Techgage logo

4GB Trials and Tribulations

Date: May 25, 2006
Author(s): Rob Williams

Have you ever wondered what it would be like to move on up to 4GB of system ram? There may not be a need for that now, but it won’t take very long until it’s commonplace in enthusiast PC’s. I take 4GB worth of top end DDR1 and DDR2 memory, and relate my overclocking and performance reports to you.



Introduction


It wasn’t too long ago that having 1GB worth of total system memory was more than enough for anything you wanted to do. In fact, the majority of people still use 1GB or less, and are content. When a few games were released though, such as FEAR and Battlefield II, people started to see clear benefits of moving up to 2GB worth of ram in their system. Those two games specifically can stress all areas of your system, including GPU, CPU and of course your system memory.

This article is not a review of any 4GB kit, but rather an article explaining the pros and cons of moving up to 4GB. I have spent a lot of time overclocking both DDR1 and DDR2 4GB configurations, so you can expect in-depth performance reports later.

Generally speaking, the better the graphics and higher the resolution a game has, the harder it will be on your ram. I found this out last year when I was conducting tests with the FEAR demo. I found, that simply going from 1GB to 2GB released a lot of stress overall on the system, and gave me far better overall FPS. The same results were found about Battlefield II, but I never personally performed any such tests.

While many people are still using 1GB kits in their PC’s, 2GB’s are catching on at a very rapid pace. Speaking with one enthusiast memory manufacturer recently, they told me that they now sell *more* 2GB kits than anything else, so the transition is really in full effect. Of course, even with an Ultra High-End system consisting of a top end CPU and Quad-SLi doesn’t really need anything more than 2GB. There just isn’t a need for it, if you are focusing on one application at a time, such as a game.

There’s a certain mystique that some people have about having 4GB worth of ram in their computer. No doubt, there are indeed actual needs for this much memory, but most of these reasons usually have to do with multimedia production. Graphic designers, movie editors and even large Photoshop projects can hog all of your memory, and cry for more.

There are also countless myths, or misunderstanding when it comes to having 4GB in your computer. The first is the fact that the more memory you have in your computer, the slower it will be, in terms of bandwidth. In fact, this is both true and false, depending on your motherboard and CPU configuration. We’ll get into that shortly.

One of the most popular myths is that Windows cannot physically see anything above 3GB. Again, this is also true and false. Due to technical ‘limitations’ or ‘fixes’ in Windows XP Home and Professional, the Service Pack 2 removed the ability for Windows to see a full 4GB. If you have Windows XP Home/Professional with SP1 still installed, then you should have no issues in seeing the full 4GB. If you have a current install of Windows x64 Professional of Windows Server 2003, then you will see the full 4GB regardless, even with the latest service pack. If for some reason you still don’t see 4GB in Windows, it may be due to your motherboard or an option in your BIOS, such as PCI MMIO.

So the ultimate fact is, if you want to see a full 4GB in Windows, you will need either Windows x64 Professional or a standard copy of Home/Professional that has not been updated to SP2. You could also use Server 2003, if that’s the version you prefer.



4GB Technical Limitations


Depending on your system, especially your motherboard and CPU, upgrading to 4GB of system memory could be slower, or just as fast as 2GB. Throughout the article, we are considering 4GB to be (4) 1GB modules, or 4 * 1GB. While 2 * 2GB modules exist, they are scarce, expensive, and slow. 2GB modules are best suited for servers, because most are buffered, which is why they are so expensive.

The performance increase or decrease will depend on whether you have an AMD or Intel. For my benchmarking tests, I will be using 4 sticks of PC2-6400 memory and 4 sticks of PC-3500 memory. Now that AMD AM2 has finally dropped, the DDR2 portion of the performance results may prove most interesting, since both the Intel DDR2 and AMD DDR2 are rather similar. One thing to bare in mind, is that as soon as you go from 2 * 1GB to 4 * 1GB, you are in effect going from 1T to 2T timings, which are latency related. 2T could prove up to 20% slower than 1T, which is why many systems will indeed suffer memory bandwidth/latency slowdowns, especially in synthetic benchmarks.

That’s where 2 * 2GB will be beneficial, because even though it’s still 4GB, it should still able to handle the 1T timings. I won’t get into how 1T and 2T timings work, because that would be an entirely separate article. To put it simply though, 1T is harder on the motherboard and memory controller, so having a 4 * 1GB configuration at 1T timings is far too taxing for any current motherboard and CPU at higher MHz, regardless of how high end it is. If you wanted to decrease the memory speeds so far in order to hit 1T with 4 1GB sticks, you may as well haul your 386 PC out of the closet.

Simply put, if you have an AMD and are going to 4 * 1GB, you will experience slower speeds than you would sticking to a 2 * 1GB configuration. With Intel’s, it could be either way. If you are using an nForce 4 Chipset, meaning you can do SLi on your Intel, then you will be taking the same route as DDR1 on the AMD. The nF4 Intel chipset allows you to use 1T timings for a 2 * 1GB configuration, and 2T for 4 * 1GB, so you will again experience slowdown with the latter config.

If you happen to have an Intel chipset instead, such as the one on my ASUS P5WD2-E, then you will be using 2T regardless of what your configuration is. This is one reason why DDR2 seems slower in some cases than DDR1 when compared, because even a 2 * 1GB config on an Intel chipset uses 2T. Why this is the case, I have no idea. What this does in turn mean though, that moving to 4 * 1GB on the same chipset will keep the 2T timings, so the performance decrease should prove minimal… if any.

Note, that the descriptions above also apply for things that don’t use all a full 2GB.

On an extra note though, even if your PC can handle DDR2 memory at 1T, chances are you will not be able to retain that 1T timing above DDR2-800 speeds. At that point, the strain is becoming increasingly painful on your memory controller. Therefore, for either a 2 * 1GB or 4 * 1GB configuration with faster than DDR2-800 speeds, the performance should not change at all.

As mentioned earlier, this is where people will actually benefit from having (2) 2GB modules, because they will retain 1T timings. The primary reason why this is not good for the average consumer, is because they are slower. I believe they max out at PC3200 speeds, and many come in ECC flavor, but not all. This means you could expect to pay an upwards of $800US for an low speed 4GB kit.

Without a doubt though, engineers are working hard to produce chips for 2GB modules that don’t suffer from a huge lack of speed that the current ones do. Ryan Peterson, CEO of OCZ Technology, was recently quoted as saying that they would be releasing 4GB kits consisting of 2 * 2GB, later in the year. The expected launch will be shortly before Vista, because that OS will prove to be system intensive. Considering Vista can take an upwards of 500MB of your ram on a fresh boot, that leaves you 1.5GB for your other app’s and a high end game. Not exactly that much breathing room, depending on what you are trying to do.

As soon at these 4GB kits are available, chances are that they will be superior to the speeds we are seeing from current 4GB kits. I am doubtful we will see the ultra high-end speeds that we do right now, but I believe 2 * 2GB kits with PC4000 and PC2-6400 speeds should be very possible.



General OC’ing with 4GB

Most computer enthusiasts enjoy knowing their computers are running at the fastest speed possible. This of course means overclocking the CPU, GPU and memory is on the top of their lists. The question is, is overclocking 4 * 1GB the same as 2 * 1GB? The answer is, yes and no. There have been a lot of yes and no answers today :)

Overclocking the memory with each configuration is exactly the same, the process at least. But, say for instance you have a DDR500 2 * 1GB kit, that easily overclocks to DDR550 speeds. Can you expect to have the same DDR550 speeds in a 4 * 1GB configuration? This again depends entirely on the modules, and your memory controller. Your motherboard can play a huge role also, though.

The first problem is whether the memory controller can handle overclocking four modules at once. Simply, it may prove to be too much strain, so you could be forced with stock speeds. You will have better success with a higher quality motherboard and CPU, but still nothing is guaranteed.

The problem is harder if you want to overclock two different types of memory. For instance, there are some out there who may currently have a 2 * 1GB kit in their computer, and find another kit on eBay for a great price, so they snatch it up. If all four modules are *identical*, then your chances of a decent or good overclock are great, as long as your memory controller can keep up. If you have two different 2GB kits though, the chances of a good overclock are around 25%.

First thing to take into consideration, is that two different kits of memory may have completely different chips. One may use Infineon, the other Micron. Because of the differences, your overclock probability is not as great as if the chips were all the same. Let’s take another example… you have two separate kits that both use Micron 5b F chips, but one kit is by OCZ and the other Corsair. This is completely a random analogy, because Corsair hasn’t released anything with these chips, so this is merely an example.

It’s up to the memory manufacturer to decide on SPD timings, which are the ‘base’ timings and configuration on the chips. One of the kits may have an SPD timings of 200MHz at 2.5-3-3-8, while the other 200MHz at 3-3-3-8. These minor differences can make a huge difference in whether you will have a successful overclock or not.

So, the moral of this story is that in order to have the *best* possible chance of a good 4 * 1GB overclock is to have 4 identical sticks. If you are a poor sap with 2 different kits as mentioned in the previous example, steps can be taken to manually change the SPD timings on one kit to match the other. This is currently only possible on DDR1, unless you have the hardware to change the timings on DDR2, which you don’t ;)

Manually changing SPD’s is dangerous though, so I am not going to give even a simple how-to on it.



Testing Procedures, DDR2 Overclocking

On this page, I will explain how our testing procedures go. Firstly, the hard drive that hosts the operating system is fully defragged using Diskeeper 10 to assure no hiccups will be due to the OS. All testing occurs on a standalone OS installation that’s dedicated to testing. Second, all unnecessary programs in Windows are closed down, minus the essential ones. This includes virus scanners, firewalls, peripheral software, etcetera. None of that is generally installed in the first place though. Before any testing occurs, the memory goes through a 4 hour MemTest run at stock settings to assure that they are error free and appropriate for overclocking and benching.

To test out the modules, I used the usual variety of benchmarking tools, including Half-Life 2 and Battlefield 2. For memory specific benchmarking, we use:

NOTE: What you gleem from these benchmarks should be different than the usual. We are not benchmarking with these games and tools to see how well they benefit using 4GB of memory. What these tests are supposed to prove though, is whether or not 4 * 1GB will *hurt* your overall performance more than help.

DDR2 Overclocking

Because I am using two different architectures here for overclocking, I am going to split them up. You can read through the next few pages for both performance reports, or just read the DDR2 or DDR1, depending on what your preferred flavor is.

The DDR2 sticks are developed by Super Talent and roll in with PC2-6400 speeds, which the very tight timings of 4-4-3-8. In our review, these proved to be extremely overclocking friendly, so paired with another kit, I am hoping for the same results. The chips that these modules use are Micron D9GKX. Not to be confused with the legendary D9 Fatbodies, these chips are some of the most -amazing- on the market right now.

My immediate goal was to see first if I could overclock the 4 * 1GB configuration to the same speeds and timings as posted in our review. I quickly found out that I could indeed overclock these just as well as I could with the 2 * 1GB setup. I didn’t test each and every setting from the review, but mainly four that I feel scales well up to our top overclock.

“But Rob, how come the 4GB solution overclocked as well as the 2GB solution?” Your results may -not- be the same as mine. My ASUS board is pretty high-end, so that may have had a large part in the successful overclock. This shows though, that it *is* clearly possible to OC 4GB as high as 2GB. On top of that, since both config’s use 2T timings due to the Intel chipset, performance drops should be minimal.

I had wanted to go to at least FSB300 for speeds of 4.2GHz, but my dual core CPU has a core that’s not as overclockable as the other. It’s not only that one core has a harder time to overclock to 4.2GHz and be stable, it has a hard time booting into Windows x64 at those speeds, even though it will boot into Windows XP no problem. This leads me to believe that the second core cannot overclock to 4.2GHz primarily due to an issue with it’s 64-Bit part.

The above settings proved 100% stable, so I am pleased with that.



DDR2 – Sandra, Everest

So let’s get started with some good Sandra benchmarks. The 2005 version was used, since 2007 was not available when these benchmarks were conducted. The differences between versions for this specific test seem to be minimal though, if any.

At absolute stock speeds, 2 * 1GB proves 100MB/s faster, or 1.7%. As the speeds and overclock increases though, the performance evens out a lot more. At the top overclock of 467MHz, they are really neck and neck.

My favorite benchmark to conduct is Unbuffered, because it stresses the memory far more than the CPU. Buffered uses all the features available on the CPU to affect the score greatly. Once again, the same holds true from the buffered tests. There is a decrease, there’s no denying that, but it’s definitely minimal. Many people consider unbuffered to mean different things, so you can check out this picture to see how I configure it.

Everest 2.22

The Sandra scores were close, but the Everest results are even closer. Take a look at the max overclock Read result.. 0.15% difference. It’s becoming evident that 2T in both configurations makes no extreme difference, but rather a very minor one.

The Latency tests are where we can see the biggest differences. Each 4GB result is at least 1ns slower than the 2GB one. Overall though, I had expected to see much larger differences than this.

Sciencemark has recently become one of my favorite memory benchmarking tools, and for good reason. Let’s jump right into that one on the next page.



DDR2 – Sciencemark 2, Super Pi, PC Mark 05

Sciencemark 2 is a great tool, because it gives you results that are far more in-depth than anything else out there. I will only relay the Bandwidth and Sciencemarks results though.

What we have seen up to now still applies here… the 4GB solution is just a wee bit slower than the 2GB. What’s interesting is the max overclock here though, because at 4GB, it proved to be 214MB/s faster. The Sciencemarks result for the same benchmark proved lower than 2GB though, probably due to the 4GB having slightly higher latencies.

Most people who want to set Super Pi records usually use a 1GB kit with ultra-tight latencies, and this is why. As we moved from 2GB to 4GB, the overall times were definitely higher. Not by a huge margin, but are higher nonetheless.

Once again looking at the top overclock, 4GB proved just 100 points lower than the 2GB. 2T timings are not looking so bad ;)

For the last of our DDR2 testing, let’s check out some 3D app’s.



DDR2 – 3D Mark 01/06, HL2/BF2

How could there be a benchmarking focused article without some Futuremark action? There can’t be. So, here’s the obligatory 3D Mark 01 and 06 results. Note that the score results for 3D Mark 01 have decimal places, but the real score does not. The decimal was added so that the graph would scale properly. Now, there wasn’t a huge difference in 01 per se, but 4GB proved slower throughout each overclock. What’s amazing though is the 06 results.. they are incredibly close. The biggest differential was 4 points overall.

To test out gaming between the configurations, I used Half-Life 2 and Battlefield 2. Half-Life 2 is not ram intensive, so the primary goal was to really see if more ram would hurt the overall FPS. BF2 on the other hand was responsible for many people running out to pick up 2GB kits, so it’s results are worth noting.

The level used in HL2 was d1_canals_07, and BF2 was Dragon Valley. Results were saved using Fraps. All of the runs were manual with no automation software. Because of each run being manual, each one can throw different things at us. When I play through each level, over and over, I try to play the level as if I am playing it for the first time, to get more realistic results.

So the question really is.. on an Intel Chipset, does going from 2GB to 4GB hurt gaming performance? It sure doesn’t look it.

Whew, that was a lot of DDR2 testing. If you can keep up, let’s jump right into some hardcore DDR1 action.



DDR1 – Overclocking

While both the 2GB and 4GB configuration on Intel both used 2T timings, we didn’t originally expect the performance to decrease by a huge margin. That proved true, with only slight differences between each test. The latency hit was likely the worst result, but that was still minor. In our DDR1 testing, 2GB uses 1T and 4GB uses 2T, which means that we will be seeing some differences. Once again though, it will be a matter of how much difference to see if upgrading is a waste of time.

The modules I will be using for testing are the Corsair 2GB 3500LL PRO. These proved to be a favorite last year, and still sell like hotcake’s. Throughout out performance tests, you will see why. Despite them ‘only’ being rated for DDR438 speeds, they include the tight timings of 2-3-2-6. Why they prove popular though, is that the Infineon BE-5 chips that they use allow them to overclock extremely well.

Surprisingly, I found that these overclocked at 4GB just as well as the Super Talent did in our DDR2 testing. I immediately tried out 275MHz, and that was completely stable.. I was impressed. After a long night of tweaking, the modules were set at 284MHz with 3-3-2-0 timings, and were Prime and 3D stable. I have been using the PC at these speeds for the past week.

As I mentioned earlier, your results may vary from mine. Overclocking success depends on your hardware and the luck of the draw when it comes to how good the chips on your modules are. As my friend Ozzimark puts it, “In the overclocking game, he who tries the most things [tweaking wise], win.” It’s very true in this case.

I was -very- impressed that these modules worked at 284MHz in a 4GB configuration. That speed is rare in a 2GB config, let alone a 4GB one. Because they OC’d so high, I was extra anxious to get on testing to see how hard the 2T timings would hurt the results.

Without further ado, let’s get right into it.



DDR1 – Sandra, Everest

Sandra needs no introduction. These benchmarks were done using version 2007 of the software. These graphs are not easy on the eyes, I admit. The easiest way to compare the results is to look at them in pairs of two. Each 2GB vs 4GB result is directly beside each other, so you can see the differences easy by looking at the bars.

If you look back to the DDR2 results, all the bars were pretty well leveled out. Not here! The difference between 1T and 2T is very evident, with the many dips throughout. Paying attention to the 275MHz, which should prove completely stable for most anyone, the 4GB method was around 17% slower, bandwidth wise. That’s not exactly a small difference.

Looking at the same speed for the Unbuffered test, once again the 4GB config was 17% slower than the 2GB config. I was originally speculating that the differences would be around 10%, but it’s almost double that!

Everest 2.22 and 2.80

I benchmarked using two versions of EVEREST, because I trust 2.22 more than any other version. 2.80 is used a lot throughout enthusiast tests though, so those results are also there if you prefer them.

Overall, the results are tighter than the Sandra ones, but the big hit is in the Write speed. 2T has an apparent drastic affect on it…

… especially with 2.80!

While a 1ns hit wasn’t too bad with DDR2, this is getting asinine! Stock speeds prove 8ns slower, while the top overclock proves ‘only’ 5ns slower. To be frank, that is not a small difference.. but huge!

Let’s see if anything get’s any better with Sciencemark and some gaming tests.



DDR1 – Sciencemark 2, Super Pi, 3D Mark 01, HL2/BF2

So far in our DDR1 testing, results have been quite congruent. 2T timings affects performance a lot, as is easily seen simply by eyeing a graph. Taking the stock speed results from Sciencemark, 2T results made a 21.4% difference! Talk about cutting an easy 1/5th of the performance straight out of your memory. Things got better at the top OC though. 2T there only was 20.8% slower ;)

Whew, this is depressing! Luckily, 2T didn’t seem to affect Super Pi results hardly at all. Our DDR2 results proved 1 second slower on average when moving to 4GB, and it’s pretty much the same situation here.

3D Mark 01 is very CPU intensive, but memory can also play a part, as you can see by the results. Overall, there’s been nothing more than really a 10% difference, which is fair compared to a few of the previous benchmark tests.

You may wonder why one result is blank. Please read the conclusion where I will tackle that issue more.

As with our DDR2 gaming tests, the same applies here. Results were recorded with FRAPS and were all manual play through’s. Half-Life 2 tests used the d1_canals_07 level, and BFII was the Dragon Valley.

Happily, moving on up to 2T timings didn’t really hurt the gaming results at all. Looking at the first HL2 result, there was a huge leap in overall FPS, but that could be due to different circumstances in that run-through. Overall though, 4GB didn’t hurt gaming at -all-, which is very reassuring after looking at the previous sets of results.

Ok! That was a LOT of benchmarking. I think I’m about ready for a conclusion!



Conclusions

Without a doubt, this is a lot of information to take in. Lots of late night benchmarking sessions happened, and I learned a lot about overclocking with 4GB. I had originally thought that it would be far more difficult to reach the same frequencies when bumping the system ram up to 4GB. However, in both sets of memory we tested, they were able to use the same overclock that the 2 * 1GB config could handle.

There was one exception though, and that was with the DDR1 kit, the Corsair 3500LL PRO. If you looked at the results for the 3D Mark 01 tests, you would have noticed one bar was missing. That’s because that test would not run, and caused the computer to reboot. Here’s where I found things to be very strange. The test would run without a hitch with 4GB at the same settings, but not so with 2GB. I figured that this must be due to the chips wanting 2T timings regardless at those speeds, but that proved false.

With more tweaking, I could likely get the 2GB overclock to run 3D Mark 01 just fine, but it will require tweaking. That’s a fact with overclocking memory in general though. All memory is different, and your motherboard and CPU can greatly affect the experience you will have.

For example, on the DDR2 side of things, I didn’t have to change anything to get either configuration stable at the max overclock. But with DDR1, I had to change the drive strength, data drive strength and bypass max. This in turn helped loosen the strain on the memory controller and allowed that OC to be possible. So without a doubt, if you plan on snagging 4GB and want to overclock it, you must be prepared to tweak. It’s not a simple thing.. and if it is.. you got lucky.

One question that this article was meant to answer was, “Is 4GB for me?” Obviously only you know if you need 4GB. The majority out there don’t at this point in time.. plain and simple. However as mentioned in the outline, with Vista around the corner, [or two or three with how things are going], people who care about performance will be required to have 4GB of memory.

As found out in tests run by enthusiasts, if you currently use 2GB for your system, then you will want 4GB when Vista rolls around. Due to the fact that Vista can eat at least 500MB after a fresh boot, that only leaves 1.5GB of free system memory. By that time also, games may even be more ram intensive than they are now, which will require an upgrade even more so. Since having 3GB of system memory doesn’t make much sense for dual channel reasons, 4GB is the only solution.

Both the DDR1 and DDR2 memory I used will cost you around $600US for 4GB worth, or $300 per kit. Though this seems expensive, it’s really not. Prices are going down steadily, and that’s a good thing. Hopefully by the time Vista is released, they will be even cheaper. Less than two years ago, my friend purchased 4GB worth of Corsair memory, which cost him over $1,200. In only two years time, the price halved. That’s the direction I like to see.

Though I didn’t get into discussions about each kit I used specifically, since this is not really a review, I highly recommend either kit to anybody who wants either a 2GB or 4GB configuration. Both kits have been the absolute best I have used on either architecture, which is why I chose to use them for this article. Not too many DDR1 kits will do 284MHZ 3-3-2-0 stable, that’s for sure. If you are interested in the Super Talent kit, the only place to currently pick it up is at eWiz, where it currently sells for $286US.

The coming year will be very interesting. Technology is moving faster than ever before, and it’s hard to keep on top of it all. By the end of the year, 4GB kits may prove to be normal.. and just not for multimedia development PC’s. As mentioned earlier, memory companies are readily working on 4GB kits (2*2GB), and the same goes for the memory controller guys. By the time an absolute need for 4GB is apparent, the performance should prove better than what we’ve found today. At least we can hope, especially after those DDR1 2T results ;)

Discuss in our forums

If you wish to discuss this article in our forums, feel free to jump into the related thread. You do not have to register in order to reply to our content threads.


Copyright © 2005-2019 Techgage Networks Inc. - All Rights Reserved.