*OFFICIAL* Alienware M18xR1/R2/18 Benchmark Thread - Part 3

Cheers for running the benchmark and nice results on 3dMark11, especially the GPU performance.

Your results on the Heaven benchmark largely align with my own, I'm almost touching 1800 with everything at stock and no custom vbios. Going further, I would like to request that you run the same benchmark, but with the custom setting (image attached) of quality set to ultra, tessellation set to extreme, 8xAA and 1080p fullscreen with stereo 3D and multi-monitor disabled.

It is at this setting where I notice two forms of throttling during certain aspects of the test. The throttling either presents itself with the cards down clocking, or in other cases, maintaining base or boost clock speed but dropping in SLi performance with each card running at around 70% as opposed to over 90%. Score is usually around 1300.

 

I just lost my os messing with memory timings, but run the heaven again later today

 

Sorry to hear about the troubles, dude. Looking forward to the results. Thanks again.

 

Unigine Heaven Benchmark 4.0

FPS:
56.4
Score:
1421
Min FPS:
9.5
Max FPS:
131.7

Never throttled

Ran 880M SLI @ 1.1Ghz/1250Mhz memory 1.056v gpu vcore

 

2920XM still runs really nice... here are some CPU benches @ 4.5GHz - Core Max 87°C using Liquid Ultra and nothing but max fans, 1.456V / 108.388W max power draw on wPrime 1024.



Below are 780M SLI Heaven and Valley benchmark result with a very modest 1033 GPU core overclock.

 

Wonderful... 2920XM is getting old but surely gives better scores than i7-4980BGA. Maybe i7-4980BGA and Alienware 17r2 manages 7.5 in Cinebench11.5 and 230-240 sec in Wprime 1024 With Core Max 95°C ?

 

Impressive as always, Mr. Fox!

Quick question: Do you think CPU clocks below 4GHz bottleneck the SLi performance of higher output cards such as the 780m/880m? Because you've really churned out some excellent FPS in this test.

 

I think it can, but probably depends largely on the game or benchmark you are running.

780M SLI is certainly no slouch, and neither is the mighty 2920XM. None of the 2015 model gamer-boy jokebooks for sale with the best available HQ processors and single 980M can come anywhere close to this level of performance from a early 2011 Alienware M18xR1 with a 2013 GPU upgrade, LOL.

 

Right on . . . I'm starting to think that since SLi does scale--to a degree--with CPU output, perhaps benchmarks like Heaven, at their most intensive settings, actually benefit from a CPU clock bump when dealing with stronger cards. Would you mind doing the same run with Heaven, but only keeping the CPU running at stock clocks? It would be interesting to see how much of a difference this can make.

 

Nice Valley score, does CPU overclock matter? I ran it at complete stock around ~3.7Ghz

 

In the earlier days of mobile SLi, people did find that there was a u-curve threshold on maximum scalability with multi-GPU solutions. For example, games like Far Cry 2--employing the open world Dunia engine--exhibited superior and consistent SLi and Crossfire scaling with an overclocked quadcore as opposed to the standard Core2Duo offering. In fact, if I remember correctly, the ATi/AMD GPUs had comparatively poorer performance at base clocks in comparison to their Nvidia counterparts, but conversely, had more overclocking headroom and peak performance when paired with a CPU that could produce higher clocks. In the end, depending on intensity of software and raw output of the cards, there tended to be a sweet spot where one could maximise the scaling output of a dual GPU solution.

Tom's Hardware did a test on older cards ( AMD & Nvidia), and as can be seen, even back then--with those cards and older software such as Battlefield 2--output would improve in certain scenarios when CPU frequency would be on the verge of 4GHz territory.

I am assuming that since the Unigine benchmarks are still reasonably intensive, and that many of these new cards pack a lot of computing output punch, the ideal CPU output for maximising dual GPU scaling has probably gone up by a considerable margin.

 

One detail not to overlook is running 8xAA with 1500Mhz memory.

I'm going to go for 3DM11 15K+ and will rebench Heaven with faster 880m memory overclock and cpu overclock.

Will report back in 40mins... stayed tuned. I will be using Canadian ambient air cooling. -13c


First run!!!

http://www.3dmark.com/3dm11/9341832

Rank# 5 for all 880m SLI systems!

 

Good point on the vram speed and high anti-aliasing.

And excellent run!

 

Unigine Heaven 4.0

880M SLI @ 1097Mhz core 1.056v & 1554Mhz memory (not sure of max)
CPU @ 3.7Ghz




You are right about GPU utilization drops, there are two points where it drops down to 21% or even lower. I will test on my Area-51 if it can be replicated.

CPU speed does not matter too much in this benchmark, but hopefully later today I will get around to rerunning @ 4.3Ghz.

 

To be completely honest, I have no idea on Heaven or Valley what difference it makes. I almost never run any of my XM CPUs under 4.3GHz for any reason. Even right now, just web browsing, mine is idling at 4.3GHz. Been running that way for years now with the 2920XM and 3920XM. I've never run these benchmarks without my CPUs overclocked simply because I have zero interest in whether stock performance is good or bad, or just "ordinary" performance.

Sometimes a little trick to avoid FPS drop is to just let Heaven (or Valley) run through a few times before starting benchmark mode. This loads up the maps in memory and they seem to run smoother. It does not have a big impact on the overall score, but it does help the minimum FPS thing quite a bit. I think those little stutters with drops to like 8 or 10 FPS are because the map isn't loading fast enough to keep up with the GPU(s). I always notice that on Heaven where you go inside the building, turn left in that short hallway, then right to go up the stairs outside again. That's the worst spot, but if you let it run through completely from start to finish before pressing F9 to start benching it does not happen. Neither do the stutters and stammers at the beginning of the benchmark. As you can see, my minimum FPS is about 30.

 

I'd say this slight variance is within the margin of error... Both machines running @ 4.5GHz CPU OC; 1110/1500/1.100V GPU OC.

AW18 was run with AC cooling. Just ran the R1 benchmark with nothing but fans and max core temp was 78°C.

3DMark11: Mr. Fox - Alienware M18xR1 | 2920XM | 780M SLI -vs- Mr. Fox - Alienware 18 | 4930MX | 780M SLI



So, no need to rush out and buy a maxed out Alienware 18 if you already have a maxed out M18xR1. At least the R1 (and R2) can use this nifty mod...

 

OK guys... the magic number for the 980M SLI in the M18xR2 is 324W. The moment you go above 324W for more than a second (maybe less) these suckers drop like a rock. Run Test 1 to 4 at 3.1GHz and it will stay at or below 324W and not throttle. I used ThrottleStop profiles for this and ran the Physics Test and Combined Test at 4.6GHz and the Combined Test stays unders 324W. With a heavy overclock on one 980M and 4.5-4.6GHz it exceeds 324W in Tests 1 to 4 without even trying hard.

Anyhow, I completed about 6 runs this way and it was consistent. I also modded my P-states with NVIDIA Inspector and clocked each P-state to the maximum core and memory clock the P-states can handle without crashing. This is using the stock vBIOS.

The way I see it, 980M SLI will never, ever be all it should be this way. Nothing short of a miracle from svl7, Prema or Johnkss in terms of firmware, drivers, something has to change for it to be worth a crap in an Alienware. I do not believe the best vBIOS mod on the planet will fix it unless it can tell lies to the BIOS. Something is broken at an engineering level and it isn't 980M from what I can tell. If you have a UPS with a watts meter or a Kill-A-Watt, feel free to prove me wrong on the 324W threshold. It's probably lower than 324W on the M17xR4 and AW17.

Apparently, these video cards are on a first-name basis with the BIOS and the BIOS tells them how much they are allowed to have. The moment they ask for more, the BIOS puts the smack-down on them. I guess the BIOS minds its own business on a Clevo and lets the GPUs do what they want to. Wish it were this way on the Alienware systems, but it ain't.

Here's the code for the batch file...

Code:

D:\nvidiaInspector\nvidiaInspector.exe -setBaseClockOffset:0,0,135 -setMemoryClockOffset:0,0,400 -setOverVoltage:0,25000 -setTempTarget:0,1,101

D:\nvidiaInspector\nvidiaInspector.exe -setBaseClockOffset:1,0,135 -setMemoryClockOffset:1,0,400 -setOverVoltage:1,25000 -setTempTarget:1,1,101

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:0,2,1050 -setMemoryClock:0,2,1600

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:1,2,1050 -setMemoryClock:1,2,1600

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:0,1,980 -setMemoryClock:0,1,950 

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:1,1,980 -setMemoryClock:1,1,950

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:0,0,950 -setMemoryClock:0,0,804

D:\nvidiaInspector\nvidiaInspector.exe -setGpuClock:1,0,950 -setMemoryClock:1,0,804 

NVIDIA GeForce GTX 980M video card benchmark result - Intel Core i7-3920XM Processor Extreme Edition,Alienware M18xR2



I still cannot believe Alienware thinks that a 180W AC adapter is "good enough" for a system with a single 980M. That just seems nuts to me.

 

Great detective work Brother Fox. So in other words, these cards will never be allowed to truly stretch their legs with a dual PSU mod..
Does this tell us where the throttling is coming from in other scenarios?

 

Thank you. As frustrating as it is, I guess the positive aspect of this effort is that I have done the same thing enough times to see the behavior is consistent... even predictable. Hardly anything I do other than using NVIDIA's crippled driver versions affects it differently. The BIOS, EC, or maybe both, seem to be calling the shots here.

The throttle point is 324W with a single or dual AC adapter. It makes no difference which one. Although, if it did not throttle a single AC adapter will not get you very far. In the very rare instance it does not throttle at 324W for a few seconds it goes over 500W very quickly with a dual AC adapter connected. I only wish it would stay there and keep running full blast... bummer that it does not.

If you don't have a UPS with a watts meter or a Kill-A-Watt you should get one or the other. Sometimes the Kill-A-Watt is not completely accurate, but it will still give you a better understanding of what your machine is doing. Playing demanding games it hits 324W with a CPU overclock. Some drivers seem to throttle even at 300W. I think the crappy new NVIDIA drivers that block overclocking are doing this on some system so they will actually stay running and not shut down from lack of power. I cannot understand how Alienware thinks a 180W AC adapter is OK for the 15 and 17. Maybe they are doing this to try to force gaming with the eGPU. These crappy low-TDP BGA CPUs will definitely help them avoid throttling, as show by my previous observations of clocking the CPU to a stupidly low 3.1GHz to Tests 1 through 4. 980W is a huge power hog... it's more efficient, more powerful, and uses more power to be powerful. It does it with less voltage, but stock clocks are near max overclock capacity for 780M, or even way beyond what most people have done as far as overclocking 780M, LOL!!!

 

NVIDIA GeForce GTX 880M video card benchmark result - Intel Core i7-4910MQ,Alienware 01W2J2

Decent GPU score, memory overclocks very nicely and runs 1650Mhz without a sweat. CPU was not set properly to run 4.3Ghz, even though 3DMark reports 4.3Ghz.

GPUs repasted with GC Extreme

 

Nice run, Brother Zero989. Nice to see somebody finally doing good things with 880M.

GC Extreme is excellent paste for GPUs. Does a good job and I really like how easy it is to clean up if you need to repaste.

 

Thanks, but it seems this laptop is not equipped to handle the 880Ms when overclocked. With maxed fans I still hit 89c nearing 90c in Crysis 3 with GPUs at 1.1Ghz. I think Liquid Ultra is in order.

 

I think you're right. Liquid Ultra does great on GPUs. Just be careful of the surface mounted components around the die. Clevo GPUs have a plastic shield covering them, but Dell GPUs do not. For those that have never looked yet, here is what the Dell GPU looks like with no protective shield around the GPU die: http://i.imgur.com/uGX05ZZ.jpg

You can make your own shield with electrical tape. See this example: http://i.imgur.com/pk6K35M.jpg

I ran Liquid Ultra on the 780M SLI in the AW18 GPUs for about six months and then, for no reason but curiosity, removed the heat sinks to see what happened to the electrical tape. I found everything pristine. The tape looked exactly like it did the day I applied it.

The 880M inductors (with blue boxes around them in this photo http://i.imgur.com/3YuCdfF.png) are part of the problem with 880M thermals. They get super crazy hot for some reason. Might want to get some more effective thermal pads in addition to Liquid Ultra.

 

Can you use adhesive tape?

 

I used black electrical tape as shown in the photo. It was recommended by a guy at PowerNotebooks and it worked perfectly. It is designed to tolerate some heat. It is inexpensive, so I would stick with that instead of some other type of adhesive tape since we already know that it works really well. I just cut it into strips the right width to fill that space using a pair of scissors.

 

What thickness ? I do not know if they sell right product here in Norway.

 

Just ordinary plastic electrical tape like you would buy at the hardware store. The thickness does not really matter since there is more than ample space for the tape. I just used some of what I had in my tool box already. Not sure if they make it in a variety of thickness. I have some of the cloth type of electrical tape as well, but using the plastic type seemed like a better idea.

 

Okay wow. Dell took out XMP enable support. And Hynix write-protects all their memory which is a permanent effect. Looks like no memory overclocking allowed on AW18... FAIL

On the bright side 20K GPU score, I could not get my CPU to behave at 4.3Ghz. After retrying to bench I hit a power limit shutdown. Extreme!

http://www.3dmark.com/3dm11/9351745

P15835

 

Yup, that power handling limitation has been the brick wall for the 18 since the day it was born. Your high score is very similar to mine with 780M SLI. Trying harder just makes it shut itself off sooner... an exercise in futility. That's why I always say it is an awesome gaming machine, but not awesome for the benching enthusiast. It sucks when artificial impediments like that block the way.

 

Yeahh I think single 980m on R2 can perform better sometimes if it wants too 3DMark 11 Graphics score 16321 (Single 980m Alienware m18x R2)

http://www.3dmark.com/3dm11/9356686

 

Fire Strike score 10295, Graphics score 12061 (Single 980m Alienware m18x R2)

http://www.3dmark.com/fs/3932054

 

I sure do like this Panther 5. Sweet machine. Just getting acquainted with it while I am waiting for the rest of the parts to get here.

 

I'm starting from scratch, so I have a lot to learn about this new machine, what works and doesn't work. With less than 24 hours of random tinkering, I'm already impressed with my noob results.

 

Even better yet...

 

http://www.3dmark.com/3dm11/9374978

 

Wish I could crack 15k's in this machine.



http://www.3dmark.com/3dm/5753914?

Funny benching though, the fans don't even spin up since until the physics test

 

http://www.3dmark.com/3dmv/5214817

Comparing the 980M SLI in the M18xR2 to single 980M in the Panther 5: http://www.3dmark.com/compare/3dmv/5203867/3dmv/5214817

 

Let's turn it up a bit guys

 

Thank you! The rest of the parts should be delivered by USPS later this afternoon.

 

Hi guys..

Just showing you benchmarks with my cheapo Area 51 motherboard

Disable spread spectrum

 

dlt

 

Does not look cheapo to me, bud. The person that made those comments a week or so ago was misinformed and making stupid assumptions.

Here are a few more single 980M benches before the second GPU goes in later tonight or tomorrow.

​

 

Mr. Fox . . . Did you get those results on the Metro benchmark with the 980M at stock clock?

 

Yes...

Here is a stock (CPU and GPU) run on 3DMark11 with the second GPU installed... pretty sweet for neither one being overclocked.

 

That's pretty damned excellent for a stock run on the Metro Redux benchmark. SSAA in Metro is meant to be a resource hog, and with a single 980M at stock, you've churned out an average of 50FPS.

In fact, the rest of those benchmark scores you have up there with the stock single 980M GPU are actually approaching, matching and/or exceeding scores I've seen on SLi 880m/780m rigs. Were you able to get roughly similar runs on the single GPU when you had the 980Ms in your M18X R2?

 

No, not even close in the M18xR2 with a single GPU and not a lot better with SLI. 980M SLI simply doesn't function correctly in an Alienware.

 

A real shame.

However, getting back to your present performance. I would really like to emulate the other variables that you've got going in addition to the hardware because that's some killer quality performance. In Metro Redux, you're almost a whole 10 FPS ahead of even the desktop GTX 970.

 

I agree the Area-51 motherboard is fine. I had an EVGA P55 FTW in my previous build and while it was an i7 870, I could only get to 4.2ghz @ 1.4v, I could never boot with 4.4ghz or higher even @ 1.5v+. Maybe it was just a bad bin chip, but 4.2ghz on a first gen i7 wasn't too bad either.

I'll push the 5820k up to 4.6ghz+ later on. Happy with temps so far considering it's a single 120mm sealed setup. 4.5ghz looks to be a sweet spot for 24/7. Hopefully I can get 4.7-4.8ghz for benching.

Great to see you have both cards working Mr. Fox, that machine is excellent.

 

http://www.3dmark.com/3dm11/9382713

 

Faster than 780 desktop SLI! Nice!!