Testing Overview
For DDR2 testing, I’ll be using Icrontic’s Intel LGA 775 test rig consisting of the following:
- CPU: Intel Core2Quad Q6600 @ varying frequencies.
- Motherboard: Asus P5K-E based on the Intel P35 chipset.
- Graphics Card: Generic passively cooled PCI based card.
- Hard Drive(s): 1x Maxtor Diamond Max 10 SATA drive.
- Optical Drive(s): Pioneer SATA DVD-RW drive.
- CPU Cooling: Noctua NH-U12P with 53CFM Noctua NF-P12 fan.
- Memory Cooling: 1x45CFM 92mm fan approximately 3cm from the modules.
- Case: Antec mid-tower with rear 120mm 45CFM exhaust fan (side left open)
- Power Supply: Corsair TX750
Environmental controls were also put in place. Ambient temperature was maintained at 19’C for all tests and was not permitted to deviate more than +/- 0.5’C.
Stability Testing:
Two applications were used for stability testing—Memtest86+ and Prime95. Memtest86+ is the industry standard in memory stability testing. It provides a quick and easy way to perform stability testing. Test number five is most demanding and is looped to look for any errors. For a test to qualify as a ‘PASS’ in Memtest86+, it must pass at least three rounds of test number five.
As good as Memtest86+ is at detecting memory instability, its weakness is that it does not put high levels of strain on the CPU and other critical components that may impact overall system stability. It is not unusual for memory to pass loop after loop of memtest, but fail under the grueling conditions that multiple instances of Prime95 can provide. A custom Prime95 test is run for our purposes, forcing Prime95 to utilize approximately 1500MB of system memory. The full 2GB can not be allocated due to the operating system utilization and the nasty hard disk paging that would occur if we increased it further.
For our purposes, all instances of Prime95 must not fail for a minimum of 30 minutes for a kit to receive a ‘PASS’. Once I get a complete picture of stability and frequency/vDIMM scaling in Memtest86+, results are validated and adjusted using Prime95.
Testing Criteria
There are numerous things I look for when I put a memory kit through the Icrontic labs. Some may be more obvious than others.
Heatspreader contact and thermal considerations: Heatspreader contact is a very important observation that is made on all modules in the Icrontic lab. In order for a heatspreader to work correctly, it must make good contact with the ICs. Observations about thermal pad interface and any gaps will be made. Although most modules employ simple heatspreaders, others have elaborate cooling solutions. All modules tested will be run at full load without any active cooling for a period of time. Contrary to popular belief, if the heatspreaders get hot, they are working correctly. Heatspreaders that do not get toasty could be insulating the ICs. Improper contact or thermal pads that are too thick can contribute to this. Special cooling solutions will be analyzed for their design and effectiveness.
SPD profiles and compatibility: One of the biggest problems the memory industry has been tackling is initial boot up compatibility. Many boards do not provide enough vDIMM by default to allow modules to run at their rated specifications. This is especially true of PC2-8500 memory, which usually requires 2.2V—much higher than the JDEC specified 1.8V for DDR2 memory. If the SPD profile is not set modestly enough, the system may fail to boot in some circumstances. Most manufacturers program the SPD defaults to a lower frequency that is known to be stable at 1.8V. The challenge is that individuals not comfortable with the BIOS will likely leave their modules as-is thinking they are defective. This is where EPP or ‘Enhanced Performance Profiles’ come into play. Nvidia and Corsair worked jointly to develop additional SPD profiles that allow EPP enabled motherboards to automatically set frequency and vDIMM correctly. During testing, we take a look at all of the programmed SPD profiles and comment accordingly.
Overclocking and Performance: Overclocking headroom is perhaps one of the most important features for enthusiasts. Aside from physical differences relating to thermal considerations, two different modules running at the same frequencies and timings perform identically. As such, there is really no point in benchmarking five PC2-8500 kits all running at their default specifications. Any difference will be within the margin of error. Overclocking headroom is what really sets modules apart.
As shown above, increasing frequency has more of an impact on memory bandwidth and latency than tightening timings. PC2-5300 memory running at 3-3-3 timings does not offer the high bandwidth and low latency that PC2-8500 does at 5-5-5 timings. This is true of both AMD and Intel platforms. As such, the higher the frequency at 5-5-5 timings, the better. There may be situations where high frequencies are not possible, perhaps due to mainboard limitations. In these situations, tightening the timings may be the best course of action to tune memory performance.
You may notice that I have not included benchmarks for the overclocking results in this roundup—there is a reason for this. It is very difficult to get representative performance figures resulting from memory performance increases alone. Increasing the front side bus frequency is necessary to increase memory frequencies, and that will also increase the CPU frequency. The CPU and FSB frequency increases would likely account for most of the performance increase shown in real world benchmarks. Even synthetic bandwidth and latency benchmarks are impacted by this. Because we are comparing numerous kits with varying overclocking potential, it is not fair to say that ‘Kit A’ can provide ‘20%’ better performance in Team Fortress 2 than kit ‘B’. Almost all of that performance benefit may stem from increased CPU frequency and FSB bandwidth inadvertently increased while overclocking ‘Kit A’ past ‘Kit B’. It is pretty simple to come to the conclusion that higher memory frequencies allow for greater tweaking potential and increased performance, so I’ll be focusing on maximum acheivable frequencies rather than application specific performance increases.
Power consumption: Although memory modules account for only a small fraction of a PC’s overall power draw, I believe in todays energy conscious world, “every little bit counts”. Power measurements are taken while each set of modules completes Memtest86+ stability testing. As you’ll see in the ‘Power Consumption’ section, there is a difference from module to module.
Price and warranty: These two criteria are important to just about every buyer. “Bang for the buck” is a term I use frequently in my reviews. Just because a dual channel kit is expensive, does not mean that I won’t recommend it. Price must be justified through all of the other criteria mentioned. In regards to warranty, the industry standard is ‘lifetime’. Reputable memory manufacturers stand behind their products.
Aesthetics: There will always be those who couldn’t care what a module looks like so long as it performs well. Lets face it though, some enthusiasts want good looking modules. Black PCB and colorful heatspreaders and unique cooling systems generally take the cake with these individuals. I really don’t let aesthetics sway my final judgement on a particular product as I believe the majority of readers prefer function over form. I will, however, comment where appropriate.
Articles RSS