This setup resembles the ultimate challenge for the WD1500 Raptor. Will two WD4000KD drives in RAID 0 outperform the Raptor in every benchmark?
State-of-the-Art: 10,000 RPM RAID 0
Although this scenario is by far the most expensive - it requires two WD Raptor drives - it certainly is the most desirable as well. Two 10,000 RPM Raptors in a RAID 0 setup should blow everything else away easily.
In theory, RAID 0 promises to be an ideal performance array, as the sequential transfer performance multiplies almost linearly with the number of drives added to the array. Files are distributed block by block across all drives, which allows the RAID controller to write data onto all drives almost simultaneously. Transfer performance measurably and noticeably increases for almost all application scenarios, though access times cannot be shortened. In real life, access times usually suffer from a marginal increase in RAID 0 setups. This is no reason for worry, as we're talking about half a millisecond, which is nothing you will ever notice.
The storage controller may in fact become a bottleneck if you set up a RAID array with multiple drives. PCI devices max out at 133 MB/s, which even two hard drives reach easily today. Serial ATA controllers that are part of a motherboard chipset usually provide much stronger backbones and typically don't limit RAID performance.
We measured up to 350 MB/s using four 10,000 RPM WD Raptor drives and Intel's ICH7 or ICH8 chipset components. This is an excellent result, as it almost matches the added throughput of four single drives. At the same time, Nvidia's nForce 680 chipset somehow showed a 110 MB/s bottleneck, which we still couldn't overcome. It shows that not every integrated controller is suitable of hosting a high-performance RAID 0, despite technically supporting it.