Depending on the action (backups, transaction log writes, etc.), SQL Server has several different types of IO sizes, so it is worth testing different block sizes.
I test with block sizes of 8K, 16K, 32K, 64K, 128K, and 256K for sequential read, sequential write, random read, and random write.
Depending on the disk size, I use a test file size of at least 50GB to make sure I am overcoming the impact of disk controller or SAN caching.
I am usually most interested in the 64K random read and random write performance, because many IO systems struggle to get good performance with these. I set up a new server recently, and saw a huge 64K random read and random write performance improvement when I reformatted a volume from the default 4K cluster size to 64K.