Serial ATA

SATA changed the physical attachment without replacing the ATA command family. The final 1.0 specification was announced on 29 August 2001, initially targeting 1.5 Gbit/s. That is a specification milestone, not a first-shipment date: the announcement described products as still forthcoming. The later 1.0a revision should not be used as SATA's origin. [2]

A 6 Gbit/s link does not carry 750 MB/s of file contents. SATA uses 8b/10b encoding: each eight data bits occupy ten transmitted bits. The coding-only ceiling is therefore 6 billion x (8/10) / 8 = 600 million bytes per second, or 600 MB/s in decimal units. Protocol traffic and device behavior reduce useful throughput further. The table applies the same calculation to all three rates; it is not a benchmark. [3]

Specification revision and signaling rate are different labels. SATA-IO recommends identifying the 6 Gbit/s interface explicitly rather than relying on the ambiguous phrase SATA 3.0. A faster-generation drive can connect to an older compatible interface, but it does not make that interface faster. Check the negotiated rate, not only the drive's advertised maximum. [4]

AHCI describes how host software programs a controller through registers and memory structures. It is not a cable format or a synonym for SATA. The specification permits fewer than 32 command slots; CAP.NCS encodes the implemented count minus one, while CAP.SNCQ reports NCQ support separately. Supported, configured and negotiated link speeds are also distinct observations. [5]

NCQ lets a supporting device schedule outstanding tagged commands. On an HDD this can reduce unnecessary head movement, but host, drive and software support are all required. A workload with little concurrency may gain little. Queue depth is an outstanding-command count, not a link rate or a promise of lower latency for each request. [6]

Historical scheduling evidence needs a boundary: Riska, Larkby-Lahet and Riedel (2007) studied three SCSI HDDs using Postmark and Linux builds. Their queue-depth results changed with cache policy and workload; request merging also changed the request sizes behind latency averages. This is not a SATA NCQ benchmark or a universal queue-depth recommendation. Compare application throughput and latency at a stated observation boundary. [1]

A direct SATA link is point-to-point; a compatible port multiplier can place several devices behind a host port. The shared upstream path does not become one independent full-bandwidth link per drive. Hot-plug support likewise needs a suitable platform and connection: a controller capability is not permission to unplug arbitrary internal hardware. [5]

USB enclosures add another layer. Smartmontools documents SCSI / ATA Translation and bridge-specific pass-through restrictions, including commands unavailable on particular bridges. An inaccessible SMART log through USB is therefore not proof that the SATA drive lacks that log. Record the adapter and access path before comparing diagnostic output. [7]