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

Hey Daveh98,

I'm no expert myself however once you see diminishing returns it usually means that is the max output of the GPU scores you will see for your cards. As you know each GPU has its limits when it comes to OCing. However you can play around and get better results by trying out different GPU drivers. As for me the 326.80 beta drivers were quite shocking for benching on my R1 Win7 rig. I would always crash and freeze. So you could start by trying a different set of drivers to try and eek out a few more points. Each GPU driver can also give you a slightly different max OC GPU clocks range. So it takes time to try and figure out what is your max settings.
Also once you have maxed out your GPU clocks you can then start focusing on the CPU OC. However I know in your case it is a little harder since it is locked down atm due to the system bios.

So yeh in most cases most games and benches will benefit more from the GPU Core OC and OCing the mem will only benefit to a certain degree before you see diminishing returns. Now bandwidth will play a key part in the 1080p benching or otherwise known as the Extreme X settings benching. To a degree the 780M with its superior bandwidth and higher clocked RAM has shown to be quite good. However in most cases you will see more score benefit from your GPU Core OC.

Unfortunately in most cases you will need to play the numbers game, trial and error, many hours of testing before you can find that range of whereabouts your max GPU OC clocks are going to be. That is where in most cases GPU driver variations can assist a bit. In most cases the GPU temps are fine so that shouldn't be much of an issue unless of course your pushing 90+ degrees C during gaming or benching. Then a repaste is in order. That is just my take on it so far. The experts will give you a better guide :thumbsup:

Cheers.

 

Same goes for me!

 

Whats the highest stable oc recorded for 780ms in sli and single? ?

 

Overclocking the memory too far is detrimental to the GPU more so than the core. If you take the core too far, the driver will frequently crash and you typically won't have anything bad happening other than instability. Bear in mind that there is no way to monitor the temps on the memory or VRMs (GPU temps are the core) so you should avoid indiscriminately overclocking the memory. You could be slow-cooking the GPUs without even being aware of it happening if you don't use good sense about GPU memory overclocking. The other thing about GPU memory overclocking is that it contributes far less to benchmark scores compared to core clock speed in most cases. Think of it the same as you would system memory. Having 2133 memory compared to 1866 or even 1600 memory with really tight CAS timings has a relatively minor to moderate effect on overall system performance, and it pales in comparison to the effect of high CPU core clock speeds.

Dave, the other thing is, each system is different. Each GPU in the same system is frequently different. Just because John or Pat.D can sustain a certain GPU overclock does not mean that your GPUs will do the same thing. That difference could be their video cards, the M18xR2 those cards are installed, or both.

It looks to me like you are getting a handle on how to do it. With the basics figured out, experimentation, trial and error are going to be required to advance. Go slow with the changes, watch what happens and write down what works best so you can replicate your results again later. Do things in a methodical manner and avoid making multiple changes at once. In other words, don't change drivers, CPU settings, GPU core setting, GPU memory settings and GPU voltage settings all at once and expect it to magically be awesome. I have gotten lucky a couple of times doing that, but those were simply lucky guesses. And, because I did not record what I did, and could not remember, I wasn't able to do it again later without starting over with the trial and error.

 

OK that clears up an assumption that I did have. I was going to start lowering my mem OC as I just don't see a giant reason to crank it up compared to working more closely with the core. I also completely agree about the differences with GPUs. I've always tended to side by the line of caution when overclocking partially for this very reason. When people say they find some clocks that work for benching and then lower clocks for 24/7, I usually start lower than the 24/7 and then work up to see how that works. I did take John's "starting point" of +180/400. I've gone to 190 without any artifacting and under 80* temps so I know I can still move up a bit more. However I have to say I am impressed with the moderate OC'ing I have done so far and want to be mindful of the fact that I have no reason to OC much at this point. Every game is playing to my expectations. I'm impressed that in the GPU intensive benchmarks how drastically the GPU score has gone up with moderate overclocking.
I'm guessing some of my scores (even stock) that have outperformed the 4930 have to do with the powerdraw and the Haswell limitations. Once that gets sorted, I think the benchmarks will really start to favor the 4930 benchers (whenever they finally decide to start benching that is...lol).

 

actually you are able to monitor vram, shader temp with hwinfo iirc. the table shows 3 temperatures for a single GPU.

 

The trouble is, I have not see any detailed information about those readings. They may all be the same thing through a different channel (Dell EC/Compal EC/PCH, etc). Those may or may not be not be vRAM and shader temps, but I know they always seem to be within 1 or 2°C of the same value for me, so that is why I tend to think it's just the GPU core reading from a different source. Do you have any information on that Brother Unityole?

---

Did some messing around with cTDP overclocking mode this evening with a modest 4.3GHz overclock and a semi-aggressive 4.5GHz. It's kind of interesting. It seems like it might possibly run just a little bit cooler than adjusting things manually. I may have to try this more with the Haswell CPU to see if it behaves any differently.

 

How do you have the Intel Dynamic Platform and Thermal Framework Settings shown in the Power Option? Is it there for Win 8 on the AW 18 too? Seems like it's available for the R2 only.

 

I will have to see if I can figure out how to enable this on the 18, although I will not be able to help with the other aspect of your question because I decided to ditch Windows 8 completely on the 18. I have no intention of reinstalling it. Late Sunday afternoon, I came within inches of removing the Windows 8 partition from my R2 as well. I am despising Windows 8 more and more over time. I am approaching the one year mark as a Windows 8 daily user and I like it less now than when I started using it, LOL. The only thing that stopped me from wiping it out on the R2 was not having enough time or desire to mess with it so late in the day.

I recently ventured into changing some settings in the DTPF menu with the unlocked BIOS on the M18xR2 that I had not messed with before. Once I enabled the extra stuff in the BIOS, an Unknown Device appeared in Device Manager without any drivers, and a new device with a problem status appeared. I manually forced driver installations for both new devices (ignoring the bitter objections and ominous warnings from Windows 8 about doing so) and extra menus showed up in my Windows Power Options that were not present before. I do not remember seeing the same option menu in the BIOS on the 18, so it may not be possible. I have not booted into Windows 7 on the R2 yet to investigate what changes might be there. (The above screen shots are working in Windows 8.)

Below is an example of a Cinebench run I did on the 18 (also at 4.3GHz) a while back and the Haswell score was not as good as the Ivy Bridge score. Of course, Cinebench is pretty brutal and, compared to Ivy Bridge, the Haswell CPU tends to get hotter and buckle at the knees when pushed extra hard.

 

perhaps for new alienware 18 or maybe just 680m card behave this way. my 6990m card shaders is about 12C higher than core, and vram is about 7C higher than core.. similar on both card, but different temperatures

 

Well, I finally had enough of these crappy old OCZ Agility 3 SSD. They gave me some fits again this weekend, out of the blue. They are working fine again. Don't know what happened, but they magically healed themselves again for about the 5th or 6th time.

I decided to try something new... almost grabbed some Sammy Pro SSD or EVO, but this SSD has been looking pretty sweet in most reviews. I am curious how the LAMD enterprise-class controller it going to perform, so I grabbed a pair of these instead. Both brand options have a 5 year warranty. It looks like the Sammy's have a performance edge at first, but steady-state performance as the drive space fills up and has more write operations performance goes in favor of the Neutron. I guess we will find out.

I like the fact that the Corsair SSD Toolbox utility works in RAID0 (most do not). Once I have them, I'll do some benching. I was looking at the GTX model and the performance was almost the same except read/write speed and IOPS was reversed.

Now might be a good time for me to retire Windows 8, seeing that in many cases RAID0 TRIM support is available in Windows 7, but sometimes not in Windows 8. I don't understand this. Seems like the newer OS would be the more likely to support RAID0 TRIM, but Windows 8 seems like more of a joke the more I continue using it. I may reinstall it on the tiny little 48GB leftover partition from my mSATA so I can still do some testing with it now and then.

Corsair Neutron Series CSSD-N256GB3-BK 2.5" 256GB SATA III Internal Solid State Drive (SSD)

Performance
Max Sequential Read:	up to 540 MB/s
Max Sequential Write:	up to 450 MB/s
4KB Random Read:	Up to 83,000 IOPS
4KB Random Write:	Up to 95,000 IOPS
MTBF:	2,000,000 hours
Features	High-performance LAMD SSD controller E-boost technology with error correction and advanced NAND management Maintains consistent read and write performance when accessing all types of data Broad PC and Notebook compatibility Compatible with Windows 7, Windows 8, Windows Vista, Windows XP and Linux and MAC systems Supports S.M.A.R.T, TRIM, and internal garbage collection

 

NVIDIA GeForce GTX 780M video card benchmark result - Intel Core i7-3940XM,Alienware M18xR2

 

Awesome score there Champ! Oles :thumbsup:

Cheers.

 

Yup, that is pretty awesome indeed.

 

WOW! That is very good. Reminds me of some scores I saw from another user with a dual 330W AC adapter setup. Nice job.

 

HAHA lol me too :thumbsup:
Fox it looks like we need to turn Super Saiyan and go up another level if we want to compete against this STAR

Cheers.

 

Ha ha... I think you're right. I am in a lower class now with 680M SLI in my M18xR2, but that's OK for now. I'm content for the time being. That gives me the rare opportunity to be in first place on a few things with formidable competition like Johnksss.

 

10 characters

 

I hate to be against the grain, but with those GPU and physics scores I think you'd be much better served cutting back on the CPU OC in order to get a good combined score. That part of the test scored so low I'm sure it pulled your overall score down a cool couple thousand points.

I'm personally jelly about that physics score.

 

Thanks for the suggestion;I'll try it.

 

Hold the horses, I just figured it out I'm sorry, that was bad advice. I've gotten so used to seeing that pattern of scores I immediately assumed it was a 3dmark11 result. Only after noticed FireStrike I hold myself instantly corrected, that's a beyond amazing score.

 

My best ever was 16616,gonna have to try a different driver.

 

Northstar that is an insanely awesome Firestrike score! The 780m SLI really is shining with that unlocked XM.

 

Thanks Dave

 

Yeah Northstar, some monster scores there. You're making me wanna go out and get some 780m's.

 

Ok, that's perverse I need a new job to be able to afford scoring like that And I thought 9.8k was nice.
Man, those overclocked 780m's are almost twice as fast as stock 7970m's.

 

Yeah that GPU score is epic. My gimps barely push 13k

 

On a roll here guys

 

Invalid score

 

Hey Guys,

Could someone with Haswell quad please run a 32MB/1024MB wPrime1.55 in W7 and post the results? Thanks in advance!

 

Here is a wPrime 32M run I did last month... I don't have a wPrime 1024M test right now, but if nobody steps up I'll do one for you.

Spoiler

On another note, got the Corsair Neutron SSD RAID0... es muy bueno! I set it up with 10% over-provision. (Corsair gives the end user the option whether to over-provision or not.)

Here is a before and after snapshot...

 

^^ holy mother nature!!! that's insane for just x46. Wish you had a stock run as well for reference.

I'm also curious how non-xm quads perform there at stock clocks. I'm trying to push my little beast to its limits and so far was only able to overclock BCLK.

 

Here's a random and zero-fill CDM benchmark with the Neutron SSD RAID0. This is a massive improvement over the Agility 3 SSD RAID0.

Compressible versus incompressible is basically equal.



Another (better) AS SSD run...



And a new MaxMEMM2 run... 1330.7 HWBOT MaxMEMM2 Marks

 

Very nice SSD benchies, Bro! - certainly MUCH better than the Agility's that you had before.....I guess it's time to retire those for use in another machine, eh?

Here's a nice SSD bench tool for you to test out: Anvil's Storage Utilities Download Version RC6

 

Still no news of a properly fixed BIOS for the 18 eh?

And by the way, this is the EVO 1TB benched. The drive is 41% filled with mostly incompressible data, so it's not at its peak performance anymore. But then it's about as good as Mr. Fox's RAID 0, and it beats it in write performance, due to all the hanky panky stuff going on. It uses page file to cache the write, and then it uses 12GB of its NAND (different amount of NAND cache on different drive, the 1TB has 12GB, the smaller drives has less) to further cache the write. I believe read is also cached too. It's at the expense of slightly higher RAM and CPU usage. For the most part I don't notice anything at all.

Sustained write performance will drop if you're writing a humongous file that fills up the cache. I am curious as to how the EVO will perform in RAID 0, since it won't have RAPID enabled, because as of now you can only enable RAPID on 1 EVO drive per machine. It will still have the 12 GB NAND cache though. I could re-run the benchmark with RAPID disabled, but I will have to reboot the machine.





6GB/s burst write for seq and 512k. Wooh...I think if I move the test size up from 500MB to 4GB, the performance will drop. But I don't want to run it since it'll exhaust the SSD. For small file size, the speed is more than I will need, unfortunately when writing to the EVO the transfer speed is limited by the read speed of other drives.

This a first world problem.
http://memegenerator.net/instance/41469134

 

unfortunately you will not be able to use rapid mode with evo if you raid two evo together as samsung magician not allowing it for consumer plateform, only for it's enterprise/server it lets you do rapid with raid SSDs. also rapid imo is just gimmick and all the performance increase so far i've seen is all about benchmarks, which aren't for real world usage. using ram as cache to speed up read/write sounds good and all but in all honesty if I were to set up ram disk, and manually move some of the stuff I need and to run it faster it'll be a lot of work, ontop of that program/software will have to be able to do just that, taking advantage of ram. so in the end, I dont think its possible for samsung magician to communicate with all other programs I use other than the benchmark software.

 

Gimmick or not, it remains to be seen. Even by itself, the EVO is one of the fastest drive out there, and it's currently the fastest 1TB SSD one can buy. Enterprise or server drives are usually PCIE, and most of them still use HDD, not SSD.

I think the benefit from RAPID is also dependent on the file type. If you have big incompressible media files, then the RAPID might end up to be slower, since it won't be cached. But if you have games with a lot of random files around, the the RAPID will learn which block to cache, and in that situation RAPID will be faster, not on the first run since nothing is cached yet. The gain in performance will only show in subsequent runs.

RAM has speed easily 10x higher than the fastest SSD now, the only downside with using RAM as cache is the increased latency. Personally, bandwidth versus very slight increase in latency I will take bandwidth first. One easy way to fail the RAPID is to have many many tiny files, choking up the IO. As for what you have said, just FYI, every IO is to the EVO has to go through RAPID, so it will be potentially cached. The performance gain is dependent on data type and the size of the data. Not everything will be cached.

A closer look at RAPID DRAM caching on the Samsung 840 EVO SSD - The Tech Report - Page 1

RAID 0 is fast, but the RAID controller takes time to initialize the drives, and the boot time doesn't get cut down significantly. RAPID doesn't work in cutting down boot time too, since it loads after the OS.

The way I see it, having multiple level of cache is a very smart way to get better speed. Simple idea but ingenious. Now it's the execution that matters, which will only get better over time.

Just like how Alienware could link up all the 3 heatsinks on the 18 to distribute the heat better, simple idea which is not impossible to execute. The only obstacle is the display hinge, but there's workaround for that. I think the OEMs never really expected the Haswell will run this hot, because Intel probably didn't tell them everything about the thermal characteristics of Haswell either. By distributing the heat around, the fan RPM will jump less. It's sporadic enough now. Beyond 75C it starts going nuts.

Back to benchmark, anyone has the benchmark with the new WHQL driver? I will probably run some later on, but as of now since I only have pathetic 1mbps internet I am not inclined to restart the comp often.

 

10 character

 

Here are two Anvil tests...

 

Nice! - I think ASU is a great little utility....

 

@kh90123 I think u might have some misconception about evo. its fast due to two to three level of cache being used, two without rapid, three with rapid. all those benchy numbers you see is the single cell TLC flash, and the rapid mode for more benchy numbers. imo if software don't take advantage of those theres no point in getting it other than storage size.

 

I see, I think what I said doesn't contradict with what you just mentioned. Like I said all IO through the disk will potentially be cached, it's not the software that's running that decides if it'll be cached or not, it's the file type and file size, and the RAPID mode service has to decide based on those. When doing writes to the drive, the process will take up CPU cycles. Just to add to what you said, the benchmark number is not just the TLC, it's a combination of 12GB TLC cache, 1GB RAM cache and the 1GB LPDDR2 DRAM cache.

And I just found out. So you're not exactly correct. Samsung is using the TLC to simulate SLC. It's not SLC. SLC is on a different process node, it wouldn't make much sense cost wise to dedicate another die area to deposit some TLC which is on another process node.

AnandTech | Samsung SSD 840 EVO Review: 120GB, 250GB, 500GB, 750GB & 1TB Models Tested
Samsung TurboWrite Adds Up To 3X Speed to 840 EVO SSDs - 2013 Samsung SSD Global Summit Update | The SSD Review
Samsung SSD 840 Evo Review - TurboWrite, Rapid Mode and Magician | bit-tech.net

Nothing beats this EVO commercial.

 

Yeah, I think the EVO is really awesome, and the pricing is similar to the Neutron as well for the 256GB model. It came down to one or the other for me. The only thing about the Neutron that peaked my interest was the possibility that the LAMD controller might offer a slight advantage over time, with less degradation... maybe. But, it was almost a coin toss in the end. I don't think a guy could go wrong with either SSD. Neutron has a 5-year warranty and that was something I considered, too... typical SSD warranty is 3 years.

I had put up with the Agility 3 for too long and I needed to do something soon. Based on the down my dirty research, these 2 SSD clearly stood out as the latest and greatest options in what I read that had solid supporting data. I often have doubts about most hardware reviews, and I had made a personal decision that whatever I did would not involve a Sandforce controller. Agility 3 was never a great performer compared to many newer SSD alternatives out there, but I got them dirt cheap and it was good enough (still much faster than HDD or SATA 2 SSD) until they started frequently acting goofy on me. I would have preferred to save that money toward a 780M SLI upgrade for the M18xR2, but the Agility 3's started behaving in such a manner that I was concerned about them kicking the bucket and leaving me in an unexpected bind.

I am disappointed about thing, and it was influential in my purchase decision to an extent. The Corsair SSD Toolbox documentation says that it supports RAID0, but that appears to be incorrect. The first thing I investigated was the over-provisioning tool and the manual TRIM tool. The former wasn't a big deal, I simply shrunk my volume to leave unallocated space (10%) for OP, but I am no longer using Windows 8 and was expecting the manual TRIM tool to work as a replacement for that feature in Windows 8. I am going to contact them about correcting the errors in their documentation. It appears nothing at all in the Corsair SSD Toolbox supports RAID0. I am going to mention this when I post my review at NewEgg as well. I don't want other customers to buy on the assumption that RAID0 support is present in the Corsair SSD Toolbox if it is not.

That's a good commercial. Anyone that has never owned SSD will experience a tremendous improvement in performance and that advertisement may effectively lure those that have been held back by fear of the unknown. That is pretty smart marketing, especially showing that gal installing her own SSD.

 

I was actually being somewhat sarcastic with the Samsung commercial. The SSD is great, but the commercial...it's somewhat overly enthusiastic, fine for the Asian (say Korea, China, and Japan) market, but a bit overdone for the Western audience. What's the word for it, tacky?

Also, how's the new WHQL driver? I see no difference in games, but I haven't run the benchmarks yet.

 

Don't know yet on the new NVIDIA driver. Will bench it this weekend and find out. I did not realize you were being factitious about the commercial, but I do think it might be powerful in luring noobs to SSD. I can see a lot of people that are computer illiterate jumping on that.

Here's what I posted on my NewEgg review...

Since nothing is supported on RAID0, that needs to be highlighted. Not a shock since nobody else has an SSD tool that effectively supports it, but misleading the way they say "all functions" without a clear bullet point or footnote indicating "single drive only, not RAID0" IMHO. The TRIM tool does not need to work for Windows 8 since it already has its own manual TRIM tool, and it does support RAID0 TRIM.

 

<laugh> I liked the commercial! I guess I don't see a lot of these so sorta fun to see these tech things. Agreed that any SSD just rocks compared to plain old HD spindles and pretty much anybody would notice massive increases in speed using it.

 

NVIDIA GeForce GTX 780M video card benchmark result - Intel Core i7-3940XM,Alienware M18xR2

 

i still think you got confused of what iwas trying to say. no point having that speedy write cache when windows just going to tell you, you have finished writing these files but if you were to shut computer down, you lose them. example imo is a dumb idea to have this on a usb drive SuperTalent USB 3.0 Express Dram Disk Review - Simplicity With Ultra High Performance | The SSD Review is simply pointless. having the extra speed looks cool but just gimmick.

and as I was saying, evo is fast but thats the SLC cache made from TLC layer you're seeing, and for 1TB of evo 12gb is all you'll ever get, which is probably enough. but do note that the files in the end are stored in TLC flash level, so w/e the TLC performance at storage level you'll never know it's performance, hence all hardware runs along with the slowest one. i mean, if you have a usb 3.0 flash drive able to transfer 300mb/s but only a HDD in your computer, the most you'll get is 100mb-200mb/s depending on the HDD.

no matter how the reviews are, this is what samsung EVO SSD is, it is what it is lol slowest TLC will probably never get tested and imo only getting 750gb+ evo version would be better idea due to it's SLC cache otherwise theres no point. unlike samsung 840 pro, is fast throughout the entire drive, where as evo only speedy for its SLC cache.

however, despite all that and goes way back to what i said b4 if evo is able to take advantage of rapid mode and software / programs we use can run off the RAM cache then only then it'll be amazingly fast. until i get confirmation of this, evo is just a bit better than 840.

 

I know I know, it's not an M18 exactly, but I did a little experiment with the 780M and Vram clocks and had the following results which I think are relevant:

These are at 967mhz core clocks.

1500mhz memory 3 runs Metro 2033 very high with 4xAA/16xAF = 43.00 Average FPS 11.32 minimum.
1650mhz memory 3 runs Metro 2033 very high with 4xAA/16xAF = 45.33 Average FPS 12.04 minimum.

= 5.4% increase to average frame rates and 6.3% increase to minimum frame rates (which were consistent among the 3 runs)

So yes, the FPS still scales nicely from 1500mhz (6ghz data rate) to 1650mhz (6.6ghz data rate).

 

Hey guys, in case you did not notice it, the final version of Catzilla benchmark was released today.

[ DOWNLOAD ]