10/15/2004 | 11:22 AM
You should be familiar with PDP Systems already, but since it is the first time we introduce the products of this company to you on our site, let me say a few words about the company itself first.
PDP Systems was founded in 1985 in the USA with the headquarters in the Silicon Valley (California). The company also has official representation offices in Taiwan and Hong Kong. Besides the regular RAM PDP Systems also manufactures flash memory and ATI based graphics cards. The company is a member of JEDEC (Joint Electronic Device Engineering Council).
The product range offered by PDP Systems is simply enormous. Besides the low-cost budget memory solutions, they also product memory “for overclocking enthusiasts” featuring the lowest timings (2-2-2-5) and 400MHz working frequency, as well as the highest for DDR timings (3-4-4-8) and working at 533MHz frequency. The storage capacity of these modules varies from 128MB to 2GB.
Moreover, the company offers memory modules with ECC support, low-profile solutions for notebooks and now gets more and more into the promising DDR2 business.
Well, I assume you are now ready to meet our today’s testing participants: the six pairs of memory modules from Patriot:
Testing the memory in dual-channel mode is especially difficult for the memory modules. However, you may have already noticed that all modules are sold in pairs, they are already selected especially for dual-channel work. Moreover, testing the pairs of modules makes the testing conditions really close to real life, because most of the mainboard chipsets today support dual-channel memory access. Moreover, all contemporary Intel CPUs require as a must that the memory is used in dual-channel mode, otherwise you will not be able to achieve maximum performance.
Let’s start with the regular DDR400 memory modules belonging to the Signature Line family. The marking states that this set consists of two 256MB modules.
Despite the fact that this is a pretty regular memory set, and maybe for this particular reason, the manufacturer’s web-site contains most of the info about it. We managed to discover such details as the fact that these modules are manufactured on six-layer wafers, their nominal voltage is 2.6V +/-0.2V, and they work at CAS Latency equal to 3.0. Although the very last fact can also be found on the memory sticker:
And this is what the SPD of these memory modules contains:
Despite this fact, ASUS P4C800 Deluxe mainboard (i875P chipset, 1016 BIOS), which we used for our test session, set the timings to 2.5-3-3-6 at 200MHz frequency for all the tested modules. By the way, the manufacturer’s site states that in dual-channel mode the memory is tested on all the most widely spread chipsets, such as i875P, i865PE and NVIDIA nForce2 Ultra 400. And moreover, it is exactly ASUS P4C800 that takes part in this testing.
We ran the check in the way you are already familiar with from our previous memory tests. We sent 2.7V voltage to the memory modules. Although it is higher that the nominal it lies within the allowed range and is quite Ok for most mainboards. We also set the highest timings equal to 3.0-4-4-8 and then revealed the maximum working frequency for the memory to function stably. Then we reduced the memory timings to reveal the overclocking potential in these particular testing conditions and then we continued the same way until we reached the most aggressive timings of 2.0-2-2-5. The frequency changed with 5MHz increment. For our tests we used a CPU, which can work at 300MHz bus frequency, which theoretically allows us to test the memory working synchronously with the CPU at up to DDR600. We used Gold Memory program as a utility to check the memory’s stability during work.
Here I have to make a short comment. You do know that there are no ideal programs for memory check, as well as for CPU or video cards check. The fact that the test has been passed, indicates only relatively stable performance, but doesn’t guarantee stable performance throughout the entire service time. However, the opposite is also true: if the program reports an error, it does not necessarily mean that your hardware will not work in these conditions or will function unstably. For instance, one of my friends has been constantly getting an error in Prime95 every time he tried to check an overclocked CPU, as well as memory. However, his system works just fine, without being shut down for weeks, without a single failure. At it really does work, and not just stays powered on in the standby mode: besides the everyday routine, it is constantly involved in some calculations and is 100% loaded all the time.
We undertook a few experiments and found out that Gold Memory tests RAM much better than Windows memory Diagnostic for instance and at least not any worse than S&M. if the memory has been tested in Gold Memory, then S&M revealed no problems. That is why we considered Gold Memory to be quite enough for our tests.
The diagrams below indicate the nominal frequencies in blue (these are the frequencies the memory should work at according to the specs on the manufacturer’s website), the frequencies and timings written in SPD in green and the actual working frequencies during our test session in red.
This is a wonderful result! The diagram very illustratively shows that the memory works at considerably higher frequencies than it should and at much lower timing settings. When we shift to CL 2.0 the frequency drops down significantly, however, to tell the truth, I was surprised to see the memory working in these conditions at all. Keeping in mind that this is regular inexpensive memory, the results should be regarded as excellent. So, what are we up to next?
Now please meet two DDR400 modules 256MB each, which are intended to work in dual-channel mode with low timings set to 2.0-3-2-5.
However, there are completely different values in SPD.
Where is the truth? Let’s find out now!
Strange as it might seem, but the memory intended to work at not very high frequencies with low timings, proved pretty successful at quite high working frequencies with pretty high timings. 245MHz (DDR400) with the timings set to 3.0-4-4-8 is a very good achievement for this memory.
Now I would like to introduce to you a pair of modules from a different category: they can’t boast low memory timings, but are guaranteed to work fine with high timings at high frequencies. The family will be represented by a pair of “younger” modules classified as PC3200 (DDR446), which work at the nominal frequency of 233MHz. This pair of memory modules boasts the biggest capacity of all our today’s testing participants: 512MB each, making the total of 1GB.
Note that any memory faster than PC3200 (DDR400) is officially not standardized, that is why the manufacturer can use any characteristics he likes. In this case the memory should work fine at 233MHz frequency with 3.0-4-4-8 timings.
The info from the modules SPD proves this info correct:
We performed the check and here are the results we got:
We can look at these results from two viewpoints. On the one hand, everything is just fine and we saw a definite increase of the working frequency compared with the nominal one. On the other hand, the situation is quite frustrating because this memory cannot work with the minimal timings at all, while 245MHz is the maximum stable frequency. So why is this memory better than a pair of PSD512400K modules, which were the first ones to be tested today?
Here I would like to make one very important comment: PDC1G3700ELK memory modules really worked stably only at 245MHz, however, they could successfully start at 270MHz too. Though they failed to pass the tests at this frequency. Don’t you think that the gap between the starting 270MHz and operational 245MHz is a little bit too wide?
The first supposition we made dealt with overheating problems. If you remember, something like that has already happened one during our memory tests. When we tested takeMS memory modules, we used a low-speed Gigabyte 3D Cooler Ultra, which is a very quiet solution with low speed fan. However, the chipset North Bridge heatsink got overheated because there were no air flows inside the case, since ASUS P4P800 mainboard, just like ASUS P4C800 mainboard uses passive chipset cooling.
This time we used a Zalman CNPS-7000A-Cu cooler and the North Bridge temperature remained within the allowed values, however, the memory turned out to be heating up too much. The aluminum heatspreaders on the memory modules heated up to 50oC. When I installed a fan to cool down the memory modules a little bit we managed to increase the stably working frequency up to 255MHz, which is a pretty good achievement I should say.
All the tests were run without any additional cooling involved. That is why the result obtained with an extra fan is not displayed on the diagram below. But please keep in mind the following two things:
The next pair of modules is 2x256MB DDR500.
These memory modules should work at the given frequency with 3.0-4-4-8 timings.
And if we check out the info provided in the modules SPD, then it turns out that the timings should be even higher:
Now let’s turn to real applications to see if the specifications claimed by the manufacturer are correct:
Well, this memory refused to work with CAS Latency 2.0, just like the previous pair of modules. The maximum working frequency when the memory remained stable didn’t exceed the nominal value, however, the timings are lower than those written in the SPD so that we managed to reach 250MHz (DDR500) for the first time.
This is the fastest pair of modules out of all we have just taken a look at. It should work at the nominal frequency of 266MHz (DDR533) with 3.0-4-4-8 timings.
However, when we checked the SPD we were surprised to find out how greatly the info differed: I have never seen such high memory timings before, to tell the truth!
Our mainboard didn’t even allow setting such high timings. Well, let’s check out the benchmark results now:
And the results are quite promising I should say. We failed to exceed the nominal working frequency, but it is quite high already. And as for the timings, they turned out notably lower than the claimed ones. Moreover, there was one more surprise waiting for us: this memory turned out stable at CL2.0, although it wasn’t expected to.
Well, now we have finally reached the most interesting testing participants of our today’s session. The memory marked as XBL is a universal solution, combining the ability of the Erased Latency modules to work at high frequencies with that of Low Latency modules to support minimal timings.
The sticker doesn’t offer us much info.
While the specifications on the site reveal all possible combinations of working frequencies and memory timings.
I have never seen any memory like that before, and haven’t even heard that there are modules supporting these features.
The info read from the SPD is quite scarce:
However, this is the only pair of modules of all the solutions tested today that managed to start on ASUS P4C800 mainboard with the default timing settings at 200MHz frequency. All other modules started only with 2.5-3-3-6 timings no matter what the SPD said.
The first benchmark results almost struck me dumb: the memory started at 300MHz frequency, just as if it were DDR600! Although we failed to get it work stably at this frequency even when we installed additional cooling and raised the voltage to 2.85V. The final performance results are given below:
285MHz is a phenomenal overclocking achievement. It would correspond to DDR570 or PC4560, if it ever existed. Unbelievable result! Here is the memory performance in SiSoftware Sandra 2004 SP2. Please make sure you check out the list of competing solutions first:
Well, what can we say here? In fact, only one thing: Good bye, our good old Kingston HyperX 2x256MB memory modules. You have done a great job providing excellent performance with the minimal possible timings of 2.0-2-2-5 at 200MHYz working frequency. But you belong to the Low Latency family and didn’t overclock that well. 220MHz was the maximum we managed to achieve, which limited our flexibility during mainboard and CPU tests. These modules from Kingston can hardly compete with Patriot PDC5123200+XBLK memory modules, as we now get new opportunities in overclocking experiments (I already have a few ideas I could finally put into life with such a good memory as Patriot PDC5123200+XBLK).
Patriot memory left a very good impression after this test session. Each of the tested memory modules will find its customer in the today’s market, and each overclocker will be able to find something suitable for his needs from the patriot’s assortment.
Patriot Signature Line is the solutions family for lucky overclockers who tend to save a bit of money while still doing what they like. This is the best value memory offer from the considered list, while at the same time it overclocks very well and can even please you with pretty low memory timings in the best case.
If you are not among those risky users and know exactly what you expect from your system memory, then you should pay attention to the Patriot Erased Latency and Low Latency product lines. The first one cannot work with minimal timings, like the latter, but unlike the Low Latency modules support pretty high working frequencies. If you are an overclocker, both product lines are worth your attention, because the choice in this case only depends on your system and on the overclocking techniques you apply.
I would like to single out Patriot XBL memory, as this is a universal solution combining the features of both other memory types. It should be a good choice for almost any overclocker. Of course, these modules cost more than the regular ones, but if you are getting yourself a kit like that (a pair of dual-channel approved modules), then you do not have to worry about the compatibility and this investment will undoubtedly pay back in the future.