Cloudfire`s road to tweaking the 4940MX (underclocking and undervolting)

Warning:
I had to do a tricky BIOS reset most likely due to too low voltage for the CPU to boot up. Be very careful using the software. I will try again later but with different approach.


So I got a new Alienware 18 with 4940MX. Was not happy with the noise from this thing despite having a 100W heatsink (even when running a simple Windows Upgrade), so I thought I could try to tweak it.
Thankfully Intel have made a tool just for that: Intel XTU

Loaded up the program and found out why my CPU fan was noisy.
Turns out Dell have overclocked 4940MX further from the Intel specs from 4000MHz up to 4400MHz which is a lot for pretty much anyone if you are not doing benchmark runs. 3/4 core have been overclocked+ 300MHz (38x to 41x). They also increased the power this CPU is allowed to draw naturally. Core voltage have also been increased +80mV which will further increase heat. Im not sure if that voltage increase was needed but thats how they did it.



Another thing I noticed is that they have done some strange things to the CPU. The processor cache have been overclocked from 800MHz (8x) to 4000MHz (40x). You see that Intel use 800MHz (8x) with a stock 4940MX. With a simple +400MHz overclock for the CPU, suddenly a 40x was needed for the cache? No way.
I also noticed that they underclocked the IGP from 1050MHz to 800MHz. My theory is that the CPU will start throttling due to power and thermals if the IGP was to run at 1050MHz. So they got some headroom by removing some of the IGPs performance (very important to notice for those who use the IGP)




Here is my first CPU stress test with Intel XTU using Dell`s specs for 4940MX (the above)
CPU max out at 91C and the fan was seriously loud running at full speed. Not very shocking considering the CPU runs at 4.0GHz and the voltage is so high.



I did not like the temps nor the noise from that benchmark. Ugh. Time to tweak it.
So I thought I could downclock the CPU down to 4800MQ but keep the 57W TDP. Should be sufficient to run 970M SLI
I also downvolted the CPU to 4940MX voltage because the 1.280V was too high.


I fixed the cache clock and the IGP speeds as well


I ran Intel XTU Stress Test again to verify it working. Temps fell from 91C Max to 81C which was very nice. Plus the fan ran at a much slower speed


I also started experimenting with voltage further. I`ve read about undervolting the CPU could do wonders on your CPU temps, so I started with -70mV. Ran HWinfo and Intel XTU Stress test again and no throttling and the CPU ran 3.5GHz (35x) on all 4 cores without any throttling.
Finally undervolted it to -100mV which is said to be ideal
Here is my final settings


Here is HWInfo results from Intel XTU Stress test. 72C max, down about 10C from before I undervolted it. The CPU runs 20C lower than what it originally did before I started playing with clocks and voltage. The fan runs on a much quieter pace and isnt noisy, and without much load it rarely runs at all. You may also notice that the "CPU Package Power" went down from 72W to 37W lol


So yeah, Im gonna keep these settings unless I notice any troubles during gaming. Don`t think I ever will because there is no way a game will utilize the CPU any more than that Stress Test which ran all 4 cores on 100%. Plus the core clocks should be sufficient for 970M SLI. The games are rarely CPU bound anyway.
I highly recommend going this way for you folks as well if you want a quiet and cool notebook. You can also create different profiles, so you can switch to higher clocks easily if you plan to benchmark

Right now the CPU runs at 50C and the fan almost never run.

 

What notebook did you buy?

 

Alienware 18
OP updated

 

You did? Did your old 18 die? I'm curious because I've seen you posting about your 970M's and 4940XM and I wanted to know which laptop offered this kind of config. Anyways, sorry for derailing the thread.

 

It's nice that you have temps and noise under control but if you were going to run your MX in a straitjacket, why not buy a 4810MQ (or, heaven forbid, 4710..) in the first place.. I'm guessing you got a nice insider deal and didn't have much of a choice w.r.t. the specs?

 

@sponge_gto: Yes, got a very good deal on this machine, plus I made an overclocking profile for the 4940MX if I want to do benchmark runs where total score matters. Plus if I find out that 970M SLI needs more power, I got plenty of it ready to be unleashed
I`m more drawn toward a balanced machine regarding specs. So this works perfectly for me

@1nstance: Alienware 18 is discontinued. I upgraded this machine myself

 

I had to do a BIOS reset yesterday because my notebook refused to boot up. Nothing would show up, the display was dark.
Most likely because the CPU got too little voltage to boot up.
Information to fix it here: http://forum.notebookreview.com/alienware-18-m18x/769152-help-alienware-18-wont-boot.html

Im pretty sure the first step I did was perfectly ok (changing the multipliers to 4800MQ and fixing the cache and changing to correct specs for the IGP) because I had to reboot when I did those settings. And the notebook rebooted just ok.
However, the undervolting changes (first -70mV and then -100mV) required no reboot so I couldnt verify if that was enough voltage at that time.
But it crashed when I switched from IGP to my GPUs while -100mV was active, so that was most likely the reason.

I noticed that using "Dynamic CPU Voltage Offset" means that the CPU will be downvolted on all scenarios: Running low clocks or high load with all 4 cores.
Notice that the 800MHz clock was running at minimum 0.660V without undervolting.
When -100mV was active, when it ran at 800MHz the voltage was at minimum 0.569V. This was probably the cause of no boot.

So what I will try next time is the first step in OP (changing multipliers, IGP and cache). Then reduce "Core Voltage" to 1.1750V because that means the voltage will only be reduced under high load.
Which in worse case scenario means it will only crash after successfully booting up and running software in high clocks which just means a simple reboot without any need to reset BIOS again

Anyone who have some thoughts about this?

 

HELP

Does anyone know how to lower the maximum voltage the CPU can get?
I tried lowering the "Core Voltage" down from 1,2V to 1.020V (I know it only needs 0.990V to run 3.5GHz) but the CPU still draw 1.18V during XTU Stress Test.
I know Dynamic Voltage Offset works and doing undervolt there, but that will affect all speeds, which will again means it get too little juice during bootup.
Choosing "Static" instead of "Adaptive" for Core Voltage preset helps to cap it at 1.020V but problem is that the lowest CPU speed, ie 800MHz, also get 1.020V too. Making this a useless tweak,

 

Very little interest on this stuff huh?

Anyhow, I think Ive found my answer from Asus forum. So it means my 1.020V is valid, just not under synthetic 100% Stress tests

 

This is seriously starting to me off.

First I made a profile with 1.020V and 37x,36x, 35x and 35x. Cache multiplier was set at 37x.
Worked fine with IGP. Switched to 970M SLI and the notebook start behaving weird. Reboot cycle, black screen etc.
Suddenly XTU is back to stock settings again and everything I did was reverted. Fine maybe it was the voltage I thought to myself.
Changed the multipliers again to 37/36/35/35. Voltage this time was 1.175V instead of stock which is 1.200V (for 40x/39x/38x/38x). Should be way more voltage than it ever needs.
Again, ton of weird issues like above when shutting off the machine. Waited maybe 3 minutes, machine got back again and rebooted successfully. XTU came with a warning that Watchdog found issues with my settings and reverted them back to stock again.

So it basically means that this 4940MX which cost about $1000 is useless even for a simple underclocking.
Is this Dell`s doing, ie their wonderful BIOS?

 

The dumb way that I've adopted is to simply use static voltage. My CPU permanently sits at 1.1V 40x, no funny business allowed. When I want the CPU to take the backseat while transferring data overnight, I just put power options to power saver.

 

Right, even 1.1V should have been enough for 35x. My own measurements using XTU Stress Test showed that it never went over 1.0V.
Good idea about using the power saver feature on Windows 8 btw.
I had enough of this XTU software for a while. Its reverted back to 4940MX default specs and uninstalled. Still didnt stop me from having a random crash
I thought AW18 played nice with XTU, because I`m certain the settings I used should have been perfectly fine.

1 core; 37x
2 core: 36x
3/4 core: 35x

Core Voltage: 1.175V
Adaptive voltage
Cache Ratio: 37x
Cache voltage: Default

Graphic multiplier: 10.5x

Didnt even touch the dynamic voltage offset

 

Would it be stupid to try downclocking it through BIOS instead of XTU?
Just change the multipliers?

 

You can do that either way. The problem was user error, not XTU error. XTU works great as long as you do not apply the wrong values. It's a Windows UI for settings (most of which are hidden) in the BIOS. The voltage being too low to power the GPU is what caused your machine to not boot any more. I have never lowered the multipliers before, so I am not sure what will happen. Probably nothing, other than the CPU running slower. The stock BIOS is absolutely worthless for all intents and purposes. Without XTU, you're going to be extraordinarily limited in what you can adjust. Unless you set the voltage to static mode with XTU, you can't even access the voltage adjustment menu in the BIOS.

 

Learning by trial and error, Mr Fox. Something you most likely are very familiar with

I still find it very strange that 50mV below Intel's own 4940MX specs suddenly means it say byebye when I fire up the GPUs. It passed just fine Intels Stress test that puts the CPU at 100% load with just 1.0V at the same multipliers!!

Ive read that XTU sometimes doesnt play nice but BIOS does, so maybe I will try that tomorrow. I left Core Voltage at Adaptive but that shouldnt matter too much. The different Core Voltages I tried was never working anyway. If the CPU wanted more than my 1.050V limit, it cared less about the XTU settings I made lol.
Adaptive means it will just scale voltage to its needs anyway. And since Im going down in clocks it will have a limit with more than enough to take from.

Yup Im gonna try it out tomorrow.
I also switched from 344.75 to 347.12 Geforce drivers. The 344.75 was giving me random freezes. 347.12 is working flawlessly this far but I havent tested it enough yet

 

You know, I always found it weird that the 49x0MX CPUs have a processor:cache ratio of 8x. The 4800MQ has 37x as default, and whenever I OC I usually set it to match the OC multiplier. I don't know how those chips work on 8x regardless of the speed multiplier O_O.

Also, apparently, even though I set my processor current limit to 120A in BIOS it's apparently reading as only being 72 in XTU. It also thinks the default is "73", which shouldn't be right. I distinctly remember it being 72 a long time ago, with 71 being my "proposed" when I was on default settings. I just launched XTU to check and be sure, then I saw it. *sigh* I don't even know anymore >_<. Might as well use XTU and be done with it. XD.

I wonder if my processor current limit in BIOS is a broken setting?

 

I have to agree with Brother Fox. It's user error.

 

I am the master of many disasters. It takes a lot of them to be successful.

 

Yup, cache ratio with Dell`s own settings was 40x when the highest multiplier was 44x. Default for the chip according to XTU was indeed 8x. I wrote about it in OP and thought it was weird so I lowered it to 8x to match Intel`s specs for the CPU.
Now however, with 40x as highest multiplier, I increased the cache multiplier to 40x as well. I`ve read about it and according to multiple overclocking sites the cache ratio should match the core multiplier. So why the heck Intel puts it at 8x is very weird indeed.

Its a bit tricky playing around with mobile CPUs that cost $800-100 and a notebook that takes forever to reset BIOS with.
The desktop guys have like $300 CPUs and easy methods of resetting BIOS. I sometimes wish I had desktop just for the fun

 

BIOS downclocking was successful :thumbsup:
I reduced multipliers within BIOS from the menu there and no crash at all, not from IGP or running the GPUs

Not sure if XTU is perfectly stable to be honest. Oh well, atleast Dell offer the option from BIOS. Voltage will have to remain where its at.

 

Wprime 1024 is a nice bench test you can test with. It is a more real test than xtu bench test. More like normal laptop use.

 

You can also undervolting with Throttlestop.