DDR2 RAM

Premium-grade RAM always photographs well. System memory comes in all shapes, sizes and colours, and is often adorned with heavy, intricate heatsinks or simple, pretty heatspreaders; you may even be forgiven for choosing the kit that looks the most impressive.

Price, looks and raw performance don't always go hand-in-hand, however. As our testing shows, there are plenty of great bargains and false economies of which to be aware before buying.

When building your first PC, it's easy to overlook the importance of memory. For most people, the temptation is to scrimp and save on the RAM, and buy a more expensive graphics card instead. But for overclockers who know how much bog-standard RAM can hold back your system, only the good stuff will do.

DDR3 memory is the latest and most expensive memory to hit the market. However, switching to DDR3 requires a motherboard upgrade and, as only one core logic chipset currently supports DDR3 anyway, it will be a long time yet before the majority of people even consider upgrading to this platform. The good news is that systems based around DDR2 are still capable of delivering daunting amounts of speed and DDR2 is currently at its cheapest - almost 250 per cent less than it was a year ago.

Two 1GB sticks of DDR2 memory are, and probably will be for some time, the sweet spot when it comes to price and performance. In this Labs test, we reviewed 19 2GB matched-pair kits in order to find out which kit is the best buy for your system, whether it's a budget gaming PC or an overclocked powerhouse.

Before beginning this Labs test, we had already decided to focus on matched pairs of 1GB DIMMs, as these currently provide the best balance between performance and price. A modern PC needs at least 2GB of RAM if you want it to run the latest games and applications. Not only are 4GB kits much more expensive, but even with the 64-bit versions of Windows Vista, there's still a 3.5GB limit on how much memory a single application can address.

For this Labs test, we decided to concentrate our test results on how well each matched pair of DIMMs performed in an LGA775 Core 2 system, as this is the platform that enthusiasts aspire to own at the moment. We began testing using an Asus Striker Extreme motherboard based on the nForce 680i SLI core logic chipset, but soon found that many kits simply wouldn't work at their advertised settings in this motherboard. Later, despite the nForce 680i SLI's excellent overclocking credentials, we also discovered that many kits wouldn't overclock as far as we expected.

For this reason, we decided to begin the Labs test again, this time using an Asus P5K Deluxe WiFi-AP motherboard based on the Intel P35 core logic chipset. Although this motherboard can't overclock the FSB as far as the Striker Extreme (480MHz vs 500MHz), thanks to its superior Northbridge, it was able to overclock the RAM much further and had far fewer compatibility issues. As a general rule of thumb, in the P5K Deluxe WiFi-AP, we were able to overclock most kits between 100MHz and 200MHz further than we could with the Striker Extreme. The P5K Deluxe WiFi-AP also supports more FSB:RAM dividers than most motherboards, allowing us to set the RAM to run at either 533, 667, 800, 889 or 1,066MHz. This meant that we were able to test nearly all the kits at their native frequencies, with the exception of the 1,000MHz and 1,150MHz kits. Since overclocking the RAM or CPU would have unfairly affected the test results, we decided to test the former at 899MHz and the latter at 1,066MHz, as this is the maximum frequency at which the modules would run in a non-overclocked PC. However, these modules still had a chance to run at their rated speeds (or faster) in the overclocking tests. Accompanying the P5K Deluxe WiFi-AP was a 2.66GHz Core 2 Duo E6700 CPU, GeForce 7900 GTX graphics card and Samsung SpinPoint P120S hard disk.

As Socket AM2 Athlon 64 systems theoretically receive more benefit from low-latency RAM than Core 2 systems, we also tested some of the kits in a Socket AM2 system using an Asus Crosshair motherboard, which is based on the nForce 590 SLI core logic chipset. The results of these tests can be found on p87. The remainder of the test system comprised a 2.4GHz Athlon 64 3800+ CPU, GeForce 7900 GTX graphics card and Samsung SpinPoint P120S hard disk. The choice of which model of Athlon 64 to use for this test rig was critical; due to Athlon 64 CPUs lacking a traditional FSB, the memory frequency in an Athlon 64 system is determined by a number of factors, as described in the formula DRAM clock = CPU clock/ceil (CPU multiplier/memory divider), in which ceil is a function that returns the highest integer value not less than the argument. The Athlon 64 3800+ allowed us to test the 800MHz DDR2 kits at their native frequency, which is the highest-frequency RAM supported by Socket AM2 systems without overclocking the CPU. We first tested each kit at the frequency, latency timings and voltage settings guaranteed by its manufacturer. It's important to note that these settings are nearly always different from the SPD settings stored in the RAM - the SPD settings are typically quite conservative so that the RAM will work in any system. If you want to get the most from your RAM, it's important to find out the voltage at which the modules are guaranteed to run. After nearly a year since our last DDR2 Labs test, we'd still like to see some improvement from the manufacturers, as these settings aren't always printed on the label or packaging, necessitating a visit to the manufacturer's website. Even then, it isn't always possible to find the information you need without contacting the company.

With each kit configured to run at its guaranteed settings, we ran a series of two benchmarks to test performance. First up was the Paint Shop Pro image editing test from the Custom PC Media Benchmarks, which uses up to 2GB of RAM while opening, closing and editing multiple high-resolution digital photos. We also ran the performance test built into F.E.A.R. at 1,280 x 960 with 2x AA and 2x AF to find out how much impact different RAM frequencies and latency timings have on game frame rates.

Once we had tested each kit on both test rigs using these benchmarks, we proceeded to overclock each kit. We first tried to overclock each kit to the maximum stable frequency at its guaranteed latency timings, increasing the voltage running through the RAM as and when required. We then relaxed the latency timings to 6 - 6 - 6 - 18 to find out if the RAM would overclock any further. To make sure that the RAM was stable and there was no data corruption occurring, we ran two instances of Windows Memtest at 100 per cent load on each kit. If Memtest discovered any errors, we raised the voltage, and then tried lowering the frequency until it passed the test.

The performance data can be found on the graphs, along with a graph displaying the results from our overclocking experiments. This information was combined to create the scores given to each kit. The Speed score is based on the performance of the RAM in the two benchmarks, plus the highest frequency to which the RAM would overclock stably. The Features score was calculated by examining the cooling system, the type of warranty that covers the RAM and whether it supports EPP (which enables easy overclocking in supported Nvidia-based motherboards). The Value score was calculated by combining the Speed and Features scores, then dividing this by the price. The Overall score is a weighted average of the first three scores.