Sequential throughputsFor these speeds, we used IOMeter. This very flexible piece of software allows us among other things to work either with standard data used by most benchmarks and which is easily compressible (series of 0s and 1s), or with random data that can’t be compressed. In the first case, SandForce’s DuraWrite makes a difference but not in the second, where it has to write and reread almost all the data bit by bit. The random data tests are labelled with an “/R” in the graphs. In practice, the large files that are moved around on a storage device are already compressed (JPEGs, DIVXs, AVC and so on) and so speeds obtained with this random data is in reality the more realistic.
Although small capacity SSDs do pretty well on reads, attaining speeds of up to 125 MB/s for the Onyx 32 GB and 266 MB/s for the Crucial C300 64 GB, the write scores aren’t as good. Writes have always been slower on flash and using controllers with fewer channels and/or less dense flash memory has a very visible impact on writes.
Although the Corsair and OCZ SandForces do well with highly compressible data, when the data can’t be compressed speeds drop to 70 MB/s on the 40 GB versions and 93.6 MB/s on the 60 GBs. Even worse, we’re at 74 MB/s on the Crucial C300 64 GB, under 65 MB/s on the Onyxs and even as low as 41.9 MB/s on the Intel X40-V! In fact, only the M225 64 GB, combining an Indilinx controller and old 43nm Samsung memory, holds firm.
We should however say that although these figures are low, they aren’t terrible: on such a small storage device, you won’t be moving large files around very much, which means that sequential write performance shouldn’t be too much of a limitation. Note that with SATA 6 Gbits, the C300 64 GB exceeds 300 MB/s, like the 128 and 256 GB versions. Nevetheless SATA 6 Gbits necessarily implies a Marvell controller and therefore Microsoft drivers for the TRIM command to work, a combination which gives lower performance than SATA 3 Gbits with Intel drivers in the other domains and therefore not necessarily the option to go for.
Staying with sequential accesses, here we tested with small 4 KB blocks. This puts the controller to the test to a maximum in terms of ins / outs. The fastest for reads here is the X25-V 40 GB, while for writes, outside of the SandForces with compressible data, only the old M225 gets away unscathed. Note that the Agility 2 is here a good deal slower than the Corsair F40: this is linked to the fact that the firmware on the Agility 2 limits it to 10,000 IOPS, which isn’t the case on the F40 which can go as high as 50,000 IOPS, as can the Vertex 2.