Fast RAM Provides Low Value

Slim gains from low-latency RAM don't justify a steep price.

High-speed memory modules from vendors Corsair, Crucial, and Kingston. With more retailers and boutique vendors touting high-speed (low-latency) RAM, two questions leap to mind: What is it, and do I need it? Last things first. Fast RAM is almost never worth the money, unless you're a dedicated gamer or gearhead obsessed with wringing every last millisecond of performance from your system.

Low-latency RAM takes less time to fetch the first bit of data after the CPU sends a column address strobe (CAS) signal to main memory. Such differences add up in tiny fractions of a second as your system processes instructions for the game you're playing or the database you're working with. Vendors list latencies for various memory functions; the first number is the CAS latency (CL), and when it's down to 2 on Intel-based systems (2.5 with AMD CPUs), you're really humming. Such memory carries a high price premium, often 20 to 40 percent more than slower RAM. A 512MB DDR400 module from Crucial, for example, costs about $100 at CL3 and $139 at CL2.

Vendors are careful to downplay any speed advantages. Latency differences may give you only a 2 percent boost in system performance, says John Stroozas, Crucial's director of engineering.

To gauge the benefits, PC World tested a PC with an AMD Athlon 64 CPU and standard memory, and then we swapped in faster RAM from Corsair Memory, Crucial Technology, and Kingston Technology, in turn. Improvement was negligible in most of our tests (see chart). Gaming tests showed the most benefit, with 4 to 5 percent gains in some cases. There was little difference among memory brands.

Low-latency RAM should eventually drop in price. But for now, if you didn't know what CAS was before reading this article, you probably don't need low-latency memory.

Low-latency DDR memory provides a modest boost to large-data-set, graphics-intensive applications (like the Doom 3 game), which make frequent random accesses of main memory. But the performance gains mostly disappear at higher resolutions for games, where systems rely more on high-powered graphics cards than on the CPU/memory subsystem. Mainstream apps show minimal improvement.

FOOTNOTES:
1Frames per second, at 640 by 480 resolution and 32-bit color depth.
2Frames per second, at 1600 by 1200 resolution and 32-bit color depth.
3Frames per second, at 1024 by 768 resolution and 16-bit color depth.

HOW WE TEST: We tested an AMD Athlon 64-based system with Windows XP Professional using a pair of low-latency DDR RAM modules, then a pair of standard DDR RAM modules, at both 512MB total and 1GB total. We used low-latency parts from Corsair, Crucial, and Kingston, in turn. See the WorldBench index page for details on WorldBench 5. In the Mozilla subset of tests, we time the loading of 115 Web pages, four times. For Doom 3, we measure frame rates during the supplied timed demo. In Return to Castle Wolfenstein, we measure frame rates during a recorded demo. The Unreal Tournament 2003 test measures frame rates during a fly-through of one game level. All tests developed and performed by the PC World Test Center. All rights reserved.

CHART NOTES: On the Mozilla test, lower is better; elsewhere, higher is better. Bold denotes best score.

David Essex