Regular tests are done by writing and reading true files. However, special Windows API functions are used to disable file caching. The maximum file size used is 1GB. For larger cards, several files are used to fill the card. All data transfers are from RAM to Card Reader or Card Reader to RAM, so there is no influence from another media.
As test data, a specially designed data sequence is used. Some interfaces have data dependent encoding, the actual speed for those interfaces might be higher or lower, depending on the actual data. Some controllers allow higher write speed for certain regular data patterns. Most - if not all - benchmark tools will get fooled by such controllers and report a much higher data rate than what can be achieved with real-world data. The easiest way to check the actual write performance is to drag&drop a large video file from a fast harddisk to the card, and check the time until the activity light of the card reader stops.
Some flash cards suffer from unaligned writes. While
SD Cards and Memory Stick PRO have strict rules how to
set up the filesystem, in order to get the clusters
aligned, CF cards are partitioned and formated in the
same way as harddisks. Proper CF card controller will
show about 1% or less speed penalty for unaligned writes,
but some SLC cards can drop to 80% of their maximum performance,
while MLC cards can easily drop to 50% or even less
of their maximum performance.
The initial card checks include alignment tests, where vulnerable cards get a properly aligned filesystem. Unfortunately, this process is time-consuming, hence there are no scientific test results available at the moment.
Detailed Test Results
These special tests are done by writing directly to the media, and reading directly from the media, in order to get closer to the maximum speed, without the influence of the filesystem.