by Anton Shilov
04/16/2004 | 06:51 PM
Memory modules are an important component of a modern computer system; their quality largely determined its performance and reliability. The manufacturers of course see to the quality and seek for better characteristics, especially with regard to high-frequency modules like PC3200 and faster.
<%BANNER[article]%>As you know, PC3200 is the latest official memory standard accepted by JEDEC. Some firms come up with modules like PC3500, PC4000 and better, but these are mostly exotic products specifically served to overclockers. Today, there are several players of repute in the high-quality high-performing memory modules field: OCZ Technology, Corsair, GeIL and Kingston.
Today we are going to put a kit of two PC3200 memory modules from OCZ Technologies into the mincing machine of our test laboratory. The full name of the product is rather long: OCZ DDR PC-3200 Platinum Enhanced Bandwidth Dual Channel. This kit consists of two unregistered non-ECC modules, 512MB each, coming in a pretty transparent package:
There are also higher- and lower-capacity versions of this product, with two 1GB and two 256MB modules, respectively. Besides that, you can purchase “EB” modules separately (in 256, 512 and 1024MB capacities). This also refers to the OCZ DDR PC-3500 Platinum Enhanced Bandwidth series, since it only differs from the other family in the operational frequency.
All products belonging to the Platinum Enhanced Bandwidth series work with 2.6V Vmem, which is the norm for a majority of modern PC3200 modules. Let's take a closer look at the contents of the package:
The modules are neatly manufactured and equipped with silver heat-spreaders. When you look at the module from the back end, you see a gray thermal pad between the memory chips and the heat-spreader. Each module carries a sticker with the company's logo, name, timings and part number. The timings and the EB abbreviation are the most exciting things here.
As you know, the performance of the memory subsystem depends on many factors and the operational frequency of the module is as important as its latency. The manufacturers usually pay attention to the CAS Latency parameter, trying to make it as low as possible. This parameter may be 3, 2.5 or 2 for DDR SDRAM, and the last value is considered perfect. And really, practice suggests that setting CAS Latency to 2 we achieve the highest performance, while secondary settings like RAS# to CAS# Delay (tRCD), RAS# Precharge (tRP) and Raw Access Strobe (tRAS) affect it less strongly. I should note though that systems with Intel processors usually prefer more memory frequency rather than lower latencies, but low-latency modules are anyway faster than high-latency ones, with the same working frequencies.
Speediest specimens of PC3200 can work with the CL2-2-2-5 timings (CAS Latency-tRCD-tRP-tRAS), but the discussed memory from OCZ supports 2.5-3-2-8 timings. At the same time, the "EB" abbreviation in the product's name means Enhanced Bandwidth. So what's wrong? The manufacturer describes its Enhanced Bandwidth technology as “a means of increasing memory bandwidth through the optimization of memory latencies for the best possible interaction between the system memory and the chipset and memory controller”. In other words, OCZ Technology claims it has found the most efficient combination of latency settings, which allows CAS2.5 modules with EB technology to work like ordinary CAS2 modules and even faster, since the latencies in data transfer are reduced to the minimum.
Besides optimizing the latencies, the new technology uses the Variable Early Read Command function, available in DDR technology. The point of this function is in sending a read command during the ongoing data transfer, rather than after it. To be precise, the command for reading the next data block is sent 2 clock cycles before the end of the current data transfer with CL2; it is sent 2.5 cycles in advance with CL2.5 and so on. Thus, the influence of the CAS Latency parameter diminishes as it is actually the number of clock cycles that pass between the received data-read command and the moment the requested data appear on the bus.
If you are interested in Enhanced Bandwidth technology, you may go to the original source at the OCZ Technology Web-site.
We decided to test the memory kit from OCZ in two systems, with AMD Athlon 64 3400+ and Pentium 2.4C CPUs. Regrettably, for some unknown reason, the Intel platform turned to be insensitive to overclocking so we couldn't find out the maximum frequency the reviewed OCZ modules could work at.
As for the Athlon 64 platform, we got the following results: we increased the Vmem to 2.7V and the system booted up successfully with 220 (440DDR) MHz on the memory, but subsequently showed us the Blue Screen of Death during a test cycle in 3DMark03. Only after we pushed the Vmem up to 2.8V (the maximum possible voltage the ASUS K8V Deluxe mainboard could provide) the system was absolutely stable. As for the latency parameters, the system turned out unable to boot the OS every time we reduced them.
Thus, the OCZ DDR PC-3200 Platinum Enhanced Bandwidth memory will hardly suit an overclocker. Their overclockability is low, just like their ability to work with lower timings. However, if you want to have maximum performance at default settings, without bothering about hardcore overclocking, you should definitely consider this memory.
At first we wanted to carry out our tests on the Athlon 64 platform, but one curious obstacle arose. We used Corsair XMS3200 modules as a reference point for the OCZ memory and these modules can work all right at 200 (400DDR) MHz with CL2-2-2-5 timings. However, they have PC2700 written into their SPD. ASUS K8V Deluxe mainboard was misled by the SPD info and refused to enable DDR400 mode for the Corsair memory, limiting their frequency to 166 (333DDR) MHz. When we increased the frequency of the Athlon 64 clock generator to 220MHz, the frequency of the Corsair memory was only 183 (366DDR) MHz, while the OCZ memory was doing fine at 220 (440DDR) MHz. That's why we switched to the Pentium 4 platform, which had the following configuration:
Reading the SPD info by means of CPU-Z version 1.20a, we got the following:
CPU-Z version 1.20a
Memory Modules Serial Presence Detect (SPD)
Module #1
General
Memory type DDR-SDRAM
Manufacturer (ID) OCZ (7F7F7F7FB0000000)
Size 512 MBytes
Max bandwidth PC3200 (200 MHz)
Part number OCZ4001024EBDCPE-K
Module #2
General
Memory type DDR-SDRAM
Manufacturer (ID) OCZ (7F7F7F7FB0000000)
Size 512 MBytes
Max bandwidth PC3200 (200 MHz)
Part number OCZ4001024EBDCPE-K
Attributes
Number of banks 2
Data width 64 bits
Correction None
Registered no
Buffered no
Timings table
Frequency (MHz) 200
CAS# 2.5
RAS# to CAS# delay 3
RAS# Precharge 2
TRAS# 8
The SPD info is nothing new to us. The maximum frequency of the OCZ modules is 200 (400DDR) MHz, as it should be. All other parameters also comply with the specifications.
Now let's see whether OCZ's claims about the miracles of the Enhanced Bandwidth technology are true. Can CL2.5 modules be as fast as CL2 PC3200 ones? We open the cycle of tests by SiSoft Sandra 2004 Professional:
The results are strange enough. First of all, we see that the Corsair modules do work better with CL2.5-3-2-8 timings than with CL2-3-3-6. It seems like OCZ Technology has found the magic combination? Well, the modules from OCZ are actually slower than the Corsair memory at these timings - they are the slowest in this test, in fact, while the Corsair memory with CL2-2-2-5 timings is the fastest, as we might have expected. Well, at least this test confirms that CL2.5-3-2-8 is better than C2-3-3-6. Let's run other tests!
Now, there's a small victory. The OCZ EB modules get an excellent score in the PCMark 2002 Memory test, closely following the Corsair modules that work in the most aggressive mode. Note also that the difference between the OCZ memory with CL2.5-3-2-8 timings and the Corsair with CL2.5-3-2-8 is big: maybe the Variable Early Read Command influences the result?
The CPU test from the same PCMark 2002 suite shows a small divergence between different memory modules and timings.
The difference is clear, however, in a similar test from 3DMark03. Moreover, the new memory from OCZ beats the Corsair CL2-2-2-5, although not very hard. Other participants showed a smaller result, even with CL2.5-3-2-8.
The results are very close to each other, which is natural since it is the graphics subsystem that determines the result in this test. The CL2-3-3-6 combination fell somewhat behind, though.
It's all quite different in Unreal Tournament 2004. Corsair modules showed the highest result with CL2-2-2-5 timings, while the OCZ EB sticks took the last place. Anyway, the difference between the first and last positions is negligible (less than 1 frame per second). The difference is bigger on the Torlan level, but only amounts to 1.7 fps. Well, game engines from Epic have never been perfect: they always prefer lower latency to high data transfer rate. I should acknowledge, though, that these results tell very little.
The results are similar, save for the Corsair modules with the most aggressive timings. Anyway, the words of OCZ Technology about the EB technology making the CL2.5-3-2-8 combination as fast as CL2-3-3-6 are confirmed once again.
The results of our archiving in WinRAR 3.0 were most interesting - the Corsair modules with the CL2-2-2-5 formula lost to themselves working according to OCZ’s magic word "CL2.5-3-2-8". The combination proposed by OCZ is really optimal, but there's one problem - the OCZ EB modules that work with the same timings show much poorer scores. We can't say what the reason is. The Variable Early Read Command function may have influenced the performance badly, although it should have done otherwise, especially at data archiving when the memory bandwidth is more important than its latency.
Overall, the OCZ DDR PC-3200 Platinum Enhanced Bandwidth Dual Channel memory kit provides excellent performance and stability. The new memory will suit all users who are not involved into extreme overclocking of computer parts and want to have the maximum performance combined with high reliability.
The Enhanced Bandwidth technology does work and sometimes brings real benefits. In some cases, the OCZ Platinum EB memory works no worse than ordinary PC3200 CL2-3-3-6 modules or even faster. Theoretically, the CL2.5-3-2-8 combination provides better stability than CL2-2-2-5 or even CL2-3-3-6. The timings combination found by OCZ Technology is really good and we can recommend it to owners of PC3200 CL2-3-3-6 and CL2.5-3-3-6 modules; this will provide some performance gain, although small.
Alas, the OCZ Platinum EB modules can't work at timings different from their specifications, but this is not a very big disadvantage since they are quite fast at their default settings. In our case, their maximum operational frequency was 220 (440DDR) MHz, but with 2.8V Vmem. Higher voltage led to higher heat dissipation, but that didn't prove to be a problem.
So, the new memory module series from OCZ, with Enhanced Bandwidth technology, will suit nicely to all users who want to have efficient and reliable memory. The only group of users who may want to refrain from purchasing OCZ's EB-marked products is PC enthusiasts and overclockers. These modules will hardly work at timings different from those written into the SPD and will hardly please you with good overclocking. On the other hand, if you are interested in frequencies just above 200 (400DDR) MHz, consider the OCZ DDR PC-3500 Platinum Enhanced Bandwidth as an option.