Configuration details for your system include the 13900K, ROG Maximus Z790 Dark Hero, RTX 4090, and a PCIe 5.0 SSD.
Configuration details for your system include the 13900K, ROG Maximus Z790 Dark Hero, RTX 4090, and a PCIe 5.0 SSD.
Many Z790 motherboards come with an M.2 port that supports PCIe 5.0 x4 SSDs. These SSDs also use the PCIe 5.0 x16 slot, but the CPU restricts it to 16 lanes of PCIe 5.0 and only 4 lanes of PCIe 4.0. As a result, manufacturers recommend using the PCIe 5.0 x16 port for the RTX 4090, though it runs at PCIe 5.0 x8. The 4.0 x16 version offers similar bandwidth to PCIe 5.0 x8. PCIe 4.0 x16 provides the same speed as PCIe 5.0 x8. Performance loss would occur if the card is forced to operate at PCIe 4.0 x8 due to lane limits. This issue has been discussed widely across forums, but no clear consensus exists.
It's a small difference, varying by game and quality. PCIe always works at the lowest common speed for both device type and connection length, meaning a Gen 1.1 x16 card in a Gen 4 x1 slot will function as Gen 1.1 x1. If you're looking for a PCIe 5.0 SSD, opting for an AM5 chipset is preferable to Intel. The Dark Hero isn't a good value, while the Apex Encore offers better performance for $50 less, and similar models are available at much lower prices.
The listing shows no bandwidth sharing between the M.2 slots and the PCIe slots supplied by the CPU. The GPU will operate at x8 only when a PCIe card is installed in the second x16 (which is actually x8) slot. The board includes a M.2 PCIe adapter, labeled as having a Gen 5 M.2 slot, without altering the lane configuration for the GPU compared to any other PCIe device in that slot.
Thanks for the update. I wasn’t expecting to purchase any additional boards yet—I already have the ASUS TUF GAMING Z790-PLUS with Wi-Fi and a 13600K paired with a Samsung 980 PRO. I’m just exploring a hypothetical situation to understand things better, since lane splitting was pretty unclear. It’s interesting how pricing has shifted; hero boards now seem to be costing more than apex models. Edited October 18, 2023 by Garborator