Overclock3D.Net :: Print Page :: 3GB, 6GB or 12GB Investigated

3GB, 6GB or 12GB Investigated Page: 1
Introduction

With the advent of a triple channel memory controller on Intel's latest CPU, enthusiasts have been forced to ask a new question: Which memory configuration is best? A simple question that usually provokes more questions and debate than perhaps any other component in a PC system.

Just a couple of years ago it was a simple choice. Get the fastest 2GB kit you can afford. With Windows Vista consuming (or should I say utilising) more memory than XP and 64bit becoming the operating system of choice, 4GB kits were rapidly becoming the norm. DDR2 prices though were and still are, very cheap so a £50 investment was not going to bankrupt anybody anytime soon. This changed however with DDR3 and for the most part, the reasoning behind most people opting to stick with DDR2. The price of DDR3 has lowered slightly but it is still significantly higher than DDR2 and if you want the latest CPU from Intel and want to get the most out of it, a triple DDR3 kit must be on your shopping list. Luckily for us, the prices have dropped enough now that kits can be bought for little more than DDR2 so the damage is not too great.

But what size kit should you buy? 3 GB would be enough for most people surely? What of the rumours that 6GB is now the standard and gives the best performance? Should I go for broke and get 12GB, after all bigger is always better right? Thanks to Corsair, who have kindly provided OC3D with 3 different kits, these are the questions I hope to answer in todays article.

All the kits we intend to use for todays review have the same timings, same frequency and being from Corsair, one of the worlds for most memory module manufacturers, compatability will hopefully not be an issue. The results will hopefully also give a true account of how memory capacity affects system performance. I will be interesed in how memory capacity limits the CPU overclock because with Intels 4 series chipsets anthing over 4GB proved difficult to get stable at high overclocks. With the memory controller now on-board the CPU rather than being controlled separately from the Northbridge, I suspect that the extra heat generated will increase with memory size as more stress is placed on the memory controller and with heat usually goes hand in hand with a decrease in potential overclocks. How much so we will find out later in the review.

Specifications

The specifications for the three kits up for review today were taken directly from Corsair's website.

	TR3X3G1333C9	TR3X6G1333C9	HX3X12G1333C9
Capacity	3GB	6GB	12GB
Speed	1333MHz	1333MHz	1333MHz
Timings	9-9-9-24	9-9-9-24	9-9-9-24
Voltage	1.5v	1.5v	1.5v
Price	£61.07	£94.42	£200.22

As you can see, there's not a great deal between them bar the price and the capacity. Let's hope there's more to distinguish the kits in our run of benchmarks but before we test the kit, let's delve a little deeper in the requirements of triple channel DDR3...

3GB, 6GB or 12GB Investigated Page: 2
Operating System restrictions explained

So you have your shiny new setup and load Windows up for the first time. You check system properties and this is where you come across something that raises more questions than it answers. Windows detects 3GB but you have 6GB installed - no fair! You blame everything, the memory, the motherboard, the BIOS, even Bill Gates but the reality is that nothing is wrong. Nothing except for the limitation of the operating system itself.

Since the advent of the home operating system, memory has seen a dramatic increase in capacity. From the humble 1KB systems of the 1980's to the behemoth multi gigabyte machines we have today. Windows has grown with this increase in memory size and recently, memory has had to play catchup to Windows.. 128MB was the standard for Windows 98, this quadrupled to 512MB for XP and as XP grew older, more and more memory was added to the operating system until we stretched the 32bit limitation with capacities of 4GB or more. For any more memory to be utilised we had to purchase the 64bit version of that operating system which, sadly was not very well supported and those who did upgrade found a plethora of driver issues and incompatibilities that rendered the upgrade a waste of time and effort for many.

Bring on Windows Vista. With Microsoft's latest and greatest OS, both 32bit and 64bit platforms are catered for and both are extremely well supported with very little difference between them, on the exterior anyway.

32bit or 64bit?

Here's where things get a little complex. A 32bit operating system can address 4GB of memory. However, you will also need a graphics card for your setup and everything else that has on-board memory such as a CPU, Hard drives, sound cards etc. You will therefore need to take that from the total amount for Windows to utilise it. MMIO (Memory mapped I/O) is the process by which Windows separates the memory available, memory provided by each device and the process used to communicate between all the different devices. In short if you intend on installing 4GB+ of memory into a 32bit operating system don't because Windows will only 'see' approximately 3GB of it . the 4096MB address space is for the TOTAL system memory not just the memory modules you stick in memory slots.

As luck would have it, the new Intel chipset allows for 3GB of memory to be used in triple channel (3x1GB) which would seem plenty and is still within the boundaries of a 32bit operating system. If you want more than 3GB then you will need an operating system that isn't limited by 32bit technology.

A 64bit operating system totally eradicates this issue as it can address a whopping 8TB of memory, thats right, 8096GB more than any motherboard around at this moment in time can support. Compare this to the 4GB limit of a 32bit operating system and it becomes clear that 64bit is the future. But do you really need this amount today, do you indeed need more than 3GB, would you use that extra capacity?

I think it's time we found out...

3GB, 6GB or 12GB Investigated Page: 3

Test Setup

To ensure that all reviews on Overclock3D are fair, consistent and unbiased, a standard set of hardware and software is used whenever possible during the comparative testing of two or more products. The configuration used in this review can be seen below:

Processor: Intel Core i7 920 (2.66Ghz)

Motherboard: Gigabyte EX58-UD5

Memory:
12GB Corsair XMS3 @ 9-9-9-24 1600MHz
6GB Corsair XMS3 @ 9-9-9-24 1600MHz
3GB Corsair XMS3 @ 9-9-9-24 1600MHz

Graphics Card: NVidia GTX280

Power Supply: Gigabyte Odin 1200W

CPU Cooling: OCZ Gladiator

Hard Disk: Hitachi Deskstar 7K160 7200rpm 80GB

Graphics Drivers: Geforce 182.0 CUDA

Operating System: Microsoft Windows Vista Ultimate x64 SP1

During the testing of the setup above, special care was taken to ensure that the BIOS settings used matched whenever possible. A fresh install of Windows Vista was also used before the benchmarking began, with a full defrag of the hard drive once all the drivers and software were installed, preventing any possible performance issues due to leftover drivers from the previous motherboard installations. For the 3DMark and gaming tests a single card configuration was used.

The testing process was a difficult one to decide upon. On the one hand it would be excellent if I could load up the memory fully, but this has to be consistant across the board for the results to be viable. Sadly, what loads a 12GB kit to 100% will cripple a 3GB kit and make obtaining results from the 3GB kit nigh on impossible (anyone who has had a system run out of memory and become unresponsive will be able to appreciate this!). Much the same in that what loads a 3GB kit will not make the 12GB kit break a sweat, so benchmarking the kits proved very difficult indeed. What I did do was analyse the kits performance thourout the testing and with regard to Photoshop (the bain of my life after this review!) after many attempts I found a happy medium for the image size to be used which stressed all kits without totally locking up the pc. This should also serve as a warning to those who intend on using a 3GB kit for processing large images, expect lots of frustration!

As an example please see the images below which show that at times 12GB was fully stressed but on others barely 6GB was used. Obviously this is run on a Vista setup with nothing installed but the drivers and programs used to benchmark but a system running programs and utilities in the background for everyday use would no doubt use a lot more memory. Consideration should therefore be given to this fact when analysing the results of the benchmarks.

To guarantee a broad range of results, the following benchmark utilities were used:

Synthetic tests
• PassMark Memory test
• SuperPI 1m, 8m, 32m

Image Editing
• Photoshop CS4

Memory Test
• Sisoft Sandra 2009
• Everest 4.60

File Compression & Encoding
• 7-Zip File Compression
• River Past ViMark
• Daniusoft DVD Ripper

3D / Rendering Benchmarks
• Cinebench 10
• 3DMark 05
• 3DMark 06
• 3DMark Vantage

3D Games
• Crysis
• Far Cry 2
• Call of Duty 4

Overall System Performance
• PCMark Vantage

Overclocking

As we are concentrating on a group test for todays review we are not looking at each kits overclocking ability, rather just the effect the capacity has on system overclocking.

Having larger capacities of memory will put a greater amount of stress on the memory controller. Considering that the memory controller is now on-die with Intel's range of i7 Nehelem processors, this could have drastic consequences for those who like to indulge in a little overclocking. Suffice to say, being Overclock3D it would be rude of me not to put this to the test so I initially noted the full timings (Primary and sub) of the largest kit on test (12GB) and set those as the standard timing to be used for all the sizes on test (see right).

These timings were determined by the motherboard used for todays review and as luck would have it, only the Command Rate and Round Trip latency setting required changing with all of the kits to ensure stability.

With these timings and the bandwidth set to 1600MHz, all of the kits allowed the CPU to be overclocked to 4GHz and with a brief 2 hour run of Prime95 26.2, I was satisfied there was a good enough sign of stability there, at least for testing purposes. Any further increase on this and the 12GB kit started to cause some issues and locked windows up requiring a hard reset. The 6GB kit was next to let go at 4.2GHz with the 3GB kit following shortly after at 4.23GHz. Obviously the limit of stability could be determined by each kit's level of overclocking as with the Base Clock frequency being increased, the memory bandwidth also increases. So it's not the most accurate of tests but it does show the overclocks are indeed limited by the capacity of the kits, be it by the memory controller or the memory itself.

To get a nice round number and primarily to ensure we encountered as few bottlenecks as possible, we overclocked all of the kits on test today from their stock settings of 1333MHz to 1600MHz. All of the kits performed at this level without issue. So much for the rumoured 'XXX motherboard will unequivocally not run 12GB at 1600MHz, only 1333MHz' - sorry to disappoint you guys (you know who you are) but 1600MHz works fine from where I am sat.

Please see below the final settings of each individual kit to ensure validity and equality of my testing.

12 GB

6GB

3GB

Let's move on to our suite of benchmarks where I hope to put a few more myths to bed...

3GB, 6GB or 12GB Investigated Page: 4

SuperPI is the benchmark of choice for many overclockers. It's lightweight to download and can give a quick indication on how good a system is at number crunching. Once again, testing was performed a total of 5 times, with an average being calculated from the middle three results.

PassMark is a popular benchmarking suite which test all aspect of PC hardware.The CPU test examines Mathematical operations, compression, encryption, SSE, 3DNow! instructions and more. Each CPU test was performed a total of 5 times, with an average being calculated from the middle three results.

Adobe® Photoshop® CS4 software provides improved access to its unrivalled power through a more intuitive user experience, greater editing freedom, and significant productivity enhancements. For this benchmark we used a variety of filters on a single image and enabled the timing mode, adding the resulting times together to produce a single result. Memory performance settings were set to maximum for the timed runs.

Results Analysis

I was surprised at the SuperPI results given that all the kits were clocked exactly the same. All 3 kits began the same but the longer the test went on, the 6GB and 12GB kits started to pull away. The only logical explanation for this is that there was more room for windows processes to run in the allocation of memory. A lame explanation but the only one I can offer I'm afraid.

More distinct performance difference were to be had with Passmark's Memory performance test, again showing the 12GB kit to be superior. Photoshop however, showed the biggest differences but as with the operating system, the 64bit version is required to fully utilise the 6 and 12GB capacities.

Let's see if there is anything to separate the kits with our run of standard memory benchmarks...

3GB, 6GB or 12GB Investigated Page: 5

SiSoftware Sandra (the System ANalyser, Diagnostic and Reporting Assistant) is an information & diagnostic utility capable of benchmarking the performance of individual components inside a PC. Each of the benchmarks below were run a total of five times with the highest and lowest scores being discarded and an average being calculated from the remaining three.

Everest is in many ways similar to Sisoft Sandra. Focusing mainly on software and hardware information reporting, Everest also comes with a benchmark utility suitable for testing the read, write and latency performance of the memory subsystem. Each of these benchmarks were performed a total of 5 times with the highest and lowest scores being discarded and an average calculated from the remaining three.

Results Analysis

Interestingly, Sisoft Sandra showed some differences between the 3 kits while Everest showed very little. The major jump comes from 3GB to 6GB and while there are still some gains with the 12GB kit, there are diminishing returns once you get past the 6GB mark. Again this was a strange set of results considering the only difference between the three kits is capacity.

Let's see how the boards perform in our next run of benchmarks...

3GB, 6GB or 12GB Investigated Page: 6

ViMark is designed to take the inaccuracies and guesswork out of measuring the time taken to encode video files, ViMark produces easily comparable and consistent results for encoding raw video into Windows Media, Quicktime, and Gif formats. As always, a total of 5 benchmark runs were performed with the highest and lowest scores removed and an average calculated from the remaining 3 scores.

7-Zip is an open source Winzip-style file compression utility that has the ability to compress and decompress many file formats including its own .7z compression scheme. 7-Zip also comes complete with its own benchmarking utility for gauging the compression and decompression speed of the system that it is installed on.

Daniusoft DVD Ripper provides powerful and easy-to-use DVD Ripper software which could ripping DVD to MPEG and almost all video and audio formats format with fast conversion speed. We converted a DVD to AVI format 3 times using this tool and then took the average from those times.

Results Observations

One area in which I expected to see dramatic differences in the capacity of the kits on test was video encoding. ViMark showed there's virtually no difference at all between 3GB, 6GB and 12GB which was hard to swallow. To gain a little more perspective and rather than using a synthetic benchmark, I decided to encode a DVD (StarWars:Attack of the Clones) converting the movie to AVI format. I ran the encoding process three times per kit just to be sure and the results gave a better, more informed picture however, the difference is still not as large as I would have thought.

Let's move on to our 3D Benchmarks...

3GB, 6GB or 12GB Investigated Page: 7

Cinebench 10 is a benchmarking tool based on the powerful 3D software Cinema 4D. The suite uses complex renders to gauge the performance of the entire PC system in both single-core and multi-core modes. Testing was performed a total of 5 times with the highest and lowest results being omitted and an average created from the remaining 3 results.

3DMark is a popular synthetic gaming benchmark used by many gamers and overclockers to gauge the performance of their PC's. All 3DMark runs were performed a total of 5 times with the highest and lowest results being removed and an average calculated from the remaining 3 results. Also included are the CrossfireX results to give an indication of how 8x PCIe lanes perform.

Results Analysis

Throughout the testing of the 3D benchmarks the 12GB kit came out on top but the differences were so small as to be insignificant. Every little helps so Tesco insist on telling us but I'm afraid that the differences are simply not worthwhile discussing...so I'll shut up right there.

If there's not a lot between the kits in the synthetic 3D benchmarks, I do wonder how actual gaming is affected, if at all...

3GB, 6GB or 12GB Investigated Page: 8

Crysis is without doubt one of the most visually stunning and hardware-challenging games to date. By using CrysisBench - a tool developed independently of Crysis - we performed a total of 5 timedemo benchmarks using a GPU-intensive pre-recorded demo. To ensure the most accurate results, the highest and lowest benchmark scores were then removed and an average calculated from the remaining three.

Call of Duty 4 is a stunning DirectX 9.0c based game that really looks awesome and has a very full feature set. With lots of advanced lighting, smoke and water effects, the game has excellent explosions along with fast game play. Using the in-built Call Of Duty features, a 10-minute long game play demo was recorded and replayed on each of the GPU's using the /timedemo command a total of 5 times. The highest and lowest FPS results were then removed, with an average being calculated from the remaining 3 results.

Ubisoft has developed a new engine specifically for Far Cry 2, called Dunia, meaning "world", "earth" or "living" in Parsi. The engine takes advantage of multi-core processors as well as multiple processors and supports DirectX 9 as well as DirectX 10. Running the Far Cry 2 benchmark tool the test was run 5 times with the highest and lowest scores being omitted and the average calculated from the remaining 3.

Results Analysis

As we saw in the synthetic benchmarks, there really is nothing to choose between the 3 kits. Greater capacity adds such a small amount it would be fruitless buying more ram for gaming alone, especially if you already have a 6GB kit. I tried each kit for a night's worth of gaming and while I did not benchmark them, the 3GB kit seemed to lag a little and was noticably slower in level loading. The 6GB and 12GB kits were indistinguishable in game but the 12GB kit did seem to be snappier in the level loading stakes but not as much as the jump from 3GB to 6GB.

Let's take a look at the kits overall performance...

3GB, 6GB or 12GB Investigated Page: 9

PCMark Vantage is the latest benchmarking suite from Futuremark. Differing significantly from their 3DMark suites, PCMark performs a series of benchmarks designed to recreate and benchmark scenarios of a PC being used for everyday tasks. Vantage has a Vista only requirement as it actually relies on several different components from the OS in order to run correctly.

Results Analysis

PCMark Vantage seemed to do what I could not; show a difference between the 3 DDR3 kits. In every benchmark, Vantage showed a performance advantage was to be had with the 12GB kit. While the differences again were not eye popping or huge by any stretch of the imagination, there does appear to be a distinct leap in performance the more capacity there is.

Let's head over to the conclusion where I attempt to put today's testing into perspective...

3GB, 6GB or 12GB Investigated Page: 10
Conclusion

Anyone who has read my reviews previously will know I'm a sucker for new technology, so when Corsair asked if I'd like to review 3 of their XMS kits I jumped at the chance. While the kits are not at the cutting edge of performance (see the Dominator kit for ultimate performance), they are all designed for i7 to run at the same timings and bandwidth so it was the perfect opportunity to answer a question I had posed myself since the advent of 64bit operating system. Does more ram give an increase in performance?

The short answer is yes. But that's an answer that could be misleading as the performance benefits we have shown are so small, so insignificant that it begs the question: Is it really worth it? Again it's down to circumstance. The benchmarks we ran today were of the game and the game alone. MSN, outlook, utilities and all those other little programs that swallow a chunk of memory were not installed on our test rig. The games used were also timedemos, which while giving the most accurate results for comparison purposes (we wouldn't have it any other way), might not be most exemplorary way to test out a 12GB kit. As stated previously, 12GB just felt more fluid overall and after a weekends worth of gaming with the 12GB kit on my own PC which has multiple programs running in the background, I can certainly tell the difference. Games and programs opened and closed like the first day of install, something that is a rarity on my own pc due to everything I have running in the background.

I personally find 3GB to be too little. While it coped with all the benchmarks fine and ran Vista without issue, it did seem a little sluggish at times and I certainly noticed a lot more egg-timers (read 'circles') when running the 3GB kit. This was most evident in the run of Photoshop which brought the system to it's knees. On the opposite side of the scale, 12GB was a dream to use. Vista appeared to load quicker, although this was later proven to be a placebo effect after I timed it. Vista was however, much more flexible to use. Think back to when you first do a fresh install of XP and realise how snappy it is compared to your old install. Jumping from 3GB to 12GB gives the same feeling, as Vista simply felt 'snappier' the more memory I used. Switching from one kit to the other just emphasised the fact, opening documents, loading programs, playing media, editing pictures just felt slick and going back to 3GB gave the impression Vista was struggling, although before using both the 6GB and then 12GB I didn't notice how sluggish it really was.

The benchmarks I ran today certainly do not scream 'buy 12GB now' and for the most part I would agree. 12GB is a ridiculous amount of DDR3 to install in an everyday system, it gives no noticeable performance gains for gaming (unless you are a system hog), little performance difference in what I assumed would be it's biggest selling point, video encoding and limited our overclocks by at least 200MHz. Photoshop showed the biggest gains overall but unless you are dealing with some very big images and applying lots of filters then 6GB would cope easily. It also had an effect in general Vista use but not enough to make me rush out and buy upgrade. Perhaps if one were to change their gaming habits and worry less about the number of open applications then the benefits of gaming with 12GB would be much more apparent.

3GB performed surprisingly well. Although we have shown it to come bottom of the pile in the benchmarks we have run, you certainly would not notice it, at least until you started to edit large files in photoshop, or have the luxury of trying out a higher capacity system. Maybe that's a little harsh but after doubling and then tripling the capacity, the 3GB kit just felt inferior.

So the 6GB kit gives the perfect happy medium. The gains 12GB had over the 6GB kit were minimal overall and this time there was minimal difference in Vista's performance as opposed to the 3GB kit. The 6GB kit did not appear to put anywhere near as much stress on the CPU, given that we achieved 200 MHz more. Once the 6GB threshold is exceeded, performance differences seemed to diminish. There were simply not enough differences between the 2 kits to warrant the extra cash injection needed. Sure, a full rack of memory looks the part and certainly grants those extra E-Peen points people yearn for but the sensible money goes on the 6GB kit, at least until Microsoft decide to munch through more of our memory with a new Operating system.

The 3GB, 6GB and 12GB kits I reviewed today are all available from Ebuyer.

A huge thanks goes to Corsair for providing the three XMS kits for todays article. Please discuss in our forums.