GullLars 12/31/2009 4:49 PM
Short summary of flaws in this article:
*Crystal Disk Mark (CDM) 2.2 is not suited for benchmarking SSD, and gives unrepresentative results. Use 3.x (beta) instead, wich has been remade to work with SSDs, and tests random 4KB at Queue Depth 32 in addition to 1 (wich 2.2 uses). Intel x25-M can do 120-160MB/s of 4KB random reads at Queue Depth 32, while crystal 2.2 only tests QD1.
*Leaving 20% of free capasity on a partition on a SSD does NOT increase performance. Leaving 20% unpartitioned however does if the drive has not been used. This shows a misunderstanding of SSD architecture.
*They have not pointed out or seemed to notice the reason the Kingston V and the Velociraptor falls behind in some PCmark tests is their appalling random write performance. The reason Kingston V beats the Velociraptor in some tests is simelarly the random read difference.
*They have not posted the PCmark scores of the tested drives.
*They stated that the reason for x25-M's high random write scores in CDM is TRIM while the others get lower results because they don't have TRIM. This is false. These numbers are entirely due to controller architecture, keeping them sustained high however does rely on TRIM.
*TRIM was not active during this test since Intel Matrix storage driver does not support TRIM, the default Microsoft Win7 ACHI driver however does.
[
Voor 96% gewijzigd door
Verwijderd op 02-03-2010 19:27
]