Alienware M18X R2 upgrade (lazy and actual guide)

I see many people have sold their Alienware laptops and got new Eurocom, Clevo etc. Well, I'm the lazy guy and thus I haven't sold mine Alienware M18X R2 Nebula Red yet. Initially it came with i7-3610QM CPU and AMD 7970M CrossfireX, 8GB RAM. After 2 years of using both GPUs have failed and I decided to get a replacement. It was difficult to choose a new GPU but now I have both Nvidia GTX 970M and 980M. Also, I got i7-3940XM CPU and 32GB RAM, SLI cable for future experiments.
I think AW M18X R2 is really a nice laptop and despite of all other followed and newer models it should not be sold. So, I will do my best to upgrade its software and hardware parts taking in consideration the common sense and money factor. I don't work with AutoCAD or graphical software. I do web browsing, sometimes play MMO, watch and edit camera recordings etc. Just a semi-casual user. Usually I don't OC all the time, I don't buy the latest Extreme CPU right after its launching on the market or even after 1 year of it. If my 2 7970M GPUs were still working I continue play with them.

 

RAM upgrade.
8 GB 1333 => 32 GB 1600 Corsair Vengeance. Lately I have discovered that my mobo doesn't work well with all 4x8GB RAM installed because it failed to pass the memory BIOS tests and started making BSOD or lost the audio occasionally. I phoned Dell's support and got the info that my motherboard's memory controller seems to be failed and buggy. I was advised to do what I have found by myself previously - work only with 3x8GB RAM, 24GB in total. With that amount of memory all the tests were finished OK and audio problem has gone also.
By some reason I can't OC my RAM but it's not so important. If you do have some audio problems after resuming from sleep mode or occasional BSODs or "memory integrity fail" error, I advise you to install less RAM memory and find your own proper working amount.

 

CPU upgrade.
I7-3610QM => i7-3940XM. This update seems to be logical because 3940XM is the last and most powerful CPU that you are able to install in AW M18X R2 based on chipset restrictions. Forget about possible BSEL mod, because on my experience this mod can limit your available RAM or other things if there will be no corresponding BIOS modification.
3940XM is a little more hotter than my old 3610QM and thus I need triple-piped CPU heatsink. Currently I have only double-piped heatsink. Also, a good thermal compound is recommended. I was advised to use Gelid Extreme. I can't say it's fantastic because I don't like how fast it becomes solid, but, yes, it's better than Arctic Cooling MX-2. Don't use MX-4 because it also becomes solid very fast and less durable. Pay attention to the condition of your heatsink's pad surface - it should be like true mirror for better heat conductivity. If you like me have some scratches on it, polish it with a special metal polishing compound prior installing. You also will need a proper version of Intel's DPTF driver. Currently I'm looking for the best one. Don't install the latest (2012-2013 dated) from the official Dell's site because in my case this causes CPU errors - the system trottles with it.
If you install Swick's A11 modded BIOS you will see some unlocked options and will be able to choose BIOS to control your CPU work and temps instead of DPTF. Personally, I decided to work with DPTF, just need to find a proper and actual driver.

 

GPU upgrade.
AMD 7970M CrossfireX => Nvidia GTX 970M / 980M. I have decided to cancel my efforts with AMD because they suck. And I tired of this fact and faulty high-end AMD GPUs. Nvidia also had some bad sheeps in the past but there are only two brands on this market, so, the choice is simple.
You need to install Swick's A11 modded BIOS prior installing 970M / 980M GPUs. In my case I had official A12 BIOS from Dell. So, I did this: A12 official => A11 official => A11 Swick's modded. In A11 Swick's make sure you disable the non-flex CPU support (thanks Mr. Fox for this info).
So, all the steps:
1. Install Swick's A11, disable non-flex CPU support.
2. Install 970M / 980M GPU. You can use Clevo's heatsink backplate if you cut some angles and holes with a metal scissors. Also, you need Clevo's screws and x-bracket. You can use your old 100W GPU heatsinks from AMD cards.
3. Download the latest Nvidia drivers for laptops, unzip. Modify the proper inf file - for Dell you need only one file (read here). Delete all the software from the installation set you don't want to install and install the drivers. Reboot.
4. In my case 970M and 980M work without any throttling after this installation. 3dmark 11 scores are about 9800 and 12500 respectively. 970M working temps are 34 (resuming from sleep) - 70 (gaming) C - 74 (max gaming Temp ever reached). 980M temps are 36 (sleep) - 74 (gaming) - 76 (max gaming). IMHO 980M scores the temperature more faster.

 

DPTF driver.
A few hours ago I have found the latest and proper DPTF driver (version 8.1.10605.221, Nov, 2016) for the system. Here is it.
Also, the latest Intel Chipset Driver has version 10.1.2.77 (Aug, 2016). Make sure you have updated it before installing DPTF. Currently I'm using Intel(R) USB 3.0 eXtensible Host Controller Driver version 1.0.10.255 (Feb, 2014) because I have not found any newer working version. And Intel Management Engine Interface (July, 2012) which was installed directly via Windows Update. Also, Windows Update has installed me some interesting update "Intel - System - 6/28/2016 - 14.35.1.69" which, I'm sure, is important for all the present work.

 

What about the 980M SLI GPU cooling? In my case since I upgraded my 780M SLI for a 980M SLI my gpus cooling does not work properly and I had to resort in install the HWInfo program to solve this problem. Would this DPTF driver could solve this problem?

 

I have not installed SLI so far. Because I have only 1 x 970M and 1 x 980M. In your case, what do you mean by saying "cooling not work properly"? I assume you are talking about fans. Currently my left (primary) GPU fan starts to spin after GPU reaches 64 C.
When I had 2 x 7970M right fan always started to spin with some delay, it's normal thermal policy for AWs, as I think. Maybe updating Compal sensor driver / FW may help to solve this. Or BIOS modification.
DPTF basically corresponds to CPU thermal behaviour. For example, unproper DPTF version have caused my 3940XM to lower its clock on all cores to 800 MHz (8x multiplier).
BTW, HWinfo is not so simple. There are some cons and pros with it. In a few words - I don't recommend to use it all the time. You will just ruin your fans soon and that's all. It will not give you more than 3-4 C of additional cooling corresponding to native AW BIOS fans cooling. If you have properlly installed all the heatsinks, cleaned fans and used good thermal compound, and your GPUs are alive, you don't need to use HWinfo. Yes, you can create custom fans profile (up to 4000+ RPM) but for a single 970M / 980M you are more than ok with stock AW BIOS profile. Max T under load for these cards is about 69-79 C which is not too much, I think. 7970M was a bit hotter. And I think average temperature 70-74 C under high gaming load for 9xxM is not bad at all.

 

Tulius, do you have throttling with your 980 SLI BTW? I was said to restart the laptop to see the throttling. I did and still nothing. Clock and memory are constantly at their maximum boost levels under highest gaming load. I have not installed Prema's vBioses. I have not disable Boost. My stock vBios let me to OC. I don't use Optimus.

 

Hello, thank you for this guide.
I have a question for you. What OS you have installed? I read that 980m + M18x R2 + W10 causes the death screen. It is so? There is no solution, perhaps with modded vbios (Prema) and modded drivers (J95)?
Thank you.

 

Hello and welcome! Windows 7x64. I have heard about some problems with Windows 10 and screens, yes. I used to work with Windows 7 on this laptop, so, I just have not upgraded... yet. IMHO, modded drivers will not help you with the issue. By modding the driver you just add one string (like 13D7.0550.1028) to the proper inf file and that's all. And as to modded vBios, I have not faced the issue that caused me to install any modded vBios yet. I advise you to ask @Mr. Fox about Windows 10 & dead screens. I'm sure he knows more about it.

 

Thanks for your quick reply.
I ordered 2 980m and a new motherboard, I would not to order a new screen.
I'll try to contact Mr.Fox and hear what recommends to me.
Thanks.

 

You are welcome I recommend you not to install any modded vBios at first. Just do all the steps I did and see will you GPU trottle or not. If there will be no throttling, try to OC via Nvidia Inspector. If everything will be ok, you don't need to use a modded vBios. I will try to install 970M SLI soon...

 

Ok, for the moment i need to wait because the gpu and the motherboard are on the way with the carrier and i don't have idea about the time to shipping and the work that I will have to do. Thanks again.

 

Yeah I hate to be the skeptic here but I don't believe youre not getting throttling with the default 980M vbios, youre just under the illusion that there's no throttling.

If youre looking at GPU utilization in OSD while benching or gaming, youre doing it wrong. It will say 99%, but if you look at your GPU frequency it will be highly erratic, dropping from say 1100MHz or so (whatever your OC is) down to 900MHz or worse every other second.

I just installed 980M in August of last year, the default vbios did the above, Nvidia isn't capable of removing this throttling behavior with a display driver as this behavior is confined to the vbios.

So yeah, again, I don't believe youre not experiencing throttling with the default vbios, just the perception of no throttling.

 

As you wish.

 

Meanwhile, I have sold my 980M and got another 970M to pair my SLI. 3940XM has the first OC BIOS preset now but my own multipliers that I have choosed based on some experiments. I have not OC'ed both GPU yet. SLI is enabled. I have used Arctic Cooling Thermal Pads and MX-2, because all my Gelid Extreme stuff gone bad. I also installed 2 GPU backplates - one partially sized copper and one metal full-sized which I had to cut a little to fit in. The SLI cable was isolated on its backside from possible heating of GPU heatsinks area. So, here are some screenshots. Unigine Valley Benchmark (High and Extreme HD presets), Fire Strike, 3DMark11 P stretched and some example of so-called "no-no-throttling" 980M.

PS. Don't forget to use Stretched mode in 3DMark11 Perfomance default test to see the real, not capped, FPS result.

 

Blade and Soul in-game FPS (single GPU and SLI enabled). Blade and Soul does not support SLI and even sli bits 0x43402005 is not good for it.

 

I have also used this trick to activate "SLI-without-reboot" feature. Works fine. Thanks, @Mr. Fox. I will add this feature to my next Nvidia's driver mod. Actually, this trick with regedit works a little weird: it does enable SLI on-the-go mode but you have to reboot to see the SLI in-game bar. But it's not frustrating anyway. The main advantage that it works as it should - when SLI enabled you can see both GPU works together if the game supports SLI. If it doesn't - you just see the main (left GPU) does the whole job.

 

Sorry to be so late to answer @kaza but let me show something interesting:
-My results with 980M SLI with no overclock and using Prema vBios, the first one with throttling and second one without throttling(using IGFX to PEG bios trick):



I'm using the full 32MB 1866MHZ RAM without BSOD errors, only occasional sound board not found bug but with some reboots it goes back just fine. If I disable the HWinfo program my GPUs goes above 93C with fans not responding so the notebook do a complete shutdown in the process, I didn't have this issue with 780M sli.

 

@Tulius, thanks. Can you show me your GPU-Z's screenshot with throttling?

 

It seems we have the same RAM sticks - 2x8Gb CMSX8GX3M1A1600C10 and 2x8Gb CMSX8GX3M1A1600C1 are here. What about yours? I can setup my SPDs to work with 1866 MHz, but this sound problem that you have mentioned is frustrating. And I don't see any obvious speed improvements while in 1866 mode (I use AIDA64 to monitor this), so, I work with default 1600 Mhz mode. I aslo can't work with XMP profiles 1 and 2 because they just won't to setup for these SPDs.
I do not use HWinfo with 970M SLI. My left fan usually starts to spin after reaching 63-64 C and works until 41 C. The same goes for the second GPU fan. Sometimes CPU fan starts to spin before GPU fans, it depends on situation. The same was with one 970M and one 980M installed.

 

I suppose, 980M SLI always throttles in M18X R2? Besides I have not seen a case with throttling 970M SLI yet... I had had some idea about 100% possibility for throttling 980M SLI, that's why I even have not tested it. Instead of it I have tested:
1. single 970M,
2. single 980M,
3. 970M SLI,
and I had chosen the third option as the most currently reliable.

 

Just to clarify:
- My memory RAM stick are not overclocked but default 1866MHZ ram sticks with part number CMSX8GX3M2A1866C10
- 980M SLI not always throttles, if the gaming with gpu usage is below 91% it works ok but if above 92% the throttle occurs, so with less demanding graphical option in some games you can play just fine. With IGFX to PEG trick you can play with 99% gpu usage with SLI enabled
- The sound bug is a somewhat rare occurrence here and when occurs I just reboot one or two times to fix it.

I still think the 980M SLI is the best option around because the throttle fix is very easy to do and its much more powerfull than 970M SLI, I can play very demanding games like Assassins Creed Unity in very high quality in 60fps.

Here's a gpu-z pic with throttling present:

 

Well, I have not seen this throttling with 970M and 980M (usage up to 98%). 1600 MHz is the fastest RAM that 3940XM supports. But if you say your laptop works handles native 1866 sticks very well, I will try test it too... Currently I have the sound bug only in two cases:
1. if I switch my RAM speed from default 1600 to 1866 MHz and
2. if I activate DPTF in BIOS and install buggy DPTF drivers version. With proper driver version and DPTF disabled in BIOS the problem never occurs.
BTW, you don't have to reboot one or two times to fix the sound. You can just go into sleep mode and resume from it one or two times, it helps in the same manner.
I don't play Assassins Creed, but I have another demanding game and FPS with 970M SLI non-OC is about 60-61. I think it's some kind of FPS cap since this game also doesn't fully support SLI.
Used 970M costs 300-350 USD and used 980M costs about 500 USD. They have different thermal profiles and basically 970M SLI non-OC'd is just about 50% more powerfull than single 980M non-OC'd. If you do like how 980M SLI works in M18X R2 it's great. Personally, I did not like the work of my single 980M. It was too hot, too fast in getting hotter and the advantage in FPS comparing to 970M was about 15 FPS in all my applications. My whole M18X R2 (i7-3610QM, AMD 7970M CF) price was about 2000 USD and paying 1000 USD for 2x980M is not the best option, taking in consideration that you also have to play with that throttling trick and stay with your default 330 W PSU... It's all personal of course. But as I said I do support "i7-3940XM+970M SLI" more because the thermal and TDP profiles of this pair are natively close to the maximum of M18X R2 capabilities.

 

@kaza , yes I can't argue about the price 1000 USD for 2x980M is a bit high indeed, but I dont think they actually run too hot because my old pair of 780M were much hotter reaching more than 81C on full load but with my maxwell cards I never past 70C even in hours of play but I think 970M SLI can run even cooler than that. Would be interesting to see how 970M SLI points can reach in 3DMark11 benchmark.

About the memory I know some users here could achieve 2133mhz speeds using native memory chips like Kington Hyper X or overcloking 1866mhz to 2133mhz in M18X-R2 just fine.
My cpu-z info:

 

@Tulius , I have worked with AMD only in the past. 970/980 are my first serious Nvidia GPUs. I can say that 7970M was a bit hotter than 970M. And 980M is hotter than 970M. My 970M SLI non-OC'ed 3DMark11 Perfomance (stretched) preset score is in the screenshot attached a few messages before.
Thanks for the tip about RAM. I would like to see actual CPU-RAM speeds with 1866 and 2133 MHz sticks. If you have some examples, post, please. Like this one (see attached). Maybe I will take some DDR3 1866 MHz stick from the office laptop and install it in M18X... The main problem is the price of a new RAM sticks. If there will be a huge improvement in speeds between CPU and RAM with native 1866 and 2133 I will try to replace my current 1600 for sure. RAM speed is a huge speed bottleneck usually. Previously I have shown my results with 3940XM and single 970M to @Mr. Fox and he said everything looks like it should be for usual M18X R2.

 

I only can show mine but I find one that show Mr. Fox using 2133Mhz ram sticks within his old M18X-R2


And that's my Aida64 score:

Hope thats help.

 

Yes, many thanks! I will start from upgrading my AIDA64. I see your version is much newer. Your memory speeds are much better too. I wonder if it's all because of 1866 sticks used or AIDA version... 3920XM and 3940XM are almost similar. And I see @Mr. Fox overclocked his 4x4 RAM sticks but I don't know their part number. It's hard to find DDR3 SODIMM 1866/2133 1x8Gb stick now. I have read about DDR3 1x16Gb sticks but they seem to be not working with Intel HM77 and 39x40 CPUs...

I have read @Mr. Fox used Thaiphoon Burner to modify his RAM sticks.

 

Here is my AIDA64 v5.80.4000 result with 32Gb 1600Mhz RAM currently installed. As you can see your memory and even CPU L3-L3 speeds are better due to a higher memory bus and bandwidth.

 

m18xr2 support low voltage ram version too, so I'll suggest to buy one of them too stress less the imc and for a superior overclockable margin.

My prefer choice will

G.Skill F3-2133C11D-16GRSL
or
Kingston hiperx impact HX321LS11IB2K2/8

Here my ram in aida64 memtest



If you find useful I can ran 3dmark (vantage/11) with ram setted 1600 vs 2133

Regards,
Luca

 

Many thanks! Yes, it would be great. I have found that it's possible to flash 1600 RAM with 1866J and even 2133 dumps. For example, my current Corsair 1600J 1.5V sticks do support 1.35V 1866J and 2133K modes. All I need is a shareware version of Thaiphoon Burner.

https://www.micron.com/parts/modules/ddr3-sdram/mt16ktf51264hz-1g6

Also, SPDTool 0.63 doesn't support M18X R2 chipset. So, it's not possible to edit and safe your own SPD dump with Thaiphoon Burner and then apply it to your RAM stick via SPDTool.

 

I tried that but I find the video tutorials not enough for me to achieve this feat.... maybe I was scared to brick my memory stick or worse.

 

No youre getting throttling, 100% guaranteed, youre just looking at the wrong indicator of throttling, the GPU utilization and not the actual frequency.

Let me elaborate because I have both data in my OSD via MSI Afterburner, Rivatuner.

980M without PEG to IGFX to PEG in a game or benchmark the clocks will hold at 1150MHz and 99% utilization for a minute and then out of the blue the clocks will drop all the way down to 500-600MHz yet utilization is still indicating 99%. It will do this on avg. every 2-3 minutes, effectively halving the frame-rate.

So if you would, run Unigine Valley or any game, set up your card(s) to display frequency, and you will see what we're talking about.

 

No actually 970M SLI may run as hot as 780M SLI and not as cool as 980M SLI, let me elaborate:

This has to do with GPU load as a contributor to heat. With 980M SLI, which is basically the equivalent of a GTX 980 Ti at default clocks, youre going to be extremely hard pressed to tap that out in any game. Youre probably going to see utilization, at the 60 FPS limit of the display, around 60, maybe 70% tops at 1080p? With my single 980M, when utilization is 99% sustained at 1150Mhz core and 1.00v the temps will continue to rise, abeit slowly, until the low-to-mid 70's. As soon as utilization drops from 99% to 70% the temps begin to decline very quickly, even though voltage is constant. If they stay at 70% the temps will drop down into the low 60's. This happens in some games, say Fallout 4 or Skyrim, when you go from an exterior environment to an interior environment.

This is why 780M SLI APPEARED to run hotter, the cards, which were effectively half as fast, were ALWAYS at 99% load, even in the interior environments and the SAME logic applies equally to 970M which is about 30% slower than 980M and is only marginally faster than 780M (50% faster, whereas 980M is 100% faster):

780M (5500): http://www.3dmark.com/fs/1461560

970M (7300): http://www.3dmark.com/fs/2724803

980M (1000): http://www.3dmark.com/fs/6327491

Another good example, I remember trying to run Planetside 2 back in say 2013 or so with a single 680M because SLI was and still is broken and I couldn't believe that the card was hitting 85C @ 1.050v (back before I dialed it back to 1.025v). Well, out of curiosity I returned to the game with 980M and because the card is 2x as fast the load was around 60% and the temps were in the low 60's with the default fan algorithm!

This is why people running 980M SLI aren't seeing more than say 70C at the most even with Prema's hot 1.065v vbios: there isn't very much load on the GPU's whereas everyone with single 980M, myself included, is experiencing 85C to include 90C induced black-screens.

I am so glad I managed to get my hands on a 1.00v vbios that doesn't throttle as this was very scary.

 

Kaza, don't waste your time, mental energy or money monkeying around with RAM speed, it has nearly no impact on real world performance!!!!



1300MHz performing within 1% of 2400MHz!

Also, I don't know how long you've owned your R2 but in my experience the most optimal OC for an extreme CPU is no more than 4.2GHz and 1.275v! (6 FLEX Voltage in BIOS). Any more than this and sure, you can run 4.4GHz but have that tube of expensive thermal compound on hand because youre going to need to tear down your unit every 3 months to change it! It gets really old, really fast. That 200MHz is absolutely NOT worth it!

The CPU has a sweet spot, where more frequency demands literally exponentially more voltage. I believe the i7 3920 =3940's sweet spot, unless you won the silicon lottery, which you probably didn't, is around 4.2GHz. Just trying to go from 4.2GHz to 4.4GHz and you need to turn voltage up from 1.275 to 1.4v! Temps get insane! Thermal compound goes on early retirement!

Also, just out of curiosity, was your 980M with Premas 1.065v vbios? Because with the vbios I have I'm at around 10,400 GPU Firestrike at 1.00v (zero throttling, GPU Boost 2.0) and with the unit elevated, Gelid GC Extreme on the core, and the fan's manually set to 3700 RPM the temp's don't exceed 75C under 99% load sustained at room temperature! 75C with 99% sustained load, my last testing session was 45 min with Dragon Age: Inquisition and the temps were in the high 60's! This is acceptable! I'm getting a copper back-plate in the mail soon, I will try that and report back to see if there is any improvement as well.

(The following data was about 20 minutes idling in a web browser with 0% GPU load which skewed the averages, eyeballing the utilization it was between 70% and 99% most of the time during the 45 minute test session)




980M at 10.5 GPU Firestrike is better than 970M at say 1300-1400k GPU as SLI has become a legacy feature and you will only have that 13-14k performance 70% of the time and usually not in that new game you want to enjoy!

Like Resident Evil 7? Better hope your primary 970M can run it, oh and expect single 980M temps under that 99% sustained load!

Same goes for Titanfall 2 and about 25% of many good games released in the last few years: SLI support is either seriously delayed, or just never arrives period!

https://steamcommunity.com/app/418370/discussions/0/141136086935515182/

(They just only now added SLI support for Battlefield 1 which was released in October and......DIABLO 3 WHICH WAS RELEASED 5 years ago LMFAO)

Batman: Arkham Knight? No SLI support, you will not be able to run that with a single 970M, I should know because I can't run it with single 980M lol. Great game BTW, despite a lot of criticism.

Wolfenstein: New Order, no SLI support.

The Evil Within, no SLI support.

Planetside 2, no SLI support.

Etc.

Etc.

Etc.

I would sell the 970's and go back to single 980M! There are ways to deal with the heat! Elevate, manual fan control, no more than 1.00v, good thermal compound and possibly the copper-plate and you won't see temps more than the low-to-mid 70's! Hell it only peaked at 76C after a 20 minute Unigine Valley loop and 99% load sustained with the default fan algorithm at 1.00v!

I mean the only reason I stayed with 680M SLI was because, hey being able to play 70% of games is better than not being able play any games with a single 680M but I patiently waited to move away from 680M SLI and with 980M being about 90% as fast as 680M we are very fortunate. Same goes for my desktop, where I took a similar hit in performance pulling the 780 Ti SLI out of there in favor of a single 980 Ti, I lost about 10% performance but hey, now I have 90% of 680M SLI and 780 Ti SLI performance respectively in all games, not just 70% of games where the other 30% I can't even play unless I turn everything completely down.

 

AS vulcan78 stated early, the ram speed doesn't effect much on system speed of everydays use but if you want the maximum from all your system and have money to blow away, why don't upgrade every single part?
In the meantime some software will thanks for the max bandwidth aviable on more speed.

So here the comparison tests, because this early morning I lose time to do it I wanna post it anyway:

1600Mhz on the leftside - 2133Mhz on the right column

Vantage Performance
http://www.3dmark.com/compare/3dmv/5580047/3dmv/5580038

Vantage Extreme
http://www.3dmark.com/compare/3dmv/5580053/3dmv/5580044

3dMark 11 Performance
http://www.3dmark.com/compare/3dm11/11960246/3dm11/11960220


We come from 0,2% to 8,4% better score with an improved ram speed at same latency.
Take it with a grain of salt, as all benchmark suite.

Best Regards to all,

Luca

 

Once again, point me - where are these drops?

 

Nope. Have you personally tested single 970M or 970M SLI? I did. I don't know about 780M SLI actually, but logic is not the same as testing results. That 3DMark score was for 970M 3Gb version, different chipset, laptop, CPU and RAM. What exactly should it show?

Yes, indeed. And that's the point? 970M works in the same manner - starts from sleep mode with 33-40 C (depends on the ambient T), moderately rises until 61-63 C, then here comes the fan, drops it to 41 C. While under 99% load (Valley, for example) both 970M achieve 69-72C and hold it until the end of test. Under game load 970M usually gets 63-69 C at max, depends on a game. Old games, like Blade and Soul, charge 970M for 63-64 C at max only. Besides the same game had charged 980M for 69-72 C constantly. I don't see a lot of extrapolated logic here. My 7970M CF was much hotter than 970M SLI. Personally, I do like more GPU with less shaders and less TDP, because I always can OC them if I need and don't worry about TDP limitations, native power throttling and PSU replacement.

 

The answer on this question can be found in some starter message of this thread I have purchased my R2 4 or 5 years ago. And it came with 3610QM initially. I have replaced it with 3940XM C0 a couple of months ago.

In my first messages I have clearly told that all my vBioses were stock, no Premas. And currently I'm on fresh MX-2, not Gelid Extreme. But I have used Gelid too. I agree, in my case 63C with 99% sustained load is acceptable. And I have two backplates installed as well. One of them is full-sized and non-copper (secondary card) and one is copper and not full-sized. All previous tests with single 970M and 980M I did with that non-copper full-sized backplate. Firestrick 1.1 result for 970M SLI is here. I had the same for a single 970M/980M, but it's somewhere in the thread on Techinferno... Can't find the screenshot now.

Everyone has its own list of games preferred. I'm not so young and I used to see FPS 70-110 with proper GPU in all my games. That's enough for me.

Well, I have ruined my last desktop a lot of years ago. Just replaced it with laptops. I'm not a hardcore maniac player and true fan. All my demands and tasks usually can fit into R2 possibilities. As I said before, 980M costs 500 USD and 970M costs 300 USD. I have 2x970M for 452 USD ((300 + 395) - 243 (sold failed 2x7970M)) now and they both are 50% more powerful than a single 980M (I was told they should be 50% more powerful than a single 980M). Yes, I don't have a lot of tasks/games with true SLI support to see that "50%" advantage, but I do have the profit in price and TDP. And it's more than acceptable for now. If I see 2x980M with price tag 400 USD each and less I of course will seriously think about it.

 

Thanks, Luca. Will investigate and compare it soon. I don't like to blow away my money, that's why, for example, I got my 3940XM only a couple of months ago and for 235 USD.

 

It's not as difficult as it looks like. I can write an instruction if you want. Or you always can ask @Mr. Fox , he did the trick (rised up to 2133 and replaced timings). One of possible benefits is adding native 1,35V support. Both your and my RAM sticks supports 1,35 and 1,5V modes. Even more, there is a great chance that your 1866 stick differs from mine 1600 with the profile loaded only, because both of them based on Micron MT16KTF51264HZ/1G64HZ.

 

@vulcan78 , here is another screenshot from one of my games showing the real GPU load. Where are those throttling drops should be? GPU Clock Core or GPU Load? My ambient/room T is about 29C now (outside the apartment we have -14C now).

 

BTW, I have no WLAN card currently installed. And no active USB 3.0 connections. They can interfere with each over and create noise.

My 970Ms were manufactured in August and October 2015 respectively. No coil whine so far under high load. ASICs are 64,2 and 62,7%.

 

No I'm not recommending going from 970M SLI to 980M SLI, I'm recommending going with a single 980M. And the performance gap between 970M SLI and a single 980M is closer to 40% (14-14.5k vs. 10k GPU Firestrike).

Oh and have a look at that combined score with both of our Physics tests being about equal:

970M SLI Combined: 3400

Single 980M: 3800

http://www.3dmark.com/fs/11587345



So yeah, this goes back to your TDP argument that you tried to make, which I believe absolutely is valid when comparing 970M SLI to 980M SLI but not to single 980M. You might be experiencing PSU wattage starvation here, meaning there isn't enough wattage for both you 970M's and your Extreme CPU, leading to not a 40% improvement in performance but an actual ~15% degradation in performance (whatever the 400 difference between 3400 and 3800 is). You don't have to worry about this if you don't play a CPU intensive game really, but if you find yourself say in downtown Boston in Fallout 4, yeah youre going to feel it.

So actually with a single 980M you have a profit in TDP and Price, sacrificing 40% performance that you will have on tap 70% of the time, even less if you don't count the games that have official SLI support but exhibit piss poor scaling, i.e. secondary GPU is at 50% utilization, this is increasingly common nowadays. 970M SLI = ~180W and $600, single 980M is 110W and $500.

63C is far more than merely acceptable, I would say that you have some thermal overhead there.

Out of curiosity, what kind of difference did you see before and after installing the copper backplates? I've looked high and low regarding this but haven't been able to find anything, I'm getting a copper backplate in the mail soon for my 980M.

No what I'm saying is that the rationale you used to justify moving from a single 980M to 970M SLI was that 980M ran too hot, what I was trying to explicate was that a single GPU tends to run hotter as the avg. load is significantly higher than having that load shared among two cards BUT that by taking a few precautions, which I laid out, you can deal with said heat.

The problem with the GPU-Z sensor data you sent me is that there is no time data, has three minutes elapsed, 5 minutes, 10 minutes, half an hour? I could be mistaken, but I was under the impression that both 980M and 970M exhibit a weird throttling behavior without doing PEG to IGFX to PEG in BIOS. I know that 100% of 980M do this, it may be different for 970M.

Also referring to your other comment about 70-110 FPS, are you using an external display? Because if youre simply turning v-sync off with the native 60Hz limited display youre simply stressing your GPU's unnecessarily.

Well I suppose I get to eat my hat unless that Physics test in 3DMark 11 is a fluke. Never seen anything like this with RAM frequency.

 

I have tested single 980M and it wasn't so great, IMHO. But maybe I hadn't perfomed enough tests. Also, I used to have 2xGPU... If my old 7970Ms were still alive I continue play with them.

I have not OC'ed my CPU a lot, just a little. At higher FPS CPU perfoms a lot of job and combined score can be different if you change your CPU TDP limit, rise multipliers etc. I don't see any PSU wattage starvation - CPU eats up to 100W, both GPUs are non-OC'ed and should eat about 80-90W as well. I have shown my 3DMark11/FireStrike results with single 970M and 980M to my friends and they said my results were relatively poor compared to other similar hardware configurations. But there is nothing about PSU or battery. I did all tests with non-OC'ed CPU and with my 1600 memory which is not so fast. Also, maybe my motherboard has some other issues, not only with memory controller... But, anyway, I'm curious why you are comparing combined and physics scores for my 970M SLI and yours 980M in Fire Strike? Why not the whole score or graphics score?

I have read that actual 970M TDP is about 80-100W and 980M has 122-130W. Single 980M may be a good option, but as I said many times, I wasn't so impressed by its work and temps in my R2. I see you had a different experience, it's really good. Actually, I wasn't sure that my possible 980M SLI will not throttle. At first I got one 970M, did some test. There was no throttling. Then I got 980M and did the same tests. Again, no throttling. Then I decided to get another 970M and build 970M SLI to see how will it throttle. And because I occasionally have found a buyer for 980M, I sold it. Now I'm with 970M SLI and despite some poor (as usual) SLI scaling support I'm more than OK with it. But you tries to argue me that I should go for a single 980M, again. LOL. Maybe, who knows. As I see no current throttling with 970M SLI it's OK for me at now. I still don't know what exactly causes this throttling all people talking about. Even @Mr. Fox doesn't know it, so do you and others here.

Have you tried to install any Clevo's backplates in R2 before? You will have to cut some angles and holes with a full-sized plate. There will be no space for the northbridge (or southbridge?) heatsink if you install it uncutted. And there will be some direct contacts with some other chips around the left GPU board. Before I installed the backplate I have tried to install 970M with some thermal pads on the backside memory chips only, Arctic Cooling thermal pads. Well, I can't say it was a great idea, but it was better than nothing. The difference between "no backplate" and "full-sized backplate installed" is 1-2 C at idle and more faster temperature drop when the fan works. The main problem with a full-sized backplate is that it can become very hot after continious game load and thus work as an owen, it also covers almost all the back side of the VGA board, so, there is a tiny air gap for a proper fan work.
On the contrary the small copper backplate covers not the whole board and leaves more air space for a fan. And it also dissipates heat better but at high and continious load only. For example, it can be visible via Valley Benchmark tests after first 10-15 min. The real difference between these two backplates is about 4-6 C. At minimal or moderate load there is no much difference. I don't have a second copper plate now, but I'm looking for it now. Currently I have a full-sized backplate on my primary GPU and a copper one on my secondary GPU. But all these my temperature tests and observations may be very unstrict - as I said my ambient T is too high now and there is no much statistics yet. But with a small copper backplate you don't need to cut its holes&angles. So, yes, basically it's a good idea to install it.

Got it. Yea, you maybe right here.

All the time. 5 min, 30 min, half an hour etc. Actually, it looks the same all the time. And I have never changed my legacy support, PEG to IGFX or something else corresponding in BIOS after I installed these 9x0M. Yes, I have read that all 980M do this and that's why I have asked the details about it on Techinferno but I got no exact answer. It seems that nobody knows. And nobody have tested 970M SLI without throttling before me.

No, I don't use an external display now. I did in the past, 2560p panel, but not now. I used to use Perfomance optimization option in Nvidia control panel, actually. I guess, this corresponds to "v-sync is maintained by 3D-application settings" or so.

 

And here is how we "rock" without SLI, just single 970M. Compare to 970M SLI result: physics scores are identical, combined scores difference is about 26% and graphics score difference is about 82%. Overall score difference is about 54%. Comparing to a little OC'ed (Core clock 1152 MHz, Memory bus clock 2601 MHz) @vulcan78 result
http://www.3dmark.com/fs/11587345
I can see that single OC'ed 980M has these benefits comparing to non-OC'ed single 970M: +37% combined score gain, +5% physics gain, +35% graphics gain and +32% whole score gain.

And for pair single OC'ed 980M - non-OC'ed 970M SLI: -8% combined, -5% physics, +35% graphics and +17% whole score gain. That strange result shows, IMHO, that my CPU works not so great as it should be, it should be OC'ed properly, because currently it just perfoms as a bottleneck at higher FPS load.

If I had mistaken somewhere, please, point me.

 

I have tried to set CPU OC preset 3 in BIOS - see the screenshots attached. Almost no difference. Thus, I was wrong about CPU OC. Or maybe I'm doing it in a wrong way. Or maybe RAM memory impact on all this is not so small.

 

No your CPU is fine, it's asking a 330W PSU to give up 350W that is the source of the performance woes when all three components are stressed simultaneously. The 680M SLI I had before upgrading to single 980M did the same exact thing, it did like 11,500 or so GPU @ 1006Mhz core and +300MHz on the memory @ 1.025v compared to the 11,300 GPU of the 980M with a baby OC at sane voltage BUT it did EXACTLY what your 970M SLI is doing in combined.

Looking at your combined score, my higher CPU score (by about 1k points) is skewing the combined result a bit but you can see what I'm referring to, wattage starvation possibly being the culprit: 680M SLI to single 980M, 680M SLI is 33% slower in combined compared to single 980M at sane voltage, and nearly 40% with the 980M at 1262MHz core and 1.065v:

http://www.3dmark.com/compare/fs/7834821/fs/9814565

13 vs 19 FPS that is a massive difference.

The only OC I have is +100MHz on the memory, 980M is doing 10k with default clocks, so maybe +200 points with the memory OC.

I would say that you have an awesome system if you have that performance without having to do the PEG to IGFX to PEG trick, it's just too bad that SLI support has become a Legacy Feature on the part of NGreedia, along with 3D Vision.

 

I can try to help you with your BIOS settings, what frequency are you at? Yeah youre down about 700 points if youre at 4.3GHz as the benches above were done at 4.3GHz. This one is 11,200 CPU at 4.2GHz with sane voltage and temperatures on all of the components:

http://www.3dmark.com/fs/11587345

I think I'm actually going to do some stress testing to see if I can get away with +200Mhz on the memory at 1.00v, I will report back.

Update:

Picked up 70 points in Combined going from +100 to +200MHz memory, testing +250Mhz memory now:

http://www.3dmark.com/fs/11634940

Update:

Well it survived both Unigine Valley and Firestrike at 1164Mhz core and 2708MHz memory @ 1.00v, but it won't do more than that without a display driver failure. I don't know if I have a dud or what, I wish it would do 1202MHz core at 1.00v but the most I can get out of mine is 1164MHz. I wonder if some time spent flirting with 85C with Prema's insane 1.065v vbios diminished the performance somewhat, not sure.

Anyhow, it's good for 10,5k GPU and 9k overall now!

http://www.3dmark.com/3dm/17835264?

I'm going to try some longer term stability testing with Dragon Age: Inquisition, I will leave the default fan algorithm on so I can return with some temp data, I know that it won't get out of the high 60's with the fans manually set to 3700 RPM.

Oh and here's Unigine Valley:

 

Update:

1164 MHz core and 2708MHz memory @ 1.00v is stable! I'm still playing Dragon Age: Inquisition, going on about 2.5 hours, oh and speaking of which, this is one of the games that has very poor SLI scaling, I remember with both 680M SLI and 780 Ti SLI there was significant stutter that was completely eliminated going to single GPU.

Loving 10.5k GPU but with sane temperatures. It's only about 700 points before lower than before and 300 lower Combined but the temps, man in Dragon Age: Inquisition they would get up to 80-85C! Now they don't exceed 73C! I am so happy to finally get this 1.00v vbios that doesn't throttle. 140W is way way too much for this poor heat-sink.

http://www.3dmark.com/compare/fs/9814565/fs/11635076