PCIe lanes (storage)
A PCIe lane consists of one transmit and one receive differential signal pair, and a PCIe link can combine multiple lanes.
A lane has separate transmit and receive pairs. Link width counts lanes, not generations [5].
Gen3/4/5 specify 8/16/32 GT/s per lane. For these 128b/130b links, GT/s x lane count x 128/130 / 8 gives a one-direction ceiling in decimal GB/s. This excludes remaining protocol overhead and is not measured SSD payload throughput [8] [9].
M.2 can use up to four PCIe lanes, but physical fit does not establish its interface or available bandwidth [7].
SD Express in SD 7.0/7.1 uses Gen3 x1. Full-size SD 8.0 adds Gen4 x1, Gen3 x2 and Gen4 x2 options [3]. The bus table lists microSD 8.0 as Gen4 x1 [4]. Card and host negotiate a common link; these capabilities are not measured throughput [3].
Linux v6.12 current_link_speed and current_link_width read Link Status, separately from maximum capabilities [11]. The upstream minimum-link calculation is a configuration ceiling, not unused bandwidth or measured I/O [10]. Shared paths can carry competing traffic [12].
Support for one narrower x4 device does not imply support for four independent x4 links. Check the CPU controller, slot routing and firmware together [13] [14].
Relations
- related
- PCIe storage attachment, M.2, U.2, SATA Express, NVMe, PCIe 4.0 storage, PCIe 5.0 storage, PCIe retimer, NVMe backplane, SD Express, CXL memory expansion, NAND channel, SSD controller, AHCI, PCIe bifurcation, PCIe switch
Advantages
- Independent transmit and receive paths allow simultaneous traffic in both directions [5].
- At the same PCIe generation, additional lanes increase the theoretical link capacity [7].
Limitations
- A shared upstream path can constrain simultaneous transfers from several devices even when each device has a wide local link [12].
- An M.2 slot may support SATA, PCIe or both, so connector fit alone does not prove protocol compatibility [7].
- Do not apply the Gen3/4/5 encoding formula to later FLIT operation. [8]
- Per-direction ceilings must not be doubled and presented as one-direction read throughput. [9]
- Lane counts at several endpoints cannot be added to infer their shared upstream capacity. [12]
Citations
- PCI Express (Wikipedia), sections on lanes, link widths, revision comparison table, link efficiency and storage devices
- M.2 (Wikipedia), sections on features, storage interfaces and keying
- SD Association: SD Express Memory Cards with PCIe and NVMe Interfaces, June 2020, pages 4, 6 and 9
- SD Association: bus modes, SD 8.0 lane configurations and microSD footnote
- AMD PG213 v1.3: PCI Express Interface, differential lane signals
- Kingston: 2 Types of M.2 SSDs: SATA and NVMe, lane widths and shared host connections
- PCI-SIG: PCIe 7.0 educational webinar, July 2025, slide 7; historical Gen3/4/5 rates and encoding. No later-generation throughput is calculated here.
- PCI-SIG: How does PCIe 3.0 8 GT/s double PCIe 2.0 5 GT/s? Encoding efficiency and direction conventions.
- Linux PCI support library: pcie_bandwidth_available; upstream path and limiting link.
- Linux v6.12, drivers/pci/pci-sysfs.c: current_link_speed_show and current_link_width_show read Link Status; maximum attributes read capabilities.
- Intel 13th/14th Gen Core datasheet v009: narrower-link support versus lane bifurcation
- ASUS Hyper M.2 compatibility: CPU, slot selection and firmware configuration
Further reading
Sources & caveats
- The SD 8.0 full-size configurations and link negotiation use the June 2020 whitepaper; the microSD 8.0 boundary uses the current bus table [3] [4].
- The coding-rate example is arithmetic, not a benchmark; the added Gen3/4/5 arithmetic is not a payload benchmark. Product availability is outside this note.
- Legacy secondary references are retained for background; new lane-structure and SD Express statements use primary documentation [1] [2] [5] [3] [4].
- The November reconciliation preserves the October physical-lane and SD Express statements and source IDs while adding negotiated-state, upstream-path and bifurcation limits.