Four-Corner Performance Testing
Flash-based products pose more testing challenges than mechanical storage products. The workload you run before a test affects the results you see immediately after. In order to make true apples to apples comparisons between products, a strict benchmark regimen must be adhered to. With some products, the type of data used for testing, whether it's compressible or not, changes the performance story.
SMI's SM2256 hits the upper performance ceiling of SATA 6Gb/s surprisingly fast in our sequential read test. The SM2246 already used in several commercially available SSDs demonstrates the same high performance at low queue depths.
At a queue depth of two, our SM2256 sample with Samsung flash falls slightly behind Samsung's own 850 Pro and SanDisk's Extreme Pro at the same capacity point.
Going forward, the difference between value- and high-performance products will come down to write performance. Most companies already ship less expensive drives with 128Gb die, so each IC holds twice as much data as performance-oriented SSDs with lower-density flash. Why is this? The higher the die capacity, the fewer dies there are for parallelizing reads and writes. In short, the more dies you can write to at any given time, the faster that operation occurs.
The SM2256 is a four-channel controller, so it's already at a disadvantage compared to eight-channel designs. That's one reason we see this drive at the lower end of our sequential write test.
4KB Random Read Performance
The SM2256 ends up behind many of today's drives in measures of random read performance. What I'd really like to know is what will happen when read retries intensify towards the end of the drive's life cycle. Samsung 840 EVO owners are already dealing with these issues, despite the company's purported firmware fix. This is why LDPC is so important to the future of low-cost SSDs.
4KB Random Write Performance
Our sample also falls toward the bottom of our random write performance results.