Question: When considering to buy a SSD, how should I interpret the different specifications of the SSD?
Here are some specific things that need to be deciphered:
- Controller (this can affect performance and endurance more than all other factors combined)
- Bus Technology
- Form Factor (Physical Size)
- Capacity
- NAND or NOR technology
- Power Consumption during Read, during Write, when Idle
- Read/Write Burst and Sustained Throughput
All of these things I would like to be explained in more detail and their actual importance in selecting an SSD.
Answer: I have a couple of laptops with SSDs – a recent MacBook Air and a 64GB Kingston V100 placed into a Lenovo Thinkpad T60p. ?Both are fast – the Thinkpad is much faster than previously, with boot times below 30 seconds, and much improved battery life.
Your points:
?
Controller (this can affect performance and endurance more than ?all other factors combined)
I don’t fully agree – if both the drive and the hard drive controller follows SATA 1.5, SATA 3.0 (aka SATA II) or SATA 6.0 (aka SATA III) standard, they’ll probably work together: the performance limits in the standard are UPPER LIMITS: many things can make your performance worse. ?I have not yet found a source of reliable test results allowing SSD performance comparison across brands, devices and OSes. ?
?
Bus Technology
Some SATA II devices claim SATA III compatibility but really just take advantage of SATA III’s downward compatibility with SATA II. ?It’s easy to see that only few of the spinning SATA III drives actually push more than 3Gb/sec of data. ?SSDs have an easier time of it, but I’ve not seen benchmarks showing any pair of SATA III SSDs + SATA III controllers reliably pushing anywhere near 6 Gb/sec.
?
Form Factor (Physical Size)
Some devices expect a 9.5mm 2.5in drive; the 7mm 2.5in drives don’t fit as well. ?The Kingston drive fit perfectly into the T60p’s 9.5mm slot.
?
Capacity
This seems linearly correlated to price, as you’d expect. ?Increasingly, SSDs are over-provisioned, where a 64GB drive actually has 72GB addressable. ?The excess allows for longer drive life in the face of SSD “bits” write cycles being limited to about 100,000. ?To compensate, SSD controllers move rather than overwrite changing file sectors, in order to balance the limited lifespan. ?Some OCZ devices destined for servers are reputedly 25% over-provisioned.
?
NAND or NOR technology
NAND technology appears to have won: NOR technology either isn’t as cost-effective, or it is hidden under more meaningful wrappers, such as wear-levelling, mean-time-between-failure (MTBF), and ever higher rates of over-provisioning
?
Power Consumption during Read, during Write, when Idle
It is difficult to measure this: as with many parameters for electronic devices, you’re trusting the reliability of manufacturer vs. the difficulty of measuring these by reviewers or consumers. ?Having said that, common SSDs are said to consume ~2W during operation, and 0.5W while idle.
?
Read/Write Burst and Sustained Throughput
Complex: ?Some SATA II SSDs can’t saturate SATA I’s 1.5Mb/sec pipe: others can. And you’re taking manufacturer’s word for it, despite the fact it’s possible to measure, with low precision, at least. ?
But other issues can obscure the results. For example: TRIM, which must be supported by the OS, the controller and the drive. ?TRIM is a technique to prevent SSDs from experiencing progressively slower write performance over time. Because SSD sectors are written all at once, if a previously used, partially full sector is to be written to, the drive must read the existing sector, add it to the new data in cache, then write the whole sector, then update the file table, which may require the same read-modify-write cycle. ?This can take some time.
Recently some reviewers tested whether Macs running Snow Leopard really needed TRIM, which isn’t supported by Snow Leopard: the reviewers built some seemingly solid tests, and reported some interesting but reasonable results. ?But other commenters pointed out the researchers assumed Apple’s secure formatting tool worked as promised: overwriting every bit of every sector with zeros, even for un-used file space. ?It turns out that Apple’s tool might not actually do this, and if the zero’d status of the tested SSD isn’t reliable, maybe the performance tests of new vs. “dirty” disks wasn’t reliable, either.