X-bit labs

DDR2 vs. DDR: Revenge Gained (page 5)

Category: Memory

by Ilya Gavrichenkov

[ 12/17/2004 | 01:46 PM ]

DDR2 SDRAM vs. DDR SDRAM: Testbed and Methods

So, it’s time to knock together two platforms: i875/i865 with DDR memory and i925/i915 with DDR2 SDRAM. We will see how the performance of the systems varies depending on the memory timings. We used identical processors and graphics cards in both platforms, so they only differed in the memory type. We took an i865PE-based P4P800-Deluxe mainboard from ASUS as an i875/i865 platform (thanks to its HyperPath technology, this mainboard delivers the same performance as i875P-based products); the i925/i915 platform was based on the ABIT AA8 DuraMAX mainboard with the i925X Express chipset.

Thus, we used the following hardware in our tests:

• Processors:
• Intel Pentium 4 550J (LGA775, 3.4GHz);
• Intel Pentium 4 3.4E GHz (Socket 478, Prescott core).
• Mainboards:
• ABIT AA8 Duramax (LGA775, i925X Express).
• ASUS P4P800-E Deluxe (Socket 478, i865PE).
• Graphics cards:
• Sapphire RADEON X800 XT (AGP 8x);
• Sapphire RADEON X800 XT (PCI Express x16).
• Disk subsystem:
• Maxtor MaXLine III 250GB (SATA150).

First, we’d like to offer you the results of synthetic memory subsystem benchmarks:

The first thing you should have noted in the table is the relatively small difference in bandwidth between various memory types. This is natural since the tests just cannot measure the “pure” memory bandwidth and the numbers are the effective bandwidth of the CPU-memory thoroughfare. The bottleneck in this thoroughfare is not the memory proper, but rather the CPU bus that has a theoretical peak bandwidth of 6.4GB/s at 200MHz frequency. The tested memory just can’t reveal its bandwidth potential here.

Anyway, we can still see the dependence of the bandwidth on the memory type and its characteristics. DDR2-533 with 3-2-2 timings has the best result, closely followed by DDR2-600 and DDR533. The memory subsystem bandwidth with DDR2-533 and 3-3-3 timings equals roughly the bandwidth of DDR400 SDRAM. DDR2-533 SDRAM with 4-4-4 timings which we used in our earlier tests has the worst bandwidth according to the synthetic tests.

As for latency, the rare DDR2-533 memory with 3-2-2 timings has the best latency of all. The more widespread DDR2-533 with 3-3-3 timings has about the same latency as ordinary DDR SDRAM. DDR2-600 and DDR2-533 with 4-4-4 timings make up a group of outsiders.

Thus, the synthetic tests suggest that the reduction of the timings of DDR2-533 memory can pull the performance of LGA775 systems up above the performance of older i865/i875-based platforms. The recently released DDR533 SDRAM with 2.5-3-3 timings cannot save the day, although adds to the speed of last-generation systems. As for the artificially implemented DDR2-600 mode, it doesn’t seem to add any performance at all.