Serial ATA performance - IOMeter
Next, we tackle IOMeter, which hammers drives with increasing I/O loads. We've restricted our testing to IOMeter's workstation and database access patterns because they should be more relevant to desktop users than the file or web server test patterns. IOMeter makes good use of the Native Command Queuing capability built into the AHCI specification, and as one might expect, it loves the quick access times of solid-state drives.

The mechanical VelociRaptor is up first, and its transaction rates don't vary much from one chipset to the next. Notice that all three hit a wall at 32 concurrent I/O requests. That just happens to be the depth of the Native Command, er, Queue.

Keep in mind that we're using a different CPU with each chipset. The 890GX's CPU utilization may be a little higher in IOMeter, but it's working with two fewer threads than the Hyper-Threading processors in our P55 and P67 systems.

With the RealSSD, IOMeter is able to extract additional performance from the 6GBps SATA controllers in the P67 and 890GX. The two are equally matched, although the P67 does have a higher transaction rate with less demanding loads. Even the P55 is quicker there. However, its 3Gbps controller can't match the peak transaction rates of the 6Gbps alternatives.

Once again, we're hitting a wall at 32 concurrent I/O requests. We've seen IOMeter transaction rates continue to scale upward after 32 I/Os with older Intel drivers and the standard AHCI drivers built into Windows 7. According to Intel, optimizations to improve performance with "common client workloads" are responsible for the change in behavior of its drivers. AMD appears to be pursuing a similar optimization strategy.

What were minor differences in CPU utilization with a mechanical hard drive are amplified by our SSD. Once again, the 890GX system consumes more CPU cycles than the Intel ones.