Date: May 9, 2007
Author(s): Rob Williams
Mushkin is the only memory company right now to offer 4GB extreme performance kits, with their top-end offering on our test bench today. Despite the fact that it already has high stock speeds, we were surprised to find out that it still had a lot of overclocking headroom.
Not long ago, if you bragged to someone about the 4GB of RAM installed in your PC, they’d call you crazy. The fact is, unless you are heavy into multi-media development or other workstation intensive applications, 2GB of RAM was more than enough. Even today, playing the latest games at the max resolutions won’t top out your 2GB of memory.
As time passes though, our needs for memory (and other components for that matter) increase. In late 2005, most games ran well on 512MB, although 1GB of RAM was the standard in a gamers rig. It was with the launch of FEAR that I realized how much benefit an increase in RAM could prove beneficial. Back then, I found that going from 1GB to 2GB increased the average FPS considerably.
I might be focusing a lot on gamers, because quite frankly, those are the primary audience who care about faster memory. If you are running a workstation, more RAM is beneficial in itself… it doesn’t need to be a sky-high frequency. Up until now though, there has not really been such thing as a performance 4GB kit, despite so many manufacturers touting their kits as such. PC2-6400 with 5-5-5 timings is far from “performance” as far as I am concerned, considering some of their 2GB kit counter-parts could run at 3-3-3 with some healthy voltage.
That’s where Mushkin has really been raising the bar. They are the only memory manufacturer producing performance 4GB kits, the PC2-6400 4-3-3 and PC2-8500 5-4-4, which we are taking a look at today. What comes with the performance is a high price-tag however. There is simply no competition at this level right now, not from a single company. That said, their PC2-6400 kit retails for $470 on average while the kit we are looking at today averages at $699. It goes without saying, these are for performance die hards with a deep wallet.
Years ago, Mushkin packaging wasn’t the most eye-pleasing. However, last summer they had a total redesign and it looks fantastic… clean and professional. They don’t go overboard with the pizazz, but instead let the performance of the modules speak for themselves.
Mushkin modules have some of the best headspreaders on the market, in my opinion. Not so much for their cooling ability, but rather their looks. Their XP (Xtreme Performance) modules are wrapped in a black spreader, and it’s only too bad that the PCB was not also black! Why is it so difficult to find black memory modules outside of Crucial?
The stickers include more information than most other companies will give, including the required voltages. I believe that voltages should be listed on all modules on the market. To me, it just makes sense. Without it listed there, people are left to guess, which normally means they will push more into them than is needed.
On the opposite end, you’ll find Mushkins sharp looking logo.
FrostByte spreaders are known most notably for their ringed top, allowing a lot of airflow across the tips of the modules.
With a look of the modules out of the way, we can jump right into the overclocking ability.
Before we go further, I’d like to clear up some concerns regarding 4GB+ of RAM. The first thing to take into consideration is that you will need a 64-Bit OS, whether it be XP x64 Pro, XP Vista 64-Bit or some 64-Bit Linux distribution. I will be covering the reasons in an upcoming article, but simply put, 32-Bit operating systems cannot properly allocate all 4GB of ram. Even with 4GB installed, you will likely only see 3GB. Some lucky folks have been able to run in PAE mode to get closer to 4GB, but the OS will never address all of it.
That said, you really have no choice but to get a 64-Bit OS unless you want to waste cash. If you don’t want to upgrade to a 64-Bit OS, you could stick to a 3x1GB setup on your 32-Bit rig, which would still be better than 2GB if you are a big multi-tasker.
64-Bit opens up a new world as far as allocation to a given application as well. In WinXP, no more than 3GB can be allocated to an application at a given time, but on a 64-Bit OS, there is no such limit (that I have seen). I have had 3DS Max use close to 7GB in the past.
Overclocking modules of a high-density always proves an interesting experience. The lower the density, the better the overclock should be, although the performance at the same specs as a larger module decreases thanks to the lack of that density. 512MB sticks sometimes clock better than most 1GB modules and the same applies with 2GB modules as we see today.
Browsing through any e-tailer at their 4GB kits gives off the idea that performance and 4GB just don’t go hand in hand. Although PC2-6400 is market standard now, it’s not “performance” by most enthusiast standards. Especially with 5-5-5 timings. In fact, the only company to really throw 4GB performance kits into the mix has been Mushkin. I am unsure why nobody else has released similar kits, but Mushkin has been sitting pretty in that spot for a couple months now.
I need to praise Mushkin just for the fact that they managed to release a kit of these speeds. Either they scored an amazing batch of chips, or they are determined to be the best when it comes to 2GB modules. Many times, reviewers will receive “cherry picked” modules from the companies, so that they know their performance will be amazing. In the case of this kit, every kit is cherry picked. They have to be. You cannot just pick up a batch of chips and assume they will do DDR2-1066 speeds. This is why we are not seeing other companies release performance kits as such, I assume. Too much time is required, which is why the kits retail for a premium.
That said, how much further could 533MHz 2GB modules possibly be pushed? I was surprised to find out, quite a bit. Here are the final OC’d settings:
There you have it, a 2x2GB kit at DDR2-1200 speeds. Granted, this is with 2.5v, but the fact of the matter is, this was a completely stable setting as you can see below.
One interesting thing to note is that for DDR2-1000 speeds and higher, I needed to use a CAS latency of 5. All of my 2GB performance kits will handle that speed and CL 4 without an issue at all. Something about these chips makes that a non-possibility though. Even while pushing 2.5v into them with CL4, the computer would simply not boot up.
Also, the stock recommended voltages for the kit is 2.3v, however I found 2.2v to be fully stable throughout all of my tests. Upon clocking any higher though, 2.4v becomes a must, as you will see with our 575MHz setting. Luckily, bumping it up to 600MHz required only a slight boost in voltage, resting at 2.5v. I was lucky, as this is the max voltage that the eVGA 680i board will allow.
While on the subject of 680i, I have to stress that such a major overclock might only be possible if you own a 680i, if possible at all. Even though 600MHz was stable, I have a hard time believing that most kits out there will handle these speeds reliably. I could be wrong, but I have to remain skeptical since I have not seen anyone else accomplish an overclock like this, on any other 2x2GB kit. I’ve seen others hit 600MHz, but with a lot more voltage and a lot less stability.
That said, regardless of your top overclock, it should prove much higher than the stock setting of 533MHz. Also, despite 600MHz being stable here, I would not recommend it be used as a 24/7 setting simply because 2.5v is too much voltage. 2.3v and under is far safer and will prolong the life of your sticks.
Throughout all of our benchmarks regardless of what we are reviewing, testing is done in a clean and stand-alone version of Windows XP Professional with SP2. Prior to testing, these conditions are met:
The testing rig used for today’s benchmarking is as follows:
Sandra is always the first benchmark to come to mind when we need to do memory benchmarking. Or CPU benchmarking. Or storage benchmarking. You get the idea. It’s a superb all-around tool that we rely on quite often.
Because this is a 4GB kit, I expected to see slight decreases when compared to the same settings with 2GB kits, but it wasn’t the case at all. We actually see some very small increases, except at lower frequencies thanks to the CL5 that we were forced to use.
Next on the list is Everest 3.5, with it’s read/write and latency tests. We broke through the 8K mark in our read tests and came close to 5,500 in our write. These benchmarks are heavily CPU bound as well though, so the higher your CPU frequency, the better effects your memory overclocks will have.
Again, at our 575MHz setting, the Mushkin kit proved slightly higher Read results than our previously reviewed OCZ Reaper 9200 at the same timings and voltages. Write scales with FSB on Intel systems, so the results will be the same regardless of the memory kit.
Latency is where we begin to see the small disadvantages of moving up to 4GB of ram. While the previously mentioned OCZ Reaper kit hit 57.4ns at 575MHz, this 4GB kit hit 58.2ns. I had originally expected much larger differences than this however, so I am impressed by the results.
This is not a memory benchmark per se, but rather one that stresses a single core of your CPU to it’s full potential. Because it crunches such an insane amount of digits, tighter timings and faster memory generally offers better results. We choose to run with an 8 million test, as anything lower flies by too fast on a Core 2 Duo and it’s hard to generally see the differences that way.
Again, great results displayed here. Comparing to the OCZ Reaper kit again, the Mushkin proved faster in all the tests that it was able to use the same timings. For instance, at the 400MHz setting, this 4GB kit proved 4s faster than the 2GB kit at equal speeds. It goes to show how a program like Super Pi loves ram, especially if it can benefit from having wider bandwidth to play with.
Though it’s no longer in development, Sciencemark is a tool I still like to keep in my chest… err thumb drive. It gives results far more in-depth than other benchmarking programs out there, although you’d have to be an engineer to care for -all- the information it delivers. It generates a bandwidth result just like Everest and Sandra does, and is effected by higher CPU clocks.
We barely broke through the 6K mark, but it happened! Let’s compare this 4GB kits performance to other kits we’ve reviewed in the past.
Here, we compare our overclocks of the memory being reviewed alongside other recently evaluated sticks. These graphs include benchmarks with each kit of ram at DDR2-1000 4-4-4-12-13 2.1v along with each kits own top overclock. While the DDR2-1000 results should not vary much, the top end overclocks will, given that each kit will top out differently.
For reference, here are the top overclocks for each kit of ram included:
* CL 5 was used for 500MHz on these kits due to inability for CL 4
One thing I want to point out is that even though I am comparing the 4GB kit to 2GB kits, the performance differences shouldn’t be taken with too much weight. The fact is, 2GB kits will always overclock higher than a 4GB kit, until we reach a day where chips are better refined and able to churn out higher speeds. This might not happen until DDR3, it’s hard to say at this point. The fact that we hit 600MHz on this kit at all is impressive. When in comparison, an overclock like that on a 4GB far surpasses any other overclock in the above list. These comparisons are to show you head on what the differences you can expect will be. These do not take into consideration multi-tasking benefits, which would undoubtedly be far better with additional memory, regardless of whether or not you lose 1,000MB/s on average.
In comparing 4GB to 2GB, the OCZ SLI-Ready PC2-7200 kit had the same top overclocked settings, so you can compare those two to see the immediate differences.
Compared to that OCZ kit, the Mushkins 4GB scored 30MB/s higher in the Read tests, but 3MB/s lower in the Write tests, which are not really memory related as much as they are FSB reliant.
Again, despite the higher bandwidth, the latency when compared to the OCZ kit was lower on the Mushkin kit. I am actually quite impressed with this fact, because I had assumed it would be lower. I believe the reason for this is that since there is always so much bandwidth to deal with, there will be more free bandwidth to access at any given time, hence the improvement.
Where do I begin? This is one of the most impressive kits I’ve laid my hands on in quite a while. With performance as displayed here, moving up to 4GB won’t feel like such a burden. As we saw throughout some of our tests, when compared to a 2GB kit at the same speeds and timings, Mushkin’s kit proved to have better overall scores, in every single test. Whether it be higher bandwidth or better latency, the XP2-8500 performed exceptionally. Because of all my tests, I need to re-think a few things. I felt strongly that a 2x2GB kit at equal speeds to a 2x1GB kit would prove far worse performance, but that couldn’t have been further from the truth.
While we are on the subject of benefits, here are a few more. With 2x2GB, you can have 4GB of ram without sacrificing 1T timings, if you wish to run them at a low frequency. I am not even going to assume that anyone will do such a thing with this particular kit, because it would be a waste of money. However, the potential is there for great latency results. As for stock settings, most DDR2 kits will run 4x1GB at the same speeds, but that means two things. More heat inside your computer and definite need of a fan (if you care that much about temp, like I do).
2x2GB allows better airflow because you don’t have four sticks of ram crammed together. Even still, this kit requires 2.3v like many performance kits on the market, so I highly recommend getting a fan even at stock speeds to keep things running well. I personally recommend Corsairs Dominator Airflow fan. It’s not as effective as ghetto-modding a 120mm fan on top of the modules, but it’s the next best thing if you want to keep your case looking good 24/7.
So, are these modules for you? That’s for you to decide, and you only. The biggest hurt is the $700 price, but from any standpoint, it’s somewhat understandable. Fact is, this is the -only- 4GB kit on the market that will handle DDR2-1066+ speeds reliably, although their PC2-6400 4-3-3 4GB kit has been known to overclock to meet the stock speeds of these sticks. If you do not care that much about added bandwidth or latency, I’d quicker recommend those since they retail for $200 less. Even if you overclock those sticks to DDR2-1000, it will be a sweet deal.
For the hardcore enthusiast, if you happen to manage the same overclock of DDR2-1200 that I did, you will be one of the very few (I still haven’t seen anyone else do this) to run 4GB of ram at those speeds stable. Running 4x1GB at the same frequencies is incredibly difficult and as far as I know, no one has ever managed it stable. The fact that is was stable at 2.5v is another impressive feat. I still don’t recommend that at a 24/7 setting, but if you have good airflow you might be alright. If you insist on those speeds, I’d touch the modules after stressing them to see how hot to the touch they have gotten. If you are using a fan and they are still too hot to keep your finger on for a few seconds, I’d clock down. It’s just not worth bricking a killer set of ram.
I am awarding Mushkin’s 4GB XP2-8500 an 8 out of 10 along with our Editors Choice award. How does a $700 kit deserve an Editors Choice award? Bear in mind that these are for hardcore gamers and overclockers only, not for your mom or dad. They are expensive, but again, they are the only kit on the market to handle these speeds, and have been since their launch. Given that, there is absolutely no competition. Even still, I recommend their still very overclockable XP2-6400 4GB 4-3-3 kit for those who don’t need the most extreme overclocks out there.
To save even more cash though, I’d recommend two kits of 2GB PC2-8500 sticks, to run 4x1GB PC2-8500. If you have a definite need for performance and 4GB though, these modules might just be for you. At the time of writing, the cheapest place to purchase a pair has been Mushkins own site, where they currently sell for $661.
If you have a comment you wish to make on this review, feel free to head on into our forums! There is no need to register in order to reply to such threads.
Copyright © 2005-2020 Techgage Networks Inc. - All Rights Reserved.