by Kirill , Oleg Golubovich
06/15/2006 | 08:45 PM
System memory… A lot of system memory… A lot of memory?!
The severe reality of the computer world more and more frequently spits out programs and applications which system requirements are shockingly high even for sophisticated computer users. And you have to start saving again to get out there and buy a new piece of hardware, which is usually a CPU, graphics card or a couple of memory sticks depending on the applications.
Speaking of the memory, then on the one hand it is a pretty easy task to buy a pair of extra memory modules and install them into the available free DIMM slots on the mainboard. Although the availability of these free slots may sometimes be an issue (especially if you used to have a pair of 256MB modules), because some contemporary mainboards may have only two DIMM slots instead of four. In this case the easiest way to save the situation is to buy two new DIMMs with higher memory capacity (1GB for instance). Ok, you bought those… But if you are into overclocking, you may be quite frustrated to find out that 1GB memory modules overclock less efficiently than the 256MB or 512MB ones.
But, don’t give way to despair, things are not so bad at all. Today we will try to find out the overclocking potential of two 2GB kits out there from G.SKILL Company that belong to the Extreme and value product series.
It is not the first time solutions from G.SKILL Company appear on our web-site, and later on you are going to see much more of them tested. For more information you are welcome to check out the following previous reviews called DDR650? - No Problem! Two Memory Kits from G.SKILL in Our Lab and Roundup of 2GB DDR SDRAM Memory Kits for Overclockers.
So, please meet our today’s testing participants: G.SKILL F1-4000USU2-2GBHZ (Extreme) and F1-3200PHU2-2GBZX (Value) solutions.
The modules were packed in G.SKILL’s traditional plastic kit with a colorful paper background insertion containing some information about the product:
Although you may consider it not really necessary, as the memory modules already have a sticker containing all the essential technical specs. These details on the sticker will give you full understanding of the memory module model, its nominal timing settings recorded in the SPD, its voltage, manufacturers name and serial number.
As for the accessories, there are actually none inside the package. So, all you will find there are the modules themselves and a paper leaflet. It is actually not a surprise for us, because I believe only GeIL provides a small tube of thermal paste with every memory kit they ship out.
Let me provide the brief list of official memory specifications for each of the kits participating in our today’s review:
We will start with G.SKILL -4000USU2-2GBHZ kit. This is PC4000 (DDR500) memory with the nominal timing settings of 3-4-4-8 (for dual-channel mode).
The manufacturer recommends using these memory modules with Socket 939 mainboards based on nForce chipset family designed by such companies as DFI, MSI, EPoX.
The first impression you get when you take these memory modules is that it is a very decent high-quality solution. A quick first look at the DIMMs proves that these modules are built on the best overclocker PCBs – Brainpower B6U808 (note that practically all G.SKILL memory modules are built on this PCB).
The solution is hidden between two black anodized heat-spreader plates pressed together with two steel clips. The company logo is put in the center of these heat-spreaders. Each memory modules has a sticker that carries almost all the necessary technical details about it. Why almost? Well, no one can say right away what memory chips are used for this solution. The manufacturer’s website claims that these are Samsung chips, although no further details are available. I decided to double-check whether this was true and removed the heat-spreaders. At first it seemed like an easy task: just remove the steel clips. However, I had to really apply some effort to safely remove one heat-spreader plate from the memory chips.
In case you decide to undertake the same investigation, please, keep in mind that if you damage the memory module sticker (and one of them goes over one of the steel clips), you will automatically lose your warranty. But even if the sticker remains virgin, you will still have visible scratches – the traces of removed clips on the glossy heat-spreader surface.
Therefore, you will have to believe the manufacturer and myself and avoid satisfying your curiosity at this relatively high expense.
Ok, I can say now that there were UCCC memory chipset from Samsung under the heat-spreaders:
The only significant drawback that catches your eye even if you refrain from removing the heat-spreaders is the thermal pad between the memory chips and the heat-spreader plate that is supposed to conduct heat from the chips into the heat-spreader. This pad is very narrow and doesn’t cover even 50% of each memory chip:
In this case it can hardly play the designated role efficiently enough…
This winds up the discussion of the first kit features and design peculiarities.
This memory kit is positioned by the manufacturer as a value (budget) solution. It is designed in a similar way to the Extreme kit we have just discussed. The major external difference is the bright-blue color of the heat-spreader plates.
Unlike the G.SKILL F1-4000USU2-2GBHZ we have just talked about, this memory boasts much more modest working frequencies, but its timing settings at the nominal frequency are truly impressive: 2-3-2-5! Far not every memory module, even the most advanced ones, can handle these aggressive timings. It is really hard to explain why this happened, but the memory maker recommends using these memory modules for Socket 939 AMD platforms (which certainly doesn’t limit the use of these memory modules to this platform only).
Unlike the Extreme series, the sticker of this memory kit doesn’t mention the nominal voltage these modules should work at:
Also there is no mention of the memory chips used for these modules on manufacturer’s site. However, if you are an experienced overclocker you should probably know that only the memory modules built on Infineon or Micro chips can handle aggressive memory timings like that at their nominal working frequency.
Our intuition proved absolutely correct: we discovered Infineon CE-5 chips manufactured with 130nm technological process when we removed the heat-spreaders.
No doubt, they are among the best memory chips in their class.
I was very pleased to discover the thermal pad covering the memory chips almost completely. So, there should be no cooling or heat dissipation issues with these modules:
Just like in the previous case these modules are built on BrainPower B6U808 PCB.
Now that we have discussed the design and features of the two 2GB kits from G.SKILL, let’s check out their performance and reveal their hidden potential.
For our test session we assembled the following testbed:
The memory power circuitry on the mainboard has been modified and we used digital multi-meter to monitor the supplied voltage.
We overclocked the memory synchronously with the CPU clock generator. If we had to get the frequency below 200MHz, the memory divider was set as DDR166.
We tested the system stability in two stages:
We decided that we will perform all the tests on our today’s memory solutions to reveal their potential and hidden features. Therefore we split the entire testing session into a few stages. First of all we took into account the specifics of the tested memory modules. Judging by this parameter we set the following timings in the following sequence: CAS Latency (tCL), RAS to CAS delay (tRCD), Row Precharge, tRAS.
1. For G.SKILL F1-4000USU2-2GBHZ:
2. For G.SKILL F1-3200PHU2-2GBZX:
Moreover, all the above mentioned timing combinations have also been checked out with the Command Rate set to 1T and 2T.
We tested the modules with Vmem at 2.66V and 2.8V, according to the recommended voltage range for both kits. We have also tried to get each kit to work at 3.0V voltage or close to that, however the results at this Vmem haven’t been included into score charts for the reasons explained later in this article.
Before we start discussing the benchmarks results I would like to draw your attention to the voltage on the 1GB memory modules. The thing is that all 1GB memory modules are built using high-density chips and so they do not really tolerate high voltages well by nature. Do not try setting the voltage over 3.0V, because the memory chips may degrade very rapidly. Also, 1GB memory modules overclock pretty well at 2.8V, which is considered to be pretty modest voltage rate from the overclockers’ standpoint. Further voltage increase may negatively affect overclocking results or have no effect at all. Therefore, we didn’t perform all the tests at 3.0V Vmem setting: in this case we experienced 2-10MHz frequency drop and saw no improvement of the overclocking potential whatsoever. That is why we didn’t include the results into any of the score charts below.
Also, our results may not necessarily coincide with your numbers, because there are multiple factors that affect the performance scores: mainboard type, CPU, BIOS settings, “lucky” or “unlucky” memory modules set. But despite this fact, our diagrams will most closely represent the statistics for solutions identical to G.SKILL modules reviewed today.
As usual we will start with the extreme series - F1-4000USU2-2GBHZ. Here is what the SPD reported:
And here are our test results:
The situation is actually quite familiar, looks we have already seen something like that somewhere… Of course, these results have been obtained from the previously tested high-quality memory modules from other manufacturer built on Samsung TCCD chips!
The results were simply impressive: I didn’t expect the reach these frequencies on 1GB memory modules. Way to go, G.SKILL! BY the way, the mainboard we used for our test session works perfectly well with 1T/2T Command Rate setting that is why most likely we couldn’t go beyond 270MHz because of the memory or the CPU memory controller. Or because of them both.
Now let’s analyze the results we have just seen on the diagrams. The picture is the following: until we hit 240MHz, neither voltage nor Command Rate influences the frequency. And the increase is non-linear in this case. The sacred 2-2-2-5 timings can be achieved at shockingly low 165MHz (there are different memory modules designed to work efficiently with these timings, however, they cannot boast 1GB capacity). However once we increase the CAS latency from 2 to 2.5, we immediately gain extra 40MHz putting the memory modules at a very acceptable performance level.
Nevertheless, the results are not that high, which gives us the right to conclude that there are no memory modules of that capacity with universal characteristics. At least, G.SKILL doesn’t have them. The memory modules they offer right now either work at low memory timings (see below for details), or reach high clock speeds. In fact, we see the same situation as 3 years ago when the first DDR500+ memory modules came out.
That is why we are not really interested in the performance of our memory modules at the frequency below 250MHz. and after that the situation gets really interesting. In the hardest conditions (with the nominal memory timings of 3-4-4-8, 1T, 2.66V) the maximum frequency we managed to achieve is 265MHz. It is an excellent result keeping in mind that these are 1GB modules. By setting the Command Rate at 2T we can achieve additional 30MHz gain. And when we change the timings even more until they reach mockingly slow 3-5-5-10 and set the Vmem to 2.8V the outcome is practically null: altogether these changes hardly get us a 10MHz increase.
Now let’s look at G.SKILL F1-3200PHU2-2GBZX. This product is also far from ordinary. Why so? I believe you have already noticed a very specific set of timing combinations in the Testbed and Methods section of this review, haven’t you? The thing is that memory built with Infineon CE-5/6 chips is absolutely indifferent to Row Precharge parameter resembling Winbond chips from this prospective: the standard value of this parameter equals 2, and raising it doesn’t have any positive effect on the overclocking potential.
The same it true for other timing settings, too. They are certainly not the 2-2-2-5, but still lower than what we have just seen by F1-4000USU2-2GBHZ. Increasing these timings (except CAS, of course) doesn’t affect the frequency potential in any way.
We were a little bit upset with the overclocking results. According to statistics that has been accumulated on a few overclocking forums, the maximum frequency achievable on Infineon CE-5/6 chips can reach the impressive 260-300MHz. however with the current combination of frequency, timings, capacity and positioning of this memory kit as a value-line solution the result is still worth our applause.
I would like to stress that F1-3200PHU2-2GBZX is not very sensitive to the Command Rate parameter, as its influence is not high enough to make to let the overclocking results go beyond the measuring error. Therefore we only offer the scores for 1T Command Rate.
These data are also given in the same diagram. We immediately notice a few tendencies.
Firstly, the behavior of F1-3200PHU2-2GBZX differs dramatically from that of the previous kit: the frequency difference between the worst and the best result is half as big. However, the difference between the best and the worst timing settings is half as big, too. Note that this memory kits lacks only 5MHz before it reaches DDR400 mode with 2-2-2-5 timings. However, the peak working frequencies of F1-3200PHU2-2GBZX kit are much higher (10-15MHz) than those of the elder model. At least until 255MHz.
Secondly, the memory modules managed to get just a little bit beyond DDR500 mode, no matter what testing conditions applied. This is exactly what I meant when I mentioned the non-universal character of these 1GB memory DIMMs. The user will have to choose between frequency and timings when purchasing the modules, because the simple settings adjustment in the BIOS, the way they do it with 512MB modules (in particular the ones built with Samsung TCCD chips), will not solve the problem.
Thirdly, the only principally important parameter that can determine the frequency potential is CAS Latency. When we change the timings to 2-2.5-3, the frequency increases linearly.
And fourthly, as we can see from the diagrams, the voltage set 2.66V doesn’t really affect the modules’ potential (no more than 5MHz increase). This is very bad news for extreme overclockers (volmodding won’t help), but very good news for those who need reliable workhorse for everyday routine. In fact, no common benchmark will reveal any significant difference in results when you compare the performance of two 2GB modules against two 512MB modules. So, if you are one of those 3DMark enthusiasts, no need to worry.
During our test session we didn’t detect any significant overheating of the memory modules, so we can conclude that our today’s heroes from G.SKILL do perfectly well with their nominal heat-spreaders (although you can also do without them).
Once again G.SKILL memory modules proved the great reputation of this system memory maker, having demonstrated that even large memory capacities can overclock pretty well.
This is is an ideal solution for those who do not suffer from low timing settings. They can clock to the level of DDR600 and perform close to the memory modules of smaller capacity. This memory kit works best in DDR500-533 mode with nominal timings and Vmem or a little bit slower than DDR600 – with Command Rate set to 2T. Both these workmodes are an excellent result for 1GB memory modules.
This is pretty specific memory that can work at maximally low timing settings supported by the modules with 1GB capacity. In the best case they can overclock as efficiently as the memory modules built with Samsung UCCC chips. They are surprisingly easy and predictable in operation: install the memory DIMMs, set the CAS Latency according to your frequency preferences and that’s it. Mainstream users will be very happy about the simplicity of setting the corresponding frequencies, timings, Command rate and voltage: there is one single CAS responsible for all of them at once. As a result, G.SKILL will be a great choice for systems designed for minimal possible latencies in the DDR400 to DDR500 frequency range.
Anyway, if you need A LOT OF system memory and there is a chance to get G.SKILL kits, then excellent quality of these products as well as carefully selected memory chips will always be an argument in their favor.