Single page Print

Memory performance
I'm not really comfortable benchmarking retail hardware with a beta operating system or drivers, but I made an exception for Sandra's memory bandwidth test to show off what a NUMA-aware OS can do for the Thunder K8W's memory bandwidth. I used the publicly available Windows XP-64 beta and a plain 32-bit version of SiSoft Sandra 2004.

With NUMA enabled in WinXP-64, the Thunder K8W doubles its memory bandwidth and scores over 9500MB/sec. In short, NUMA rules. Unfortunately, the only operating systems to support AMD64 NUMA are beta versions of Windows and a handful of Linux distributions.

Looking at our 32-bit results, the Tiger K8W's memory bandwidth is disappointing. Given that all three boards are using the same Opteron integrated memory controller, I expected the Tiger to be more competitive.

The Tiger's lackluster performance continues in Cachemem's bandwidth tests, but the results are a little closer this time. The Thunder K8W leads the field in read memory bandwidth, but it's slightly behind the K8T Master2 in write bandwidth.

In Cachemem's memory latency test, the Thunder K8W comes out on top again, trailed by the K8T Master2 and the Tiger K8W.

Disk controller performance
Our disk controller performance tests use a Maxtor 740X-6L 7,200RPM hard drive for "parallel" ATA (PATA) and a Western Digital Raptor WD360GD 10,000RPM hard drive for Serial ATA (SATA). Because we use different drives for PATA and SATA, scores aren't comparable between the two. PATA scores should only be compared with each other. The same goes for SATA scores.

The boards are all closely matched in terms of "parallel" ATA disk performance, but the K8T Master2's south bridge Serial ATA offers much faster burst speeds than the K8Ws' PCI-bound Sil 3114 SATA controller.

Disk access times are closely matched across the board, but the K8T Master2 comes out ahead again when we look at CPU utilization. The K8Ws' Sil 3114 SATA controller is particularly greedy with CPU cycles.