Navigate: Advice Beginners BIOS Guide CPUs Links Mainboards Memory Network Storage Video/Sound Cards Contact Forum SiteMap Sponsors WebNews Home	.	.
Prices: Mainboards ABIT ASUS Chaintech Shuttle Soyo Tyan CPU Intel P4 2.4C-800 P4 2.6C-800 P4 2.8C-800 P4 3.0-800 P4 3.2-800 AMD AthlonXP XP 1700+ XP 2000+ XP 2400+ XP 2500+ XP 2700+ XP 3000+ XP 3200+ Athlon64 Athlon64 3200+ Athlon64 FX-51 Opteron Opteron 240 Opteron 242 Opteron 244 Opteron 246 Memory Corsair Crucial Kingston Mushkin OCZ Search Prices:

LOSTCIRCUITS	SHORTCUTS: The Missing Pin The Numbers Game / Test Configurations Latencies Memory Bandwidty Sysmark2004 Sysmark Continued 3ds max, Amorphium3 Cinebench3 Gaming Benchmarks More Gaming Benchmarks Overclocking, Conclusions Give Us Some Feedback to Help Us Improve our Reviews
AMD Athlon64 FX53 Back to the top
(Review by MS, July 1, 2004)

Memory Bandwidth: SiSoft Sandra

The probably most widely distributed memory benchmark is SiSoftware's Sandra memory benchmark. There is not much that has to be said about Sandra other than that there are two different modes in which the benchmark can be run. The default uses prefetching and buffering on the chipset level, that is, the chipset can use more or less intelligent prediction algorithms to prefetch data in bus idle periods and, therefore, increase the bus utilization. Needless to say that in that case the overall performance is going to be substantially higher than without prefetching or buffering enabled. We have the full gamut of the buffered benchmarks across most current platforms and show the comparison between the ASUS SK8V (FX53) and the K8V (3800+). All benchmarks on the Athlon64 3800+ platform were run using the OCZ PC3700 EB modules unless otherwise indicated.

Buffering & Prefetch Enabled

In raw streaming benchmarks such as Sandra with buffering / prefetching enabled, the Socket 939 pulls ahead with a substantial lead, even over the so far undisputed champion in this area, that is, the Socket 940 platform.

Buffering & Prefetch Disabled

It has been shown in a number of reviews (e.g. Anandtech) that the unbuffered bandwidth greatly depends on the number of populated ranks. The rationale in that case is that the number of open pages equals the number of populated ranks multiplied by the number of internal memory banks (4 per device) and that the page hit vs. miss probability increases with the number of open banks. Interestingly, though, there is also an impact of the distribution of ranks across the memory channels, that is, whether only one "dual" channel (128-bit wide) is populated or whether the physical memory banks are associated with four separate DIMM slots. (1): 2:3:2; (2): 2.5:2:2.

next page:    => General Usage: Sysmark2004 =>