SATA vs NVMe in AS SSD: Why the Scores Differ So Much
Run AS SSD Benchmark on a SATA SSD and on an NVMe SSD and the results look like different sports: the sequential row multiplies several times over, the 4K-64Thrd row explodes, yet the humble single-thread 4K row barely moves. None of that is a benchmarking error. Each row is faithfully reporting a different structural difference between the two interfaces. This page explains, row by row, where the gap comes from — and backs it with published results measured on a single test bench, so the comparison is honest. If you want to reproduce any of this, the tool is on our download page.
The sequential gap is a link-speed gap
SATA III carries data over a 6 Gb/s link that was designed for mechanical drives. However fast the flash behind it becomes, a SATA drive’s sequential throughput stops at what that link can carry — in practice, review results for good SATA drives cluster right up against the link’s ceiling. NVMe drives attach directly to PCIe lanes, and each PCIe generation roughly doubles the available bandwidth, so the sequential ceiling is far higher and keeps moving. AS SSD’s Seq row measures a single 1 GB file streamed end to end, which is precisely the workload that hits the link ceiling and nothing else. That is why the Seq row shows the largest SATA-to-NVMe gap of the whole grid: it is mostly measuring the cable, not the flash.
The cleanest illustration comes from a single review bench. RelaxedTech tested both interfaces on the same system: the SATA Samsung 870 EVO 500 GB recorded sequential speeds of 558 MB/s read and 549 MB/s write, while the NVMe Samsung 970 EVO Plus 500 GB on the same bench recorded 3,503 MB/s read and 3,285 MB/s write — more than six times the read throughput, from the interface change alone.
The 4K-64Thrd gap is a queueing gap
The second structural difference is how commands queue. AHCI, the protocol SATA drives speak, was built around a single command queue suited to a spinning disk that services one request at a time. NVMe was designed for flash from the start, with deep parallel queues feeding many flash channels at once. AS SSD’s 4K-64Thrd test — random 4 KB blocks spread across 64 threads — exists to expose exactly this difference, and it doubles as the classic illustration of why Native Command Queuing matters: drives stuck in IDE mode, where NCQ is unavailable, fall flat in this row even on fast hardware. This row is also the biggest single contributor to the total score (see the score formula), which is why NVMe totals pull so far ahead: the scoring rewards the queue depth that NVMe was built to exploit.
The single-thread 4K gap is small — and that is the interesting part
Now look at the row that barely changes. Single-queue 4K random access is limited less by the link than by latency: one small request, issued, waited on, completed. Doubling link bandwidth does almost nothing for it, which is why a strong SATA drive’s 4K row can sit surprisingly close to an NVMe drive’s. A sourced example: Legit Reviews measured the entry-level NVMe Intel SSD 660p 1 TB at 54 MB/s 4K read, while the same outlet’s SATA-era results for good drives are in a similar neighbourhood — the interface revolution mostly passed this row by. This is also why moving from SATA to NVMe rarely feels six times faster in daily use, whatever the Seq row says: the workload your operating system generates looks much more like the 4K row than like a 1 GB file stream. We expand on that in Sequential vs 4K scores.
Same bench, two interfaces: the totals
RelaxedTech’s same-bench results pull the whole story together. The SATA drives in that roundup — Samsung 870 EVO 500 GB at a total of 1,242, Samsung 860 EVO 500 GB at 1,238 and Crucial MX500 500 GB at 1,231 — finished within a few points of each other, all pressed against the same interface ceiling. The NVMe drives on the same bench finished far ahead: the Crucial P1 500 GB at 2,691 and the Samsung 970 EVO Plus 500 GB at 4,931. Notice what the totals are really saying: within one interface, drives are separated by small margins, because the interface is the binding constraint; across interfaces, the totals separate by multiples, driven first by link bandwidth in the Seq row and then amplified by queue depth in the 4K-64Thrd row.
How to use this when comparing your own results
Three practical rules follow. First, never judge an NVMe drive by SATA expectations or the reverse — judge each drive against published results for its own model, as our scoring guide sets out. Second, if an NVMe drive returns SATA-shaped results — sequential speeds stuck at SATA-like levels and a flat 4K-64Thrd row — suspect the installation before the drive: it may be seated in a slower slot, running at reduced lane width, or using a legacy driver mode, all of which AS SSD’s top information strip helps reveal. Third, remember that the totals also embed the AS SSD version, so compare only within a version. The full set of sourced results behind this page is collected in AS SSD scores compared. Read that way — each drive against its own interface class, each row explained by the structure behind it — the SATA/NVMe gap stops looking like a verdict on your drive and starts looking like what it is: two different connection technologies, measured honestly by the same small tool.