noticed the 4930MX performs better under Windows 7 than it does Windows 8/8.1. It does with Ivy Bridge also, but for some reason the Windows 8/8.1 performance hit seems worse to me with Haswell.

I agree with Dufus... you should not put too much stock in Linpack, XTU or prime95 benchmarks. They do not provide any meaningful results relating to real-world performance. These are the CPU equivalent of hardware abuse that Furmark represents for GPU abuse. It's sort of like testing the engine in your car by throwing a cinder block on the accelerator pedal and saying everything is good if it doesn't explode. If you adjust your system to handle these CPU benchmarks without overheating then your performance results in other benchmarks will be poor. I would recommend using 3DMark11 and Vantage physics testing if you want something that more closely resembles real-world CPU performance measurements. A Haswell CPU will get a higher Gigaflop than Sandy or Ivy Bridge because it throttles methodically instead of shutting down, but it gets raped by Ivy Bridge when it comes to real-world performance testing.

 

I meant I capped it at 77W to prevent from thermal shutdown when running the AVX2 Linpack. That benchmark will make the CPU overheat. It's like temps you get from wPrime/Cinebench +10C to +15C.

I have also seen 130W power draw in HWiNFO. Not sure why. The power estimation is probably an estimation. It's perhaps based on digital sensors rather than analog sensors. It's like an estimated power, or so I read.

I have disabled SecureBoot since Microsoft patched the watermark away.

 

The 4930MX definitely draws 130W+ under extreme load at 4.5-4.6GHz if you give it enough power to not throttle from power starvation. You can verify that with a Kill-a-Watt meter. That's one reason why 780M SLI and 4930MX can't function well with a wimpy 330W AC adapter. Ivy Bridge XM can barely hit 100W overclocked to 4.8GHz.

I much prefer having a processor that quickly takes the system to a thermal shutdown instead of throttling. When that happens you know that you have to find a way to keep it cooler or slow it down. When it throttles like Haswell, you end up with a CPU that has a mind of its own, refuses follow its commander's instructions, and puts the entire system at risk for achieving poor results. Haswell deserves to receive a dishonorable discharge for insubordination. It needs to be replaced by a CPU that does exactly what its told to do, without wavering or rationalization.

 

Hello Mr. Dufus.

I finally decided to go ahead and take out the washers that I put in last time. Had to repaste the liquid metal TIM. That's a pain to get off the copper. Basically, everytime repasting you have to lap the copper. But I guess it won't really affect the thermal performance as long as the same TIM is used.

I saw that you ran your 4700MQ at 1V in AVX2 Linpack, and it runs at 3.4GHz. So I decided to go down a notch and lock the multiplier at 39x, which enables me to run the CPU at 1.06V, instead of 1.115V for 4.0GHz. Some results here:

85W long power limit; 1.06V core voltage; Observed power ~86W. Observed multiplier in Linpack ~ 35x.

Spoiler

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Thu Dec 26 17:30:55 2013

CPU frequency: 3.890 GHz
Number of CPUs: -1
Number of cores: 1
Number of threads: 4

Parameters are set to:

Number of tests: 9
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Number of trials to run : 4 4 4 4 4 2 2 2 2
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=705536800, at the size=25000

=================== Timing linear equation system solver ===================

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.011 62.2142 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 113.6096 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.007 90.0060 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 108.8817 1.083189e-012 3.693953e-002 pass
2000 2000 4 0.050 107.7279 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.048 111.9556 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.047 114.1562 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.055 97.1062 4.220901e-012 3.671667e-002 pass
3000 3000 4 0.145 124.2460 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.142 126.9220 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.145 124.6248 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.144 124.9706 1.016483e-011 3.914231e-002 pass
4000 4000 4 0.315 135.5143 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.313 136.4687 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.318 134.2748 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.317 134.8370 1.906425e-011 4.155234e-002 pass
5000 5000 4 0.593 140.5745 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.594 140.4588 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.586 142.3109 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.600 139.0268 2.299338e-011 3.206242e-002 pass
10000 10000 4 4.115 162.0452 9.420734e-011 3.321846e-002 pass
10000 10000 4 4.151 160.6372 9.420734e-011 3.321846e-002 pass
15000 15000 4 13.295 169.2748 2.137378e-010 3.366406e-002 pass
15000 15000 4 13.271 169.5702 2.137378e-010 3.366406e-002 pass
20000 20000 4 30.427 175.3082 3.723286e-010 3.295924e-002 pass
20000 20000 4 30.350 175.7516 3.723286e-010 3.295924e-002 pass
25000 25000 4 59.333 175.5843 5.678848e-010 3.229357e-002 pass
25000 25000 4 59.459 175.2122 5.678848e-010 3.229357e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 93.6779 113.6096
2000 2000 4 107.7365 114.1562
3000 3000 4 125.1908 126.9220
4000 4000 4 135.2737 136.4687
5000 5000 4 140.5928 142.3109
10000 10000 4 161.3412 162.0452
15000 15000 4 169.4225 169.5702
20000 20000 4 175.5299 175.7516
25000 25000 4 175.3983 175.5843

Residual checks PASSED

End of tests

85W long power limit; 1.01V core voltage; Locked multiplier at 35x; Observed power ~80W. Observed multiplier in Linpack ~ 35x.

Spoiler

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Thu Dec 26 17:37:45 2013

CPU frequency: 3.490 GHz
Number of CPUs: -1
Number of cores: 1
Number of threads: 4

Parameters are set to:

Number of tests: 9
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Number of trials to run : 4 4 4 4 4 2 2 2 2
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=705536800, at the size=25000

=================== Timing linear equation system solver ===================

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.006 106.1590 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 115.1227 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 116.5691 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 116.0956 1.083189e-012 3.693953e-002 pass
2000 2000 4 0.048 111.2845 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.047 114.1266 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.048 112.3607 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.048 111.0342 4.220901e-012 3.671667e-002 pass
3000 3000 4 0.147 122.6452 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.149 121.1677 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.149 121.2757 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.149 120.8900 1.016483e-011 3.914231e-002 pass
4000 4000 4 0.319 133.9843 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.312 136.8215 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.312 136.8505 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.312 136.7054 1.906425e-011 4.155234e-002 pass
5000 5000 4 0.588 141.7851 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.582 143.2959 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.585 142.5036 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.583 142.9912 2.299338e-011 3.206242e-002 pass
10000 10000 4 4.109 162.3024 9.420734e-011 3.321846e-002 pass
10000 10000 4 4.098 162.7138 9.420734e-011 3.321846e-002 pass
15000 15000 4 13.215 170.2923 2.137378e-010 3.366406e-002 pass
15000 15000 4 13.259 169.7317 2.137378e-010 3.366406e-002 pass
20000 20000 4 30.107 177.1711 3.723286e-010 3.295924e-002 pass
20000 20000 4 30.084 177.3055 3.723286e-010 3.295924e-002 pass
25000 25000 4 58.946 176.7361 5.678848e-010 3.229357e-002 pass
25000 25000 4 58.725 177.4020 5.678848e-010 3.229357e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 113.4866 116.5691
2000 2000 4 112.2015 114.1266
3000 3000 4 121.4947 122.6452
4000 4000 4 136.0904 136.8505
5000 5000 4 142.6439 143.2959
10000 10000 4 162.5081 162.7138
15000 15000 4 170.0120 170.2923
20000 20000 4 177.2383 177.3055
25000 25000 4 177.0690 177.4020

Residual checks PASSED

End of tests

I wasn't that happy with the drop of performance in Cinebench, so I went back and increase the long power limit to 87W. This run core 1 hits 100C, but there was no thermal shutdown.
87W long power limit; 1.06V core voltage; Locked multiplier at 39x; Observed power ~86W. Observed multiplier in Linpack ~ 35x.

Spoiler

Intel(R) Optimized LINPACK Benchmark data

Current date/time: Thu Dec 26 17:53:47 2013

CPU frequency: 3.889 GHz
Number of CPUs: -1
Number of cores: 1
Number of threads: 4

Parameters are set to:

Number of tests: 9
Number of equations to solve (problem size) : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Leading dimension of array : 1000 2000 3000 4000 5000 10000 15000 20000 25000
Number of trials to run : 4 4 4 4 4 2 2 2 2
Data alignment value (in Kbytes) : 4 4 4 4 4 4 4 4 4

Maximum memory requested that can be used=705536800, at the size=25000

=================== Timing linear equation system solver ===================

Size LDA Align. Time(s) GFlops Residual Residual(norm) Check
1000 1000 4 0.012 55.9287 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 106.4210 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 111.9178 1.083189e-012 3.693953e-002 pass
1000 1000 4 0.006 109.1584 1.083189e-012 3.693953e-002 pass
2000 2000 4 0.050 105.8239 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.047 114.6049 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.047 113.1357 4.220901e-012 3.671667e-002 pass
2000 2000 4 0.059 90.8991 4.220901e-012 3.671667e-002 pass
3000 3000 4 0.147 122.3610 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.149 120.8920 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.147 122.7019 1.016483e-011 3.914231e-002 pass
3000 3000 4 0.143 126.3004 1.016483e-011 3.914231e-002 pass
4000 4000 4 0.317 134.5583 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.317 134.7931 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.317 134.5030 1.906425e-011 4.155234e-002 pass
4000 4000 4 0.313 136.3514 1.906425e-011 4.155234e-002 pass
5000 5000 4 0.596 139.9921 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.582 143.3454 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.586 142.3104 2.299338e-011 3.206242e-002 pass
5000 5000 4 0.584 142.8865 2.299338e-011 3.206242e-002 pass
10000 10000 4 4.121 161.8108 9.420734e-011 3.321846e-002 pass
10000 10000 4 4.591 145.2547 9.420734e-011 3.321846e-002 pass
15000 15000 4 13.276 169.5167 2.137378e-010 3.366406e-002 pass
15000 15000 4 13.260 169.7148 2.137378e-010 3.366406e-002 pass
20000 20000 4 30.381 175.5760 3.723286e-010 3.295924e-002 pass
20000 20000 4 30.131 177.0316 3.723286e-010 3.295924e-002 pass
25000 25000 4 59.498 175.0960 5.678848e-010 3.229357e-002 pass
25000 25000 4 59.132 176.1802 5.678848e-010 3.229357e-002 pass

Performance Summary (GFlops)

Size LDA Align. Average Maximal
1000 1000 4 95.8565 111.9178
2000 2000 4 106.1159 114.6049
3000 3000 4 123.0638 126.3004
4000 4000 4 135.0515 136.3514
5000 5000 4 142.1336 143.3454
10000 10000 4 153.5327 161.8108
15000 15000 4 169.6157 169.7148
20000 20000 4 176.3038 177.0316
25000 25000 4 175.6381 176.1802

Residual checks PASSED

End of tests

At the end of the day, undervolting the CPU benefits a lot in Linpack. Each 10mV drop will drop the power by 2-5W. Now the best of both worlds would be to have adaptive voltage working on the AW 18, but with offset that I want. The max computation power from mobile Haswell, is probably about 180 GFlops, unless one uses some exotic cooling.

It's perhaps only in AVX2 Linpack, that Haswell 4930MX will beat Ivybridge/Sandybridge XM processor conclusively. I am curious to see if 3920XM/3940XM will even come close to the 177 GFlops performance that I got. Other than that, due to heat and BIOS issues, Ivybridge mobile CPUs are still faster in most other benchmarks that don't use AVX2 code. Using AVX Linpack, I consistently get about 98 GFlops. I could tweak it for more but heat is overwhelming. A 3770K would get ~105 GFlops, not sure about how the mobile XM CPUs would fare, but I am pretty sure it's still faster than 4930MX.

I have mentioned time and time again, the CPU cooling on the AW is far from adequate. There are inches of wasted space, which could have been where the extra length of heatsink goes.

EDIT: Someone at Puget System did some testing: Haswell Floating Point Performance - Puget Custom Computers.

4770K got 177GFlops running at 3.5GHz, the same as what I got. 3570K got 105GFlops running at 3.4GHz. Hyperthreading is not a factor here as Linpack should only have # of threads running = # of cores. It seriously uses up any resources available.

I bet Intel is betting on AVX2 pickup. But tech industry is usually a bit slow on the pickup.

Hopefully by the time the next gen CPU comes around, the heat issues would have been resolved. We want no more sucky overheating CPUs.

 

My 3920XM gets around 100 GFlops. It does not matter what I set the multipliers to, the core ratios end up running around 3.5-3.7GHz. CPU core overclocking seems to be disabled somehow when running those tests.

If the only benchmarks Haswell can win at are GFlops tests, I would say "Gee, what a flop" . The Futuremark and Unigine titles, and gaming benchmarks, are the ones that are truly important for the things I care about most. Maybe if productivity applications and computational tasks are what matters for a particular niche, the GFlop performance would be more relevant.

 

What sort of current values are you setting? 130-140 amps?

 

If the other benchmarks is coded in AVX2, then Haswell will have an advantage. Just based on the information on the architecture, it should be superior. The cache will be tons faster, but it's not. Only L1 is faster, L2 is not much faster, L3 is slower because the ring cache is uncoupled in terms of freq. There are some bottlenecks in there. Maybe I should get a job with Intel and find out what went wrong.

Anyway we will get to know more stuff when Broadwell comes out. I believe that on the mobile CPU the IVR will still be integrated, and in some low power CPU the PCH will be integrated too. CPU is a vacuum cleaner/blackhole, it will suck up anything and everything that it can, it's all heading the way of SoC (system on chip), like I said sometime ago.

120A. I have heat issues with the power before the CPU runs out of juice.

Something funky that I found out with the 4930MX is that, when you drop the multiplier from 40x to 39x, I went from like 925 in Intel XTU benchmark to 850, and clock speed would waver around in XTU benchmark. On the other hand, I get roughly the same performance in wPrime/Cinebench/Linpack (for eg about 6s for wPrime 32M). Not sure why clock speed would drop in XTU benchmark, there's no throttling, no overheat, nothing. I set it to adaptive voltage, and the results are worse than with static voltage.

Hopefully after Christmas when people go back to work, there will be a BIOS update. Maybe it's coming soon.

 

Wow, I wouldn't have expected such a big change in voltage for that frequency with that processor.

Sounds like the EC may be controlling things instead of the CPU. I have no trouble with dynamic offset mode (adaptive) dropping vcore across the board. At the LFM (8x) core voltage is under 0.6V, at least that's what the CPU reports.

Except for a few niche softwares it's probably going to take a few years.

Well unless there is a change in fabrication materials I'm thinking it's just going to get worse for every shrink, at least that is for running the higher clocks. Maybe instead we'll see an increase in the number of cores.

 

I agree with you. Desktop CPUs will become more like the server CPUs, with more cores, wider execution path, more cache, and runs at a lower clock speed. Perhaps we will see something like a Haswell-E 8 cores, which has like 12MB or more L3 cache, but the cores run at less than 3.5GHz. That will be the way to keep power down for next gen CPU.

In the end we might start to see ARM cores in CPU beside the x86 big cores. In as soon as next gen GPU I expect to have some ARM trickery in there.

 

I found the xtu test was quite sensitive to the current limit.

I hope that the arm core is able to do the 2d work and use system ram so the shaders and gddr5 can be shut off like optimus but we can drop the intel chip outputs like a sack of stones.

 

Thanks for running these tests. This will help when my machine comes in.
I'm a little disappointed that this has gone on since the laptop's release.

 

I have a 4900MQ.

After 30 days of fiddling around I came to understand that there is no way to sustain the default OC Level 1 - 4Ghz.

There are no amount of settings that I can play with that makes this thing go at the advertised specs.

Am I crazy?

 

Try setting core and processor cache voltages to about 1.1V and the core/processor cache voltage modes to Static. Also allow plenty of power and current limits and that should stop you from ever dipping below 4.0GHz. Show us a screenshot of your XTU settings if things don't work right the first time.

 

Ok, first of all, my AW18 is bricked. It just didn't boot anymore for some reason and because you can't really reset the bios unless you take apart the computer and use another CPU...

They are sending me a new one (after Haggling a bit).

So, I was talking to the tech guy and he says that they are not working on a fix because they can not reproduce the problems and I shouldn't believe everything that's written on the net (talking about our messed up BIOS).

You know what? That might be true since months have passed and nothing changed... I don't know what makes us believe Alienware will come out with a "fix".

Anyways, I just needed to vent since I will have to wait 3-4 weeks for the build.

 

I've bricked mine once that way too, so have a few other people around here. I believe that after doing some OC in XTU, enabling Extreme Edition in the BIOS will result in a brick like what you described. A motherboard replacement can solve the problem but there has been no record of recovering from this issue otherwise. The tech guy you spoke to may or may not know what he's talking about although I have a bad feeling that he might be right about there being no fix on the horizon.

 

Yeah, this is crazy, it take 5 minutes to make the case that there is an issue with the processor current limit and the fans table.

Any computer hardware engineer will acknowledge the issues.

Also, even if you tweak the settings and make your AW run at specs, the fan table is so bad that sustaining turbo is almost impossible.

 

That sir, is where you're incorrect. Only very few have recovered/reset the BIOS. I am among one of them. After 2 mainboard replacement, I finally figured out how to reset the BIOS.

Refer here: http://forum.notebookreview.com/ali...-if-one-runs-into-no-post-messed-up-bios.html

So basically the easy way to do it, would be to put in a 4700MQ in place of the 4800MQ/4900MQ/4930MX that's in there. It will beep a few times if I am not mistaken. There's a post over at T|I which you can look up, where an user explained his experience.

Second method which I found out, would be to induce a BIOS error. To do that, you have to remove the CMOS battery, take out the RAM, and boot it up. It will beep 2x, following the beep code here: My Alienware Computer is not showing image and is beeping (No POST Beep Codes Table) | Dell US

Wait 1-2 mins, then the machine will reboot. If it reboots and start beeping 5x, then you're done. Put the RAM and CMOS battery back, and boot it up. Sometimes it will still beep 5x even after you connect the CMOS battery back. In this case, you have to hold down Fn and press the power button, refer this: [Solved] 5 beeps on startup after tweaking some options on Unlocked A11 BIOS

Most of the time, it won't induce the error. So rinse and repeat, power on the machine without the RAM and CMOS batt connected, wait 1-2 mins till it restart. If you get 5 beeps you're good. If not repeat.

Anyway, I believe removing the CPU while booting up, will do the same thing. You will get 7 beeps (CPU error) instead of 2 beeps (RAM error). The point is to induce CMOS error so it will reset the BIOS.

After the BIOS is reset, first thing you want to do after the machine POST, is to go into BIOS by pressing F2 and turn off the CPU overclocking option. Either that or press F9 to reset the BIOS to its default state.

The only thing that I am pretty sure will brick the machine, is when someone mess with the cache voltage, specifically in my case, applying a negative cache offset will brick it upon next reboot.

 

Thank you kh90123 for potentially saving gqman69's behind. Am I right though in assuming there is still no safe way to tweak the BIOS OC settings alongside XTU? I guess we have at least a bit more reassurance while experimenting than I initially though..

 

Sometimes a simple ram change can do it too, introduce a new stick of different ram and it forces a reset.

 

The ridiculous part is that my 4900 is not even able to run stock without some tweaking because of:

1- the cpu fan which kicks in too late and the system is unable to sustain max turbo multipliers.

2- the obligatory bump of the 55A --> 85A which means that you need some software or service to set the 0x601 msr register properly.

This means that even if someone forgets about the provided (+$) OC you will NOT get what you paid for.

-

Now my AW18 is kaput, everything failed, no beeps, nothing and I tried everything.

Not to worry Dell is replacing it with a new one (3-4 weeks?).

-

However, why can't a simple Dell engineer give a definitive answer, say something like: "yes, problem is there and a fix is in the works". Why do we need all this drama?

 

I don't think I've seen and don't know why Dell can't at least acknowledge the problem and state whether they plan to fix this when every reputable tech website has complained about the same issue...

 

so my 4930XM auto boost to 3.8xxGhz ..thats normal right? Thats just running stock and benching 3Dmark11 , got a little over P12,XXX - Dual 780m GPu -

 

Out of curiosity, what settings did you change for it to brick?

 

Messing with XTU settings will have chances to cause the machine to brick. For me, I think it's setting a negative cache voltage offset.

Also, it bricked when I had to disable Samsung RAPID. Weirdly, now even if I disable it doesn't brick anymore.

 

Actually i didn't touch a settings since a couple days befor3 it bricked. I came back from gym and it was bricked.

The last setting i touched was the cpu voltage which i put at 1.115v.

 

So it seems the BIOS has a few bugs more.

I wouldn't mind paying the extra 10c it would cost to have a jumper in an accessible place (under an easy to remove existing bottom cover) for CMOS reset. Don't know why they don't do that or maybe even go that extra step and have a socketed BIOS chip. You'd think it would save a fair bit of money in RMA'd bricked BIOS's.

kh90123, FWIW on the GE60 I run cache with a -100mV offset with no ill effects. No longer have a need for XTU, it was useful for initially setting things up though.

 

Sadly, this reverts to a basic business tenet: If you don't admit to a problem.... there is no problem.

 

Don't get me wrong. It played games beautifully. And i am By no means an Overclocker nor was i planning on it, That's why I hesitated so much on sending it back. The performance was more than enough for me. The reason I decided to send it back was due to the small issues I had with it in the week I had it, from my experience if I am having problems (whether they be small or not) from the get go chances are they probably won't stop there. I was thinking if something major had gone wrong after my return grace period I would have been sending it away anyways for repair or waiting around to have some joe come to my house to rip it apart. I had the warranty to cover it but I wasn't prepared to be going through that after the first week. Plus reading other people's reviews once they started having problems they usually had a few more once it was fixed and I don't want that. The advice here was helpful too. I also live in Canada not the UK so it's not so bad. I was given around $450 in discounts to buy it without spending an overly large amount of time haggling. The guy I told the guy who accepted the return I just was scared off by the immediate issues I had and said I may look into the second revision if I hear good stuff about it so he told me to give him a call if I do decide to re purchase and we can work something out. Again I love the way this monster looks I think it is way nicer than the previous version in terms of looks. Much cleaner IMO. If I never had the small problems I had from the start I would still have the thing. Anyways who knows I might end up with an R2 if the price is right or I might just buy the aurora. I could get a maxed out version for the price I paid. Anyways, thanks again guys hope to stay in touch.

Cheers

 

Quoting this customer in this thread "for the record" since it applies here.

 

http://forum.notebookreview.com/ali...ng-potential-overclockers-alienware-17-a.html

 

This thread needs to be bumped because new threads are cropping up that revolve around the same underlying problems.

I am changing the thread title to something broader than 4930MX since it is only attracting owners with that CPU.

Here is the new thread title: How to Overclock the Alienware 18 and Haswell CPUs with a Messed Up Stock BIOS

 

Wow my i7-4900 can't even run at stock speed, I must be a huge fool to think I can over clock it. 3.6ghz top speed, what a RIP, I lost 400 mhz, EVEN ON SINGLE CORE/THREAD Apps

 

My Samsung Magician with EVO 1TB SSD was freezing my AW18 (2-4) times per day, randomly, sometimes with nothing running in task manager. Mouse or keyboard became inoperative requiring a hard shutdown. This happened with and without RAPID Mode enabled with OS Optimization set to (Maximum Performance). I switched the OS to (Maximum Reliability) and still had the same weird problems with no increased reliability. I updated the BIOS to AO3 but that didn't solve anything. Many others where having this same problem and most gave up.

I was about ready to throw in the towel and get the Samsung Pro SSD at the much higher cost. After fiddling around for a few months and taking a few hints from others I selected the Advanced tab of OS Optimization and did a custom setup and now the EVO 1TB performs fast and without any more frustrating freeze ups. Seq Read=835 & Seq Write 762 using Magician Performance Benchmark. Here are the settings that work for me under Advanced tab. Hibernation eactivate, Virtual Memory 7670 MB:Shortcut, Indexing Service eactivate, Prefetch:Activate, Write Cache Buffer; Deactivate, Power Options:Shortcut, System Restore OC(C -On PBR ImageOff. (All with Rapid Mode enabled) using Bios A03. I have not yet overclocked my 4930 MX, but at least I do have a deep sign of relief that the EVO SSD with Rapid works great now and runs super reliable.

 

A friendly reminder that we need to let Alienware know how much we would appreciate having a fix for the BIOS and EC fan table issues. Please take action using the following links. Your voice is more likely to be heard in a Dell Community Forum or Dell IdeaStorm thread. Thanks in advance for taking action, especially if you have not already done so. If you promote something at IdeaStorm, please also include comments about why you feel the idea needs to be implemented.

Alienware 18 CPU Throttling | Dell Community Alienware Club
^^^please share your story here and ask when a fix will be released for the BIOS and fan tables

Alienware 18, 4930MX Overclocked up to 4.3GHz only reaching to 3.9 GHz | Dell Community Alienware Club
^^^please share your story here and ask when a fix will be released for the BIOS and fan tables in this thread as well

Dell IdeaStorm | Alienware: Enhanced / Customizable Fan Control
^^^this has 67 votes, no action has been taken to fix the problem, but it got archived for some reason, so please vote and comment this up again

Dell IdeaStorm | bigger power supply or dual psu for high end 18s please (18x-18
^^^without this, there is no point in buying one with top specs unless you are merely a gamer that is happy with stock performance

Dell IdeaStorm | Unlock the Bios on the new Alienware 18s ~and~ IdeaStorm Petition: Unlocked BIOS for the 4930MX
^^^this is "Enthusiast 101" - they need to either remove the restrictions so we can unlock and flash it for ourselves, or provide it to us already unlocked

Dell IdeaStorm | Desktop CPU's in AW M18
^^^the "Throttle King" (4930MX) ain't good enough for a true enthusiast - they need to think about moving in this direction because this is what many of us want in an Alienware 18
​

---

The opening post in this thread had been updated with the above links for easy reference.

 

In my last thread, to answer all the questions:
-The XTU benchmark just won't run. It would just say error and that I can't run a benchmark for some reason.
-Yes I have rebooted.
-I am running on my nvidia cards. Nvidia control panel's showing them and when disconnecting the AC adapter, I get the message to switch to the integrated cards.

 

Can you post the XTU error message? That might be helpful identifying what problem is.

 

Its in my old thread. In the screenshot that I uploaded

Here's what is says:
"Error Log
The benchmarking has failed and no score was generated as a result."
That's all it says.


Edit: For some reason, ANY setting I change requires a reboot now. Can't even change processor Current Limit...

 

If you have HWiNFO64 and/or ThrottleStop running it could be what is forcing a reboot. Disable them from starting with Windows if they currently do. Reboot and launch only XTU. You should be able to change Base Clock, Multipliers, Core Voltage, Core Current Limit, Processor Current Limit, Turbo Power and Time Window limits without rebooting.

 

I only have intel XTU installed. It happened after I bricked my mobo and reset bios earlier today. Was playin around, all of a sudden CPU stopped holding anything past 3.6Ghz, went to bios to change some settings, and then it bricked. Reset, now I'm stuck with that issue.

Edit: I reset my 18 today from a brick, and now all settings in xtu stick... Weird how they only work if bios oc is disabled.

Edit 2: And again XTU is requiring a reboot for changing ANYTHING. ARG what the hell is wrong with this contraption!?

 

I'm searching this thread, how do you reset XTU to stock settings,I played around with mine but honestly the performance seems the same.. in fact throttling is kicking in and on my i7-4900 I see 4 cores drop to 2.8 ghz during the heaven engine demo.

 

Also, how good is the stock ram, the DDR3L thats included, junk for overclocking?

 

DDR3L is required on the Haswell platform. The stock RAM is fine. You can buy faster memory, but the difference is almost indiscernible. You can measure it, but it doesn't do much to enhance performance in way that you can "feel" an improvement.