by Ilya Gavrichenkov
01/16/2009 | 06:37 PM
Rapid drop of memory prices that coincided with the growing popularity of Windows Vista OS and release of multiple resource-hungry applications and computer games made 4GB of system memory an absolutely justified choice even for mainstream platforms. Most users currently equip their systems with this amount of memory. However, you won’t see the same unanimity in other memory subsystem configuration matters. For example, what is preferable: two memory modules 2GB each, or four memory modules 1GB each? Or which memory type, DDR2 or DDR3 would be a better choice today?
In fact, the answer to the first question is quite evident: the advantage of 2GB modules pairs is that it leaves two DIMM slots on the board free. First, it allows to increase the amount of system memory in the platform without replacing the already installed memory modules, and second, it lowers the electrical load on the memory controller thus improving the general system reliability. Of course, four memory module configurations have every right to exist and work problem-free in most cases, but we would still recommend using 2GB modules, especially if you intend to overclock your system.
The second question is much more interesting, despite its seeming simplicity. Of course, the owners of Socket AM2+ and LGA1366 platforms cannot choose the type of memory to be used on their platforms, however, what about those users who focus on extremely popular LGA775 platforms boasting numerous advantages? Intel as well as mainboard makers do not pose any restrictions to the memory types to be used on their platforms: there are both types of mainboards in the market today – for DDR2 as well as DDR3 SDRAM. Nevertheless, buying DDR3 SDRAM was considered an unjustified waste of money a few months ago, because DDR3 SDRAM modules cost several times more than DDR2 ones of the same capacity. The frequency advantages of the new memory type, however, didn’t provide an adequate performance improvement, as we have proven by our tests of previous-generation LGA775 platforms based on “third-series” Intel chipsets.
At first glance, things today are very similar to what they used to be. 4GB DDR3 SDRAM kits are still 2-3 times more expensive as DDR2 SDRAM kits. However, if we regard this difference from absolute rather than relative positions, we will see that the recent price drop made the price difference between 4GB DDR3-1600 SDRAM and high-quality DDR2-800 and DDR2-1067 kits less than $100. And this is a pretty insignificant sum, considering that the price difference between the two top LGA775 processors is often much greater. Therefore, the choice of DDR3 SDRAM is justified not only from the emotional prospective, but also from the economical one, if this memory provides any performance improvement in contemporary systems.
Actually, our today’s article will be devoted to checking out these particular assumptions. We are going to investigate the performance of systems equipped with 4GB DDR3-1600 SDRAM and based on the latest LGA775 chipset – Intel P45.
To test our DDR2 SDRAM and DDR3 SDRAM we put together two identical platforms built around similar mainboards from Asus on Intel P45 chipset. They were Asus P5Q3 and P5Q Pro and differed only by the type of supported memory. As for the CPU, we used the top dual-core processor – Core 2 Duo E8600. This processor can be regarded as a mainstream solution according to today’s standards.
The complete system configuration looked as follows:
Note that we ran the tests in Windows Vista Ultimate SP1 64-bit, because only a 64-bit system can use all 4GB of memory available to it and applications to the full extent.
First of all I would like to clarify why decided to focus on DDR3-1600 SDRAM, while more mainstream DDR3-1333 memory modules are currently much more affordable and wider spread in the market. It really doesn’t make much sense to prefer DDR3-1333 over f faster DDR3 SDRAM, because its high latencies put it behind DDR2-1066 and even DDR2-800. However, the results of our practical tests will be much better evidence to that. We compared the performance of a system equipped with 4GB DDR2 SDRAM against that of a similar system with DDR3-1333 memory. For more illustrative results we tested DDR3-1333 with two sets of timings: 9-9-9-27 and 7-7-7-20. The first set of timings is typical of most mainstream memory kits, while the second set is only available in products from well-known memory makers positioned as overclocker solutions.
The tests were performed in a system with a CPU working at its nominal speed, i.e. at 1333MHz processor bus frequency.
First of all we decided to check out the results of synthetic Everest Ultimate 4.60 test that measures the practical bandwidth and latency of the bus between the processor and the memory during data reading, writing and copying.
Read speed from the memory and memory latency prove to be the most dependent on the memory subsystem parameters. They also affect the system performance in different applications most of all that is why you should pay special attention to them. In fact, the corresponding diagrams show very clearly that DDR3 memory cannot demonstrate any impressive advantage, as we have just said above. In terms of real bandwidth, regular DDR3-1333 with 9-9-9-27 timings turns out equivalent to DDR2-1067, and in terms of latency it can only compete against DDR2-800. DDR3-1333 for computer enthusiasts that can work with CAS Latency 7 performs a little better. It reads the data slightly faster than DDR2, but features latency comparable to that of DDR2-1067.
Judging by these numbers we can expect slower DDR3-1333 to be no rival to the good old DDR2 memory. As for fast DDR3-1333, it may only be able to reach the performance level of DDR2-1067. Further tests will show if these assumptions are correct.
To estimate the complex performance rate of our test systems we used two testing suites: graphics Futuremark 3DMark06 and “general performance” CustomPC Benchmark. For our 3D test we chose not the latest version of the popular Futuremark application for a reason. Its results depend more on the processor and memory subsystem performance than the results of 3DMark Vantage. We chose CustomPC Benchmark to replace the traditionally used PCMark Vantage and SYSmark 2007, because the performance index readings it generates do not fluctuate too much. As a result, we can achieve higher measurement precision with fewer test runs when testing similar platforms differing only in memory subsystem.
It would be silly to expect 3DMark06 to demonstrate any significant performance difference between the systems equipped with different memory types. This test serves to measure the performance of the graphics subsystem, which works almost identically with any type of system memory employed, as we have already seen.
However, general system performance measured during work in resource-hungry applications does depend on the memory, and this dependence is often quite significant. The testing suite we used checks the system performance under several types of workload, such as: digital photo editing in free GIMP graphics application, encoding MPEG-2 video into H.264 format using Handbrake utility, data archiving and encoding with 7-zip, and high-definition video playback.
Overall, the performance difference between the tested systems equipped with the fastest (in our today’s test session) and the slowest memory makes almost 5%, which makes us choose the system memory seriously. The memory parameters matter most for its performance under multi-threaded load, when fast DDR3-1333 ensures a 14% advantage over the platform equipped with slow DDR2-800 SDRAM. At the same time systems equipped with DDR3-1333 memory cannot outperform identically configured systems equipped with the best DDR2 SDRAM. In fact, the tests in real applications repeat the results obtained in synthetic benchmarks. DDR3-1333 with 9-9-9-27 timings can only be regarded as a performance alternative to DDR2-800 with 4-4-4-12 timings, and DDR3-1333 with more aggressive 7-7-7-20 timings works as fast as DDR2-1067.
We should pay special attention to the systems performance in games. As you know, games are extremely sensitive to memory subsystem characteristics. To ensure an even more evident effect, we ran all tests in low resolutions, where the graphics subsystem has minimal influence on the frame-per-second rate.
True, games turn out to be these particular tasks where high bandwidth of DDR3 SDRAM affects the performance. Honestly, the performance advantage demonstrated by DDR3-1333 is not too impressive, however, you can still notice it with a naked eye. Nevertheless, DDR3-1333 with CAS Latency 9 can’t outperform DDR2-1067, and DDR3-1333 with 7-7-7-20 timings is barely 0.5% faster than DDR2-1067.
In addition to the results discussed above we also measured the testbeds performance in a number of other applications creating significant system load:
In fact, there is nothing surprising here. All the conclusions made before prove absolutely correct during archiving, video encoding, rendering and chess gameplay. The archiving utility turned out the most dependent on the memory subsystem performance; however, even in this case DDR3-1333 SDRAM was no more than 3% faster than the platform with fast DDR2-1067.
The conclusion is evident: using DDR3-1333 SDRAM in contemporary LGA775 platforms doesn’t make any sense considering how expensive it is. Therefore, today we are not going to talk about 4GB DDR3-1333 kits. Instead we are going to discuss DDR3-1600 SDRAM.
It turns out DDR3-1600 SDRAM faces a pretty serious problem. The today’s LGA775 systems, even those based on the latest Intel P45 chipset, officially do not support DDR3-1600 SDRAM. And it is not just a missing formal mention ion the specifications of the chipset and mainboards based on it. Things are in fact much worse: the memory controller in the chipsets for LGa775 processors supporting DDR3 memory cannot clock the memory at frequencies more than 4x higher than the FSB speed. And its means that the top memory type on the list of supported ones will be DDR3-1333 SDRAM even for CPUs with 1333MHz Quad Pumped Bus, and as we have just seen, this memory doesn’t improve the performance in any significant way.
The way-out of this situation is actually pretty evident: you can clock the memory at 1600MHz by setting the FSB at 400MHz. Unfortunately, there is only one CPU in the market today that supports this FSB frequency. It is Core 2 Extreme Q9770, which price exceeds all reasonable boundaries, that is why it is of no immediate interest to most users out there. In this case, there is practically only one solution remaining for those willing to use DDR3-1600 SDRAM: overclocking.
Of course, the need to increase the FSB speed past the nominal values does set certain limitations for the use of DDR3-1600 SDRAM. However, these limitations will definitely not discourage us from checking this memory inside out. Especially since contemporary system allow increasing the FSB to 400MHz without any difficulties. Since this frequency is in fact nominal for certain existing LGA775 processors, chipsets and mainboards should work just fine in these conditions. As for the CPUs, processors with 1333MHz Quad Pumped Bus can easily cope with 33% FSB frequency increase from 333MHz to 400MHz. In most cases, contemporary 45nm processors overclocked only that far will work stably without any additional voltages adjustment or special cooling solutions.
To illustrate this statement I would like to add that Core 2 Duo E8600 processor we used in our today’s test worked perfectly fine at 4.0GHz when the FSB frequency was raised to 400MHz. Experiments showed that this overclocking didn’t require any increase of the processor core voltage.
However, if you are not one of the overclocking fans, you can always lower the processor multiplier when increasing the FSB frequency. In this case the resulting CPU clock speed will not exceed its nominal value.
Note that at 400MHz FSB the processor bus bandwidth increases to 12.8 GB/s. The memory subsystem built on DDR2-667 or slower memory modules cannot provide the same bandwidth, and DDR2-800 can only guarantee that the processor bus bandwidth and memory bus bandwidth will match. Therefore, LGA775 chipsets do not support slower memory than DDR2-800 SDRAM at 400MHz FSB. And it means that you will not be able to increase the FSB frequency any further with this type of memory in your system. This is actually the major drawback of DDR2-800 SDRAM: it may not just limit the system performance at 400MHz FSB, but can also turn into a serious problem during further overclocking attempts.
Moreover, slow memory in systems with 400MHz FSB (and higher) may cause another problem. Low frequency of the memory subsystem requires higher setting of the Performance Level parameter – the chipset uses this latency to synchronize the processor and memory busses. This factor also affects the performance. Fast DDR3-1600 SDRAM supports Performance Level 6 or 7, while DDR2-800 SDRAM requires this parameter to be set at 8 or 9. Faster DDR2-1067 SDRAM is a little better in this respect: it allows lowering Performance Level to 7. The BIOS’s of contemporary mainboards are very well familiar with this chipset peculiarity that is why this factor is yet another reason for performance differences between systems using different memory types.
So, it looks like the use of faster DDR3 memory may really pay back in systems with 400MHz FSB. Let’s take a look at the results of our experimental test session performed in a system with Intel Core 2 Duo E8600 processor overclocked to 4.0GHz by raising the FSB frequency to 400MHz. Dual-channel 4GB memory kits we are going to discuss here worked in the following modes:
The above listed DDR3-1600 timings configurations were selected basing on the specifications of solutions available in the market these days. It is also important to keep in mind that DDR3-1600 memory with CAS Latency 7 is the most widely spread type of memory. Therefore, we will focus on this type of SDRAM when analyzing the advantages of fast memory subsystem.
The latency of the memory subsystem based on DDR3-1600 is close to that of DDR2-1067. However, the bandwidth tests show that DDR3 boasts a serious advantage. This is a typical situation that is hardly surprising. What you should in fact pay special attention to, is the fact that transition from DDR2-800 to DDR2-1067 provides a much greater parameters improvement than transition from DDR2-1067 to DDR3-1600. You may consider this fact a perfect illustration of how insufficient the bandwidth of DDR2-800 is in a system with 1600MHz Quad Pumped Bus.
However, the results of synthetic benchmarks should not be basis for conclusions; let’s check out the results in real applications before jumping to them:
Unfortunately, just like in DDR3-1333 SDRAM tests, we have to state that fast memory is often not enough to speed up the entire system. Nevertheless, there also exist opposite examples, for example, multi-threaded load created by CustomPC Benchmark. The performance difference here between slow DDR2-800 and fast DDR3-1600 reaches the impressive 22%. However, DDR2-1067 SDRAM falls just a little behind DDR3-1600: the fastest DDR3 memory can get 5% faster at best.
We observe a similar picture in games. DDR3-1600 results look impressive only against the background of a system with DDR2-800 SDRAM. In this case the performance advantage lies between 5-10%. If we look at the performance difference between systems with DDR2-1067 and DDR3-1600, the advantage of faster memory will no longer be that evident.
We can draw the same conclusions basing on the results obtained in application tests. It is evident that DDR2-800 SDRAM is not the best choice for an LGA775 system working at 400MHz FSB. Using DDR2-1067 memory is quite justified in this case, and the advantage of DDR3-1600 will only be a slight performance improvement.
So, it turns out that DDR3-1600 is not a highly justified choice even for systems running at 400MHz FSB. The performance improvement it provides compared to DDR2-1067 SDRAM is insignificant, while the price difference between them is quite serious. Nevertheless, we don’t give up hope to find suitable application for the DDR3-1600 SDRAM, so let’s see what happens during more serious system overclocking.
The next series of tests will reveal how well the system equipped with DDR3-1600 SDRAM will perform with the FSB frequency increased to 500MHz. Overclocker platforms often work at this or even higher bus frequency, so memory with high bandwidth may turn out exactly what one needs in this case. This overclocking increases the peak processor bus bandwidth to 16 GB/s, which makes the efficient use of dual-channel DDR3-1600 SDRAM potential much more likely.
For our tests we used the same Core 2 Duo E8600 processor overclocked to 4.0GHz. However, we lowered its multiplier and obtained the resulting frequency as 8 x 500MHz.
Note that this FSB frequency increase should be taken more seriously than the previous one to only 400MHz. In this case, the system may only be stable if running on a high-quality mainboard equipped with an advanced chipset North Bridge cooling system. Moreover, you will most likely have to increase the chipset NB voltage and CPU VTT. Nevertheless, many computer enthusiasts overclock their systems with FSB set at 500MHz or even higher. Many mainstream LGA775 processors support relatively low frequency multipliers that do not allow reaching high frequencies with any other overclocking methods. So, 500MHz FSB is not extreme overclocking exotics, but more of a widely spread occurrence that deserves special attention on our part.
The memory tested within this series of experiments was configured as follows:
Here I have to point out two things. First, we had to seriously ease the Performance Level North Bridge timing during overclocking with such aggressive FSB frequency increase. To ensure system stability we had to increase it to 9-10. Second, we had to use the test DDR2-1067 SDRAM in a slightly slower DDR2-1000 mode. The reason for that is the fact that existing Intel chipsets do not support any dividers that could help us set the memory frequency at anything closer to 1067MHz. By the way, for same exact reason when we overclock to 450MHz FSB or higher we have to use “synchronous” divider for the memory frequency (when DDR2 frequency equals twice the FSB frequency) for most DDR2 SDRAM modules.
Nevertheless, despite the above mentioned “allowances”, we hope that additional FSB overclocking will have a positive effect on performance.
The measurements taken with the synthetic Everest Ultimate 4.60 benchmark reveal pretty curious occurrence. It is for the first time that any DDR3-1600 SDRAM turns out faster than DDR2 in bandwidth as well as latency. And it means that our earlier suppositions are correct, namely when the processor bus frequency increases significantly over its nominal value, memory speed is of greater importance. This way, there is no doubt that DDR3-1600 is extremely fit for overclocker systems. Now we just have to estimate how significant the gain from this expensive memory will be.
Let’s resort to benchmarks estimating the “real” performance.
3DMark results are not impressive; as we have already seen, this test barely reacts to memory subsystem parameters. However, CustomPC Benchmark shows that in real, mostly multi-task scenarios, fast DDR3-1600 with 7-7-7-20 timings may provide a pretty significant performance improvement. The performance may increase by up to 10%, while the total score reported by the benchmark shows almost 3% advantages of the system with DDR3-1600 SDRAM.
High-speed memory is also great for contemporary games. Even DDR3-1600 with “weak” timings of 9-9-9-27 works a little faster than DDR2 SDRAM. More widely spread faster DDR3-1600 with 7-7-7-20 timings provides a tangible fps advantage. Note that with higher processor bus frequency the value of fast memory also increases. And although at 400MHz FSB the performance gain was more of a symbolic nature, then at 500MHz FSB it reached 5%.
We could say almost the same about the results obtained in applications. DDR3 memory improves the performance differently depending on the type of load; however, it is indisputably better than the old memory standard in all tests.
So, those overclockers who have their systems running at comparatively high bus speeds will be the first to feel the actual effect from DDR3-1600. According to our tests, high-speed memory starts really paying back when the FSB frequency gets far beyond the nominal levels. Moreover, DDR3 SDRAM offers a wide range of dividers setting the FSB-to-memory frequency ratio. It not only gives you more freedom during overclocking, but may become the ultimate success factor for setting new overclocking records.
The obtained test results suggest one more interesting thing: how efficient high bus frequency could be in LGA775 platforms. Just compare the benchmark results obtained at 4.0GHz CPU speed set as 10 x 400MHz and 8 x 500MHz. If you use DDR2-1067 memory with 5-5-5-15 timings, the results turn out pretty unpredictable:
Although it may seem illogical at first glance, the 500MHz FSB doesn’t provide any advantage to the system equipped with DDR2 SDRAM. Moreover, faster bus will, on the contrary, lower the system performance in most cases. In fact, there is a very simple explanation to that. First, by raising the processor bus frequency you increase the internal latency of the chipset North Bridge (Performance Level), and higher performance of the memory subsystem cannot make up for it. Second, when we tested our system at 500MHz FSB, we had to slightly lower the memory frequency, so it actually worked as DDR2-1000. It is important to remember that we didn’t do it in order to ruin the race for DDR2. It is the only possible mode in this case, because Intel’s LGA775 chipsets support very limited range of memory frequency dividers. In case of dramatic FSB frequency increase there is only one divider we could use for DDR2 memory: the “synchronous” one.
The picture turns completely different if we compare the results obtained for DDR3-1600:
All test results indicate a performance increase resulting from the higher bandwidth of the processor bus. However, it is a pretty relative win we are talking about here. It doesn’t exceed 1% on average. Inevitable Performance Level increase during FSB overclocking has its negative effect also with DDR3-1600 SDRAM.
So, the obtained results indicate clearly that FSB overclocking doesn’t have any individual practical value. It only makes sense to set high front side bus frequencies in order to reach high processor clock speeds or to overclock the memory. If you can get the same processor and memory frequency from higher FSB speed as well as from higher multiplier, you shouldn’t focus on the bus frequency increase. However, if your overclocking experience requires setting the FSB at 400-500MHz or even higher, it is important to understand that DDR2 memory will “slow down” the system and you will be able to fully reveal the system potential only with high-speed DDR3.
The results of this investigation also allow making one important conclusion: DDR3-1600 may be demanded. However, you can really benefit from its high speed only if you use high FSB frequencies or 500MHz+. In these operational conditions it does provide better performance than previous generation memory. Therefore, this memory may be of interest only to overclockers, who also value DDR3 for a richer selection of memory frequency dividers and wider opportunities for system performance optimization.
If you have read that far already and are still interested in DDR3-1600 SDRAM, we would like to introduce to you a number of 4GB memory kits for computer enthusiasts available in the today’s market from the leading memory makers.
The memory modules sold under Aeneon brand name are manufactured by Qimonda. It is an excellent recommendation already for those who “know their way” around memory products, because Qimonda Company is one of the leading DRAM chip makers. So, they are selecting the best chips for their brand name enthusiast products, which determines the exclusivity of Aeneon XTUNE solutions.
In our today’s article we decided to line-up all products participating in the test session in alphabetical order, therefore, we had to put the Aeneon XTUNE AXH860UD20-16H-K-4G description in the very beginning. And unfortunately, it may not give you the best opportunity to notice the uniqueness of this solution, best seen against the competitors’ background. These are the only memory modules designed to work at 1.5V voltage – the nominal voltage setting for DDR3. In other words, while most overclocker memory makers increase the voltage in order to improve the chips overclocking potential, Qimonda uses an absolutely “honest” approach: Aeneon XTUNE memory works at 1600MHz at the voltage level defined by JEDEC specification.
However, you shouldn’t be misled by the fact that Aeneon XTUNE memory modules work at high frequencies and low voltage. The peculiar thing about the memory chips used for them is that increasing the voltage doesn’t improve their overclocking potential. Therefore, do not expect this memory to conquer unattainable frequencies once you push its voltage setting up: the maximum frequency we managed to achieve during our overclocking experiments with different voltage and timings settings was even lower than what we could squeeze out of DDR3 memory with 1.8-1.9V nominal voltage.
Aeneon XTUNE AXH860UD20-16H-K-4G kit ships in standard clear plastic packaging. The memory modules look quite common. Their only peculiarity is the blue PCB. The heat-spreaders used to dissipate heat from the chips are common stamped aluminum plates of black color with Aeneon and XTUNE logos on them.
The manufacturer mentions the following specifications for their Aeneon XTUNE AXH860UD20-16H-K-4G kit:
Note that the manufacturer puts only the frequency and the key CL value on the modules themselves.
I also have to say that compared with what other memory makers have to offer, Aeneon XTUNE modules also feature comparatively high latencies, which must be determined by the low voltage. Therefore, Aeneon XTUNE AXH860UD20-16H-K-4G should be considered an exotic product that may be of interest only in some particular cases. For example, if a mainboard doesn’t allow increasing the DIMM voltage, or if low power consumption and no heat dissipation matter the most. Luckily, Qimonda has no intention to make a profit out of exotics lovers and offers their memory kits at even lower price than any other “mainstream” DDR3-1600 SDRAM.
Memory modules SPD have been pretty sloppily programmed for DDR3-1333, DDR3-1500 and DDR3-1667 frequencies. The nominal mode is only available to you in XMP profile, however the voltage setting there is increased.
So, those users who decided on Aeneon XTUNE AXH860UD20-16H-K-4G memory kit will have to correct the memory subsystem parameters manually in the BIOS Setup.
Corsair Company doesn’t need a special introduction. This is undoubtedly one of the most eminent overclocker memory makers. The memory from this manufacturer usually stands out due to remarkable specifications and outstanding quality. However, for the current project we received Corsair TW3X4G1600C9DHXNV memory kit that yields to many competitor solutions out there in characteristics. We could say in Corsair’s defense that their TW3X4G1600C9DHXNV memory is targeted for Nvidia nForce 790i Ultra SLI based systems, however, in reality it is absolutely identical to TW3X4G1600C9DHX modules that have no special optimizations of any kind.
The main peculiarity of Corsair memory is obviously their brand name DHX heat-spreaders that have two advantages over traditional memory cooling systems. They have larger heat dissipating surface area due to their comb shape and hence remove heat from the chips more effectively. Besides, they also feature special additional plates cooling the PCB and not the memory chips. However, these unique heat-spreaders also have a drawback. They are taller than standard cooling systems, so Corsair memory modules may not fit into systems equipped with certain large CPU coolers.
Unfortunately, this sophisticated cooling system did affect the price, but not the features. Corsair TW3X4G1600C9DHXNV costs about $40-$50 more than competitors, while the specifications are overall quite modest:
As we see, at 1.8V voltage Corsair could only reach the same specifications as Aeneon XTUNE modules working at much lower voltage. In Corsair’s defense we could say that the modules did perform pretty well in our tests: at 1600MHz frequency we could get them to work stably at CAS Latency 7.
Corsair engineers are always very thorough when it comes to SPD programming, which makes their memory compatible with a wide range of different mainboards. This time is also no exception. The modules’ SPD contained a broad variety of operational modes. However, taking into account that Corsair TW3X4G1600C9DHXNV is targeted for use in Nvidia based systems in the first place, there was no XMP support.
Instead the modules support EPP 2.0 profiles that are quite useless for the majority of users. So, if you want to properly configure your Corsair TW3X4G1600C9DHXNV, you will have to do it manually through BIOS Setup.
Mushkin is one of the oldest manufacturers of high-quality memory modules; they have been in the market since 1994. Since then they have earned very good reputation, which is totally confirmed by the results of our tests. This time we had a chance to check out a kit of two 2GB modules with 996601 part number that belongs to the fastest DDR3-1600 modifications in the market.
Nevertheless, Mushkin 996601 modules cannot boast any striking appearance. They are shipped in pretty common plastic casing, and the modules are covered with the simplest stamped aluminum heat-spreaders colored black. It is only the manufacturer logo and slightly unusual shape that may catch your eye.
The specifications, however, are worth a much closer look. Mushkin 996601 4GB XP3-12800 kit boasts very aggressive timings, even for DDR3-1600:
As you can notice, these modules require higher voltage than the common 1.8V. Nevertheless, Mushkin 996601 4GB XP3-12800 modules worked with their default settings at 1.8V voltage just fine, which may indicate that the manufacturer simply wanted to play it safe by setting higher voltage requirement.
It is very nice that great Mushkin 996601 specifications didn’t affect its price. It costs considerably less than the solutions from OCZ or Corsair, however, it is not any slower in real tests.
As for the SPD contents, Mushkin memory is just as good as the best competitors’ solutions in this respect.
It has everything necessary for broad compatibility, and the XMP profile contains the settings for DDR3-1600 mode.
OCZ solutions are extremely popular among overclockers. And there are very good reasons for that: the memory modules from this maker often boast better specifications that the other manufacturers’ solutions.
The 4GB DDR3-1600 SDRAM kit participating in our today’s test session is an indisputable bearer of this good tradition. It belongs to the Enhanced Bandwidth Edition series, which indicates that it supports exceptional timings. OCZ memory modules offered even more aggressive parameters than Mushkin kit: their default timings are 7-6-6-18. However, the remarkable parameters of OCZ DDR3 PC3-12800 Platinum EB are the reason why this kit costs so much: it is more expensive than the alternatives from other makers. Nevertheless, this memory is not the most pricy solution: Corsair kits hold the lead here.
OCZ Company used meshed copper heat-spreaders for their modules. According to the company engineers, this shape provides better air convection and hence improves cooling. It is pretty hard to check how true this actually is, but one thing is undeniable: the modules have very attractive flashy exterior, as their heat-spreaders are polished to mirror-shine.
The nominal specifications of OCZ DDR3 PC3-12800 Platinum EB modules look just as good as the modules themselves:
Just like Mushkin, OCZ had to increase the memory voltage to 1.9V in order to ensure more aggressive timings support. But it will hardly upset real computer enthusiasts. OCZ specifically stressed that their OCZ DDR3 PC3-12800 Platinum EB modules could be used freely at up to 1.95V voltage setting.
Unfortunately, these memory modules are shipped with unfinished SPD that contains no XMP profiles, which have become an unannounced standard for overclocker DDR3 SDRAM. However, there are quite a few configurations available ensuring maximum compatibility.
In order to get these modules working at 1600MHz frequency in the nominal mode, the user will still have to go through manual configuring in the BIOS Setup.
In conclusion to our introduction of OCZ DDR3 PC3-12800 Platinum EB I would only like to add that the best official specifications do not always mean the best practical performance. These memory modules yielded to their competitors in a few benchmarks. However, it would be unfair to overlook the fact that OCZ memory kit demonstrated the absolute best results during frequency overclocking among all other today’s testing participants.
PDP Systems Company that owns the Patriot trade mark usually tries to make its products popular among computer enthusiasts by making them a perfect combination of price and features. The Patriot PVS34G1600LLK kit we got for our tests is an excellent example of how this strategy works.
However, the first thing that will win your heart when you see Patriot PVS34G1600LLK will be their unusual exterior. Unlike other manufacturers’ solutions, these modules ship in a bright cardboard box with the modules sitting in clear blister packs inside.
The modules are also not that common: they look pretty similar to Corsair solutions due to their comb-shaped heat-spreaders. However, you shouldn’t accuse Patriot of copycatting: their heat-spreaders are of completely different design. Although they feature similar comb-like shape, Patriot modules have two plates instead of four. Besides, patriot heat-spreaders are not solid aluminum. The surface contacting the chips has a thin layer of copper that improves heat distribution over the entire heat-spreader surface. Besides, assembled Patriot DIMMs are shorter than Corsair ones, so they will most likely fit into any system configuration.
Patriot PVS34G1600LLK modules cannot boast remarkable characteristics, they are more on the mainstream side:
By raising the voltage setting to 1.9V the developers guarantee that their memory will work with 7-7-7-18 timings, but nothing more than that. Setting RAS# to CAS# and RAS# Precharge below 7 cycles is not mentioned anywhere: this is why Patriot memory modules are being offered at such attractive price point.
And although we managed to get Patriot PVS34G1600LLK kit to work stably at 1600MHz frequency with 7-6-6-18 timings, it couldn’t compete against more expensive DDR3-1600 SDRAM modules from OCZ or Mushkin in maximum frequencies.
The parameters Patriot engineers wrote into the modules SPD, are a perfect example of how it should be done:
There are two standard parameters sets the ensure Patriot PVS34G1600LLK compatibility with a wide range of mainboards. There is also an XMP profile with timings for DDR3-1600 mode.
Wintec Industries solutions are not so well-known among computer users. However, this manufacturer is not a new name in the memory world. Wintec used to focus mostly on the OEM market before. However, they decided to pay more attention to the retail channel lately that is why the AMPX product series emerged. This series includes our today’s hero - Wintec 3AXH1600C9-4096K modules. So, you shouldn’t be afraid of Wintec branded products – this memory maker has stood the test of time well enough. Besides, being not so well-known, Wintec Industries products can offer a good financial win, because their AMPX modules are on average cheaper than the similar solutions from other makers.
Wintec 3AXH1600C9-4096K exterior doesn’t have any distinguishing features. These are common memory modules with standard stamped heat-spreaders colored black. They are shipped in traditional plastic packaging. Even the embossed AMPX logo on the heat-spreader is none other but a painted piece of plastic glued to the heat-spreader.
The official modules specs do not make us feel any better about the Wintec 3AXH1600C9-4096K modules, because they seem to be considerably weaker than what competitors have to offer:
Moreover, we couldn’t find the complete spec sheet for this memory type on the official Wintec Industries site.
However, the contents of the modules SPD turned out a real shocker! Part Number line for the modules from this maker reported that they belong to the DDR3-1600 Titanium Edition series from OCZ. I believe even the biggest skeptics will now have no doubts about the quality of Wintec memory modules: it turns out that OCZ sells the exact same memory modules under their own trademark. Moreover, if these particular modules are available under OCZ brand, they have much better specifications, namely, CAS Latency 7. Wintec’s specifications claim CAS Latency 9, which means we can expect great overclocking results.
Full set of SPD parameters of Wintec 3AXH1600C9-4096K kit also indicates that these modules can do much more than the manufacturer promises. The existing XMP profiles suggest using this memory with 8-8-8-28 and 7-6-6-28 timings, which prove quite operational. In other words, these memory modules are a great opportunity to save some money for those who care more about the actual functionality rather than the popular brand. In fact, Wintec offers top OCZ memory with different heat-spreaders at a low price.
Trying to shape up our verdict about the 4GB (2x2GB) dual-channel memory kits we got for this review, we tested them in order to determine their maximum stable operational frequency and bets timings configuration. For this purpose we used a perfectly suited Memtest86+ 2.11 benchmark testing how well the memory subsystem works under maximum load.
First we decided to find out the most aggressive timings configuration for the memory to work as DDR3-1600. We used the highest safe voltage setting of 1.9V for all memory kits except Aeneon XTUNE AXH860UD20-16H-K-4G that performed best at 1.6V voltage.
Only one kit of all tested - Aeneon XTUNE AXH860UD20-16H-K-4G – failed to work at 1600MHz frequency with CAS Latency 7. However, we can easily forgive it, taking into account phenomenally low voltage for an overclocker memory kit. As for other testing participants, we saw the best timings configurations by Mushkin 996601 4GB XP3-12800 and, surprisingly, Wintec 3AXH1600C9-4096K. They managed to outperform even the OCZ modules with much better original specifications.
Now let’s check out how far we can push the maximum frequency of our 4GB memory kits. We performed the first test with 9-9-9-27 timings and 1.9V voltage.
The results show the OCZ memory has fully rehabilitated itself. It overclocked to a record-high frequency, although it still couldn’t conquer a psychologically important DDR3-1800 barrier. The second prize goes to Aeneon modules that have suddenly revealed very good overclocking potential even despite low voltage setting. All other memory modules overclocked similarly except Corsair TW3X4G1600C9DHXNV kit that turned out a disappointment. Although it is one of the most expensive DDR3-1600 memory kits, its frequency potential is lower than that of all other competitors even with the “weakest” 9-9-9-27 timings.
The second overclocking test was intended to find out the maximum frequencies at which 2GB memory modules could work with aggressive 7-7-7-20 timings. This time we also set the voltage at 1.9V for all DDR3 SDRAM except Aeneon modules tested at 1.6V voltage setting.
And the winner with CAS Latency 7 is Wintec 3AXH1600C9-4096K! This is something out of the ordinary because according to the specs these memory modules are designed to work with CAS Latency 9! Nevertheless, the fact is undeniable: close family ties to the best OCZ products doesn’t go unnoticed in practical experiments. The “authentic” OCZ modules demonstrated very close results, too.
In fact, I have already made all the necessary comments above. Now all I should do is sum up everything I have already said and try to answer the question about how demanded DDR3 SDRAM is for contemporary LGA775 platforms. As our tests have revealed, the launch of new processors and chipsets has hardly changed anything. We have already expressed some doubts about the need to go for DDR3 SDRAM in a Core 2 based platform, and our opinion hasn’t really changed. If you cannot explain to yourself why exactly you need DDR3 SDRAM, then go for DDR2. In most cases, previous generation memory performs just as good as the fashionable DDR3 SDRAM, which is actually way more expensive.
There should be a serious reason for using DDR3 SDRAM in an LGA775 platform. Intention to perform some extreme overclocking experiments is good enough. And if you have made up your mind about getting DDR3 SDRAM, we would strongly advise to stay away from low-frequency products. The available DDR3-1333 SDRAM may only be of benefit in Core i7 platforms; LGA775 platforms with this memory perform as fast as platforms with DDR2-1067.
In this respect, the best performance per buck DDR3 choice will be DDR3-1600 SDRAM. However, you should keep in mind that only overclockers will be able to take advantage of all its features in contemporary LGA775 platforms. Only a significant front side bus frequency increase will let DDR3-1600 SDRAM seriously outperform previous-generation memory. Besides, with this memory running at high frequency you get extra flexibility in system parameters configuring. That is why DDR3-1600 SDRAM may come in very handy for computer enthusiasts trying to reach high FSB speeds.
As for the specific recommendations we could give you regarding 4GB DDR3-1600 SDRAM kits, our tests showed that it is pretty hard to recommend anything specific. Memory modules from difference manufacturers have different availability in retail channels and their price differs dramatically, while the practical functionality is often very similar. Overall, we would recommend to pay attention to the memory that works well with low latencies. Among the memory kits tested today these are the solutions from Mushkin and OCZ. And of course, we have to point out DDR3-1600 kit of two 2GB modules from Wintec. Despite relatively modest claimed specifications, it suddenly proved identical to one of the best solutions from OCZ