VIA KT333 Motherboard Roundup - June 2002

IDE RAID 0 Performance Disk Winbench 99 (Transfer Rate in MB/s) Shuttle AK35GT2/R ABIT AT7 Soyo KT333 Dragon Ultra ABIT KX7-333 ECS K7VTA3 EPoX 8K3A+ MSI KT3 Ultra Gigabyte GA-7VRXP ASUS A7V333 62.0 62.0 61.8 61.7 60.4 60.0 59.8 59.7 59.0 | 0 | 12 | 25 | 37 | 50 | 62 | 74

When we first introduced this test, the motherboards with the Promise based solutions had some serious performance problems and couldn't hold a candle to the HighPoint offerings. This time around, the performance is virtually identical across the board which is good news for everyone. But what about Promise's 64KB stripe size limitation, does that get you into any trouble with performance? Let's compare the application performance of the HighPoint controllers with a 512KB stripe size to the only selectable stripe size (64KB) on the Promise chips.

IDE RAID Comparison High End Disk Winmark 99 (Transfer Rate in MB/s) Soyo KT333 Dragon Ultra ABIT KX7-333 ABIT AT7 Shuttle AK35GT2/R EPoX 8K3A+ ECS K7VTA3 ASUS A7V333 MSI KT3 Ultra Gigabyte GA-7VRXP 32.4 32.4 32.3 31.9 30.0 27.1 25.8 21.0 20.9 | 0 | 6 | 13 | 19 | 26 | 32 | 39

Here you can see the 64KB stripe size limitation of the Promise chip holding it back, although it's clear that not all of the performance limitations stem from the stripe size limitation. The five top performing boards all use the HighPoint chip (the AT7 uses the HPT374, the rest use the 372), the remaining boards use some variant of the Promise chip yet they offer very random performance. It seems like there are still implementation issues with the Promise chip on some of these boards, we shouldn't be seeing such great variations in performance between boards.

IDE RAID Comparison Disk Winbench 99 (CPU Usage - lower is better) MSI KT3 Ultra ECS K7VTA3 Shuttle AK35GT2/R Soyo KT333 Dragon Ultra Gigabyte GA-7VRXP ABIT KX7-333 EPoX 8K3A+ ABIT AT7 ASUS A7V333 1.76 1.85 1.86 1.87 1.95 2.27 2.59 2.61 3.56 | 0 | 1 | 1 | 2 | 3 | 4 | 4

As you can see here, there's very little correlation between CPU utilization and the chip used as well as the performance of the particular solution.