Pcie Specification 〈360p〉
Old specs (Gen 1-5) used NRZ (Non-Return to Zero)—simple, clean signaling. Gen 6 introduced PAM4, which is more susceptible to noise but necessary for physical limits. The spec includes new Forward Error Correction (FEC) logic to clean up that noise.
Marketing loves bandwidth (GB/s). Engineers love latency (nanoseconds). The spec carefully defines latency budgets for things like NVMe over PCIe. A GPU might not need 128 GB/s of bandwidth for a simple draw call, but it cannot tolerate a 1-microsecond delay. Why You Should Care (Even if You Aren't an Engineer) For the Gamer: Higher PCIe generations ensure that future GPUs won't be bottlenecked by the bus. While a Gen 3 x16 slot is mostly fine for an RTX 4090 today, that won't hold true for the GPUs of 2027. pcie specification
The spec dictates how fast your OS can boot and games can load. PCIe Gen 5 NVMe drives are now saturating the connection, pushing the bottleneck back to the NAND flash itself. Old specs (Gen 1-5) used NRZ (Non-Return to
For most users, the spec is invisible. For hardware designers, system architects, and serious enthusiasts, it is the rulebook that dictates the speed of your graphics card, the bandwidth of your NVMe SSD, and the future of I/O connectivity. Marketing loves bandwidth (GB/s)
Running large language models locally requires moving gigabytes of model weights from RAM to GPU. The PCIe specification determines how long that "warm up" time takes. The Future: PCIe 7.0 Don't look now, but PCI-SIG is already finalizing the 7.0 specification (expected 2025). It will double the data rate again to 128 GT/s using PAM4.
If you have ever opened a computer, you have seen them: those standardized beige or black slots on the motherboard. We call them PCIe slots. But while we often talk about "PCIe Gen 4" or "PCIe Gen 5," we rarely discuss the dense, complex document that makes it all work: The PCIe Specification.
Previous PCIe versions wasted about 2% of bandwidth on "packet headers." Starting with PCIe 6.0, the spec mandates FLIT mode, chopping data into fixed-size cells. This improves efficiency but required a complete rethinking of how retry buffers work.
